HK1230300B - Systems and methods for analyzing and synthesizing complex knowledge representations - Google Patents

Description

Systems and methods for analyzing and synthesizing complex knowledge representations

Divisional Application Instructions

This application is a divisional application of the Chinese invention patent application with the international application date of June 22, 2011, which entered the Chinese national phase on December 21, 2012, with application number 201180031005.5 and the name “System and method for analyzing and synthesizing complex knowledge representation”.

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority to U.S. Provisional Application No. 61/357,266, filed June 22, 2010, entitled “Systems and Methods for Analyzing and Synthesizing Complex Knowledge Representations,” which is hereby incorporated by reference in its entirety. This application also builds upon concepts disclosed in multiple prior applications filed by the same applicant and/or assignee, including the following, to which the reader is referred by way of background in addition to the background discussed below: U.S. patent application Ser. No. 13/161,165, filed Jun. 15, 2011, entitled “Systems and Methods for Analyzing and Synthesizing Complex Knowledge Representations”; U.S. patent application Ser. No. 12/477,977, filed Jun. 4, 2009, entitled “System, Method and Computer Program for Transforming an Existing Complex Data Structure to Another Complex Data Structure”; U.S. patent application Ser. No. 11/469,258, filed Aug. 31, 2006, entitled “Complex-Adaptive System for Providing a Faceted Classification”, now U.S. Patent No. 7,596,574; and U.S. patent application Ser. No. 13/161,165, filed Jun. 15, 2011, entitled “Systems and Methods for Analyzing and Synthesizing Complex Knowledge Representations”; U.S. patent application Ser. No. 12/477,977, filed Jun. 4, 2009, entitled “System, Method and Computer Program for Transforming an Existing Complex Data Structure to Another Complex Data Structure”; U.S. patent application Ser. No. 11/469,258, filed Aug. 31, 2006, entitled “Complex-Adaptive System for Providing a Faceted Classification”; and U.S. patent application Ser. No. 13/161,165, filed Mar. 30, 2006, entitled “System, Method, and Computer Program for Constructing and Managing Dimensional Information Structures"

Background Art

Broadly speaking, knowledge representation is the activity of making abstract knowledge explicit as concrete data structures to support machine-based storage, management, and reasoning systems. Conventional methods and systems exist for utilizing knowledge representation (KR) constructed according to various types of knowledge representation models, including structured controlled vocabularies such as taxonomies, thesauri, and faceted classifications; formal specifications such as semantic networks and ontologies; and unstructured forms such as natural language-based documents.

A taxonomy is a KR structure that organizes categories into a hierarchical tree and associates categories with related objects such as physical items, documents, or other digital content. Categories or concepts in a taxonomy are typically organized in terms of inheritance relationships, also known as parent-child relationships, generality-specialization relationships, or parent-child relationships. In such a relationship, a subcategory or concept has the same properties, behaviors, and constraints as its parent, as well as one or more additional properties, behaviors, or constraints. For example, the knowledge statement "a dog is a mammal" can be encoded in a taxonomy by concepts/categories labeled "mammal" and "dog" linked by a parent-child hierarchical relationship. Such a representation encodes the concept that a dog (child concept) is of type mammal (parent concept), but not every mammal is necessarily a dog.

A dictionary is a KR that represents terms (such as search keywords for information retrieval), which are often encoded as single-word noun concepts. The links between terms/concepts in a dictionary are generally divided into the following three types of relationships: hierarchical relationships, equivalence relationships, and association relationships. Hierarchical relationships are used to link terms that are narrower and wider in scope than each other, which is similar to the relationship between concepts in a taxonomy. Following the previous example, "dog" and "mammal" are terms linked by a hierarchical relationship. Equivalence relationships link terms that can replace each other as search terms, such as synonyms or near-synonyms. For example, the terms "dog" and "canine" can be linked by an equivalence relationship in some contexts. An association relationship links related terms whose relationships are neither hierarchical nor equivalent. For example, a user searching for the term "dog" may also want to see items returned from a search for "breeder," and the association relationship can be encoded in the dictionary data structure for this pair of terms.

Facet classification is based on the principle that information has a multidimensional quality and can be encoded in many different ways. The subject matter of an information domain is subdivided into facets (or more simply, categories) that represent this dimensionality. The attributes of the domain are related in the facet hierarchy. Objects within the domain are then described and classified based on these attributes. For example, a collection of clothing offered for sale in a physical or web-based clothing store can be classified using color facets, material facets, style facets, etc., where each facet has multiple hierarchical attributes representing different types of colors, materials, styles, etc. Facet classification is often used in facet search systems, for example, to allow users to search for clothing collections according to any desired facet sorting (e.g., color first, then style, style first, then color, then style, or any other desired facet priority order). Such facet classification contrasts with classification by taxonomy, in which the category hierarchy is fixed.

A semantic network is a network representation of various types of semantic relationships between concepts, or a data structure that encodes or instantiates the network structure. A semantic network is typically represented as a directed or undirected graph consisting of vertices representing concepts and edges labeled with the type of relationship linking pairs of concepts. An example of a semantic network is WordNet, a lexical database for the English language. Some common types of semantic relationships defined in WordNet are whole-part relationships (A is a part of B), hyponymy (A is a type of B), synonyms (A and B are synonyms), and antonyms (A and B are antonyms).

An ontology is a KR structure that encodes concepts and the relationships between those concepts, and is limited to the specific domain of the real or virtual world it is used to model. The concepts included in an ontology generally represent the specific meanings of terms when they are applied to the domain being modeled or classified, and the included conceptual relationships generally represent how those concepts are related within the domain. For example, the concept corresponding to the word "card" can have different meanings in an ontology about the domain of card games than in an ontology about the domain of computer hardware.

Generally speaking, all of the KR types discussed above, as well as other conventional examples, are tools for modeling human knowledge in terms of abstract concepts and the relationships between those concepts, and for making that knowledge accessible to machines (such as computers) for performing various knowledge-requiring tasks. Thus, human users and software developers conventionally use their human knowledge to construct KR data structures, and manually encode the completed KR data structures into a machine-readable form as data structures to be stored in machine memory and accessed by various machine-executed functions.

Summary of the Invention

One embodiment relates to a method for generating a complex knowledge representation, the method comprising: receiving input indicating a request context; applying, with a processor, one or more rules to a basic data structure representing at least one basic concept, at least one basic concept relationship, or at least one basic concept and at least one basic concept relationship; synthesizing, based on the application of the one or more rules, one or more additional concepts, one or more additional concept relationships, or one or more additional concepts and one or more additional concept relationships according to the request context; and generating a complex knowledge representation according to the request context using at least one additional concept among the additional concepts, at least one additional concept relationship among the additional concept relationships, or at least one additional concept among the additional concepts and at least one additional concept relationship among the additional concept relationships.

Another embodiment relates to a system for generating a complex knowledge representation, the system comprising at least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by at least one processor, perform: receiving input indicating a request context; applying one or more rules to a basic data structure representing at least one basic concept, at least one basic concept relationship, or at least one basic concept and at least one basic concept relationship; synthesizing one or more additional concepts, one or more additional concept relationships, or one or more additional concepts and one or more additional concept relationships according to the request context based on the application of the one or more rules; and generating a complex knowledge representation according to the request context using at least one additional concept among the additional concepts, at least one additional concept relationship among the additional concept relationships, or at least one additional concept among the additional concepts and at least one additional concept relationship among the additional concept relationships.

Another embodiment relates to at least one non-transitory computer-readable storage medium encoded with a plurality of computer-executable instructions for generating a complex knowledge representation, wherein the instructions, when executed, perform: receiving input indicating a request context; applying one or more rules to a basic data structure representing at least one basic concept, at least one basic concept relationship, or at least one basic concept and at least one basic concept relationship; synthesizing one or more additional concepts, one or more additional concept relationships, or one or more additional concepts and one or more additional concept relationships according to the request context based on the application of the one or more rules; and generating a complex knowledge representation according to the request context using at least one additional concept among the additional concepts, at least one additional concept relationship among the additional concept relationships, or at least one additional concept among the additional concepts and at least one additional concept relationship among the additional concept relationships.

Another embodiment relates to a method for deconstructing an original knowledge representation, the method comprising: receiving input corresponding to the original knowledge representation; applying one or more rules with a processor to deconstruct the original knowledge representation into one or more basic concepts, one or more basic concept relationships, or one or more basic concepts and one or more basic concept relationships; and including in a basic data structure a representation of at least one basic concept among the basic concepts, at least one basic concept relationship among the basic concept relationships, or at least one basic concept among the basic concepts and at least one basic concept relationship among the basic concept relationships.

Another embodiment relates to a system for deconstructing an original knowledge representation, the system comprising at least one non-transitory computer-readable storage medium storing processor-executable instructions, which, when executed by at least one processor, perform: receiving input corresponding to the original knowledge representation; applying one or more rules to deconstruct the original knowledge representation into one or more basic concepts, one or more basic concept relationships, or one or more basic concepts and one or more basic concept relationships; and including in a basic data structure a representation of at least one basic concept among the basic concepts, at least one basic concept relationship among the basic concept relationships, or at least one basic concept among the basic concepts and at least one basic concept relationship among the basic concept relationships.

Another embodiment relates to at least one non-transitory computer-readable storage medium encoded with a plurality of computer-executable instructions for deconstructing an original knowledge representation, wherein the instructions, when executed, perform: receiving input corresponding to the original knowledge representation; applying one or more rules to deconstruct the original knowledge representation into one or more basic concepts, one or more basic concept relationships, or one or more basic concepts and one or more basic concept relationships; and including in a basic data structure a representation of at least one of the basic concepts, at least one of the basic concept relationships, or at least one of the basic concepts and at least one of the basic concept relationships.

Another embodiment relates to a method for supporting semantic interoperability between knowledge representations, the method comprising: for each input knowledge representation in a plurality of input knowledge representations, applying, with a processor, one or more rules to deconstruct the input knowledge representation into one or more basic concepts, one or more basic concept relationships, or one or more basic concepts and one or more basic concept relationships; and comprising, with the processor, in a shared basic data structure, a representation of at least one basic concept among the basic concepts, at least one basic concept relationship among the basic concept relationships, or at least one basic concept among the basic concepts and at least one basic concept relationship among the basic concept relationships for each input knowledge representation in the plurality of input knowledge representations.

Another embodiment relates to a system for supporting semantic interoperability between knowledge representations, the system comprising at least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by at least one processor, perform: for each input knowledge representation in a plurality of input knowledge representations, applying one or more rules to deconstruct the input knowledge representation into one or more basic concepts, one or more basic concept relationships, or one or more basic concepts and one or more basic concept relationships; and including, in a shared basic data structure, for each input knowledge representation in the plurality of input knowledge representations, a representation of at least one of the basic concepts, at least one of the basic concept relationships, or at least one of the basic concepts and at least one of the basic concept relationships.

Another embodiment relates to at least one non-transitory computer-readable storage medium encoded with a plurality of computer-executable instructions for supporting semantic interoperability between knowledge representations, wherein the instructions, when executed, perform: for each input knowledge representation in a plurality of input knowledge representations, applying one or more rules to deconstruct the input knowledge representation into one or more base concepts, one or more base concept relationships, or one or more base concepts and one or more base concept relationships; and including, in a shared base data structure, for each input knowledge representation in the plurality of input knowledge representations, a representation of at least one of the base concepts, at least one of the base concept relationships, or at least one of the base concepts and at least one of the base concept relationships.

Another embodiment relates to a computer-implemented method for synthesizing a complex indication representation, the method comprising: receiving a context from a data consumer; identifying one or more basic components in a basic knowledge representation including a first concept based on the context; and generating a complex knowledge representation by applying one or more rules to the one or more basic components through execution of stored instructions by at least one processor, wherein generating the complex knowledge representation comprises synthesizing complex concepts that do not exist in the basic knowledge representation, and including the synthesized complex concepts in the complex knowledge representation, wherein synthesizing the complex concept comprises joining the first concept and basic concepts that are not hierarchically related to the first concept to form the synthesized complex concept.

Wherein, in the method, generating the complex knowledge representation includes including internal relationships in the complex knowledge representation, the internal relationships joining concepts in a set to create a complex concept.

Wherein, in the method, generating the complex knowledge representation includes encoding, in the complex knowledge representation, an intrinsic relationship between the first concept and the complex concept synthesized by joining the first concept with the basic concept.

Wherein, in the method, the first concept and the basic concept joined with the first concept to synthesize the complex concept form a concept definition for the complex concept.

Wherein, in the method, generating the complex knowledge representation includes including extrinsic relations in the complex knowledge representation, the extrinsic relations describing features between pairs of concepts.

Wherein, in the method, generating the complex knowledge representation includes encoding both the intrinsic relationship between the first concept and the complex concept and the extrinsic relationship between the first concept and the second concept in the complex knowledge representation.

Wherein, in the method, the extrinsic relationship between the first concept and the second concept encodes a hierarchical relationship between the first concept and the second concept.

Another embodiment relates to at least one non-transitory computer-readable storage medium storing computer-executable instructions that, when executed, perform a method for synthesizing a complex knowledge representation, the method comprising: receiving a context from a data consumer; identifying one or more basic components in a basic knowledge representation that include a first concept based on the context; and generating a complex knowledge representation by applying one or more rules to the one or more basic components through execution of the stored instructions by at least one processor, wherein generating the complex knowledge representation comprises synthesizing complex concepts that do not exist in the basic knowledge representation, and including the synthesized complex concepts in the complex knowledge representation, wherein synthesizing the complex concept comprises joining the first concept and basic concepts that are not hierarchically related to the first concept to form the synthesized complex concept.

Wherein, in the at least one non-transitory computer-readable storage medium, wherein generating the complex knowledge representation comprises including internal relationships in the complex knowledge representation, the internal relationships joining concepts in a set to create a complex concept.

Wherein, in the at least one non-transitory computer-readable storage medium, generating the complex knowledge representation includes encoding, in the complex knowledge representation, an intrinsic relationship between the first concept and the complex concept synthesized by joining the first concept with the basic concept.

Wherein, in the at least one non-transitory computer-readable storage medium, the first concept and the basic concepts joined with the first concept to synthesize the complex concept form a concept definition for the complex concept.

Wherein, in the at least one non-transitory computer-readable storage medium, wherein generating the complex knowledge representation comprises including extrinsic relations in the complex knowledge representation, the extrinsic relations describing characteristics between pairs of concepts.

wherein, in the at least one non-transitory computer-readable storage medium, generating the complex knowledge representation comprises encoding both the intrinsic relationship between the first concept and the complex concept and the extrinsic relationship between the first concept and the second concept in the complex knowledge representation.

wherein, in the at least one non-transitory computer-readable storage medium, wherein the extrinsic relationship between the first concept and the second concept encodes a hierarchical relationship between the first concept and the second concept.

Another embodiment relates to an apparatus comprising: at least one processor; and at least one storage medium storing processor-executable instructions, the processor-executable instructions, when executed by the at least one processor, performing a method for synthesizing a complex knowledge representation, the method comprising: receiving a context from a data consumer; identifying one or more basic components in a basic knowledge representation including a first concept based on the context; and generating a complex knowledge representation by applying one or more rules to the one or more basic components, wherein generating the complex knowledge representation comprises synthesizing a complex concept that does not exist in the basic knowledge representation, and including the synthesized complex concept in the complex knowledge representation, wherein synthesizing the complex concept comprises joining the first concept and a basic concept that is not hierarchically related to the first concept to form the synthesized complex concept.

Wherein, in the apparatus, generating the complex knowledge representation comprises including internal relationships in the complex knowledge representation, the internal relationships joining concepts in a set to create a complex concept.

Herein, in the apparatus, generating the complex knowledge representation includes encoding, in the complex knowledge representation, an intrinsic relationship between the first concept and the complex concept synthesized by joining the first concept with the basic concept.

Wherein, in the apparatus, the first concept and the basic concept joined with the first concept to synthesize the complex concept form a concept definition for the complex concept.

Wherein, in the apparatus, generating the complex knowledge representation includes including extrinsic relations in the complex knowledge representation, the extrinsic relations describing features between pairs of concepts.

Wherein, in the apparatus, generating the complex knowledge representation includes encoding both the intrinsic relationship between the first concept and the complex concept and the extrinsic relationship between the first concept and the second concept in the complex knowledge representation.

The foregoing is a non-limiting summary of the invention as defined by the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings are not intended to be drawn to scale. In the accompanying drawings, each identical or nearly identical component is illustrated in various figures. For the sake of clarity, not every component may be labeled in every figure. In the accompanying drawings:

FIG1 is a block diagram illustrating an exemplary system for implementing an atomic knowledge representation model according to some embodiments of the present invention;

FIG2A illustrates an exemplary complex knowledge representation according to some embodiments of the present invention;

FIG2B illustrates an exemplary basic data structure of an atomic knowledge representation model according to some embodiments of the present invention;

FIG3 illustrates an exemplary data plan according to some embodiments of the present invention;

FIG4 illustrates an exemplary method for analyzing complex knowledge representations according to some embodiments of the present invention;

5 is a block diagram illustrating an exemplary distributed system for implementing analysis and synthesis of complex knowledge representations according to some embodiments of the present invention;

6 is a flow chart illustrating an exemplary method for analyzing a complex knowledge representation to generate an elementary data structure according to some embodiments of the present invention;

FIG7 illustrates an exemplary method for synthesizing complex knowledge representations from basic data structures according to some embodiments of the present invention;

FIG8 is a table illustrating an exemplary knowledge processing rule set according to some embodiments of the present invention;

FIG9 illustrates an example of a knowledge representation that can be derived from an exemplary natural language text;

FIG10 illustrates an example of a basic data structure that can be parsed from an exemplary thesaurus; and

11 is a block diagram illustrating an exemplary computing system for use in implementing some embodiments of the present invention.

DETAILED DESCRIPTION

As discussed above, knowledge representation (KR) data structures created by conventional methods encode and represent a specific set of human knowledge modeled for a specific domain or context. Since KRs are typically constructed by human developers and programmed into machine memory in a complete form, conventional KRs contain only a subset of the human knowledge that the human user originally used to program it.

For example, a KR may encode the knowledge statement "Dogs are mammals," and it may also express statements or assertions about animals that are mammals, such as "Mammals produce milk to feed their pups." The inventors have recognized that such fact combinations, when combined with appropriate logical and semantic rules, can support a wide range of human reasoning, thereby making explicit inferences (such as "Dogs produce milk to feed their pups") that were not initially seeded as facts within the KR explicit. The inventors have appreciated that extending the KR data structure with such inferences can be used to support a variety of knowledge-based activities and tasks, such as inference/reasoning (as described above), information retrieval, data mining, and other forms of analysis.

However, as discussed above, methods for constructing and encoding KRs have conventionally been limited to manually inputting complete KR structures for access and use by machines (e.g., computers). Continuing with the above example, while a human acting as a KR designer may implicitly understand why the fact that "dogs produce milk to feed their pups" holds true, the properties that must hold for it to hold true (in this case, properties such as transitivity and inheritance) are not conventionally an explicit part of the KR. In other words, any underlying set of rules that can guide the creation of new knowledge is not conventionally encoded as part of the KR, but is actually applied from outside the system by the human designer when constructing the KR.

A previously unrecognized consequence of conventional approaches is that knowledge can be expressed in KRs for machine consumption, but the KRs themselves cannot be created by machines. Humans are forced to model knowledge domains for machine consumption. Unfortunately, because human knowledge is so vast and, in many cases, subjective, modeling all knowledge domains is technically infeasible.

Furthermore, because so much knowledge must be explicitly encoded as data, the resulting data structures quickly become enormous as the knowledge domain grows. Because conventional KR does not encode its underlying theories or practices for knowledge creation as part of the data that makes up the knowledge representation model, its resulting data structures can become complex and impractical. In other words, because knowledge representation cannot be created by a machine, it must conventionally be provided as explicit data or otherwise deduced or inductively summarized through logical or statistical means.

Therefore, the conventional KR approach leads to multiple problems:

Large and complex data structures: Data structures that encode knowledge representations are conventionally complex to construct and maintain. Even relatively simple machine-readable knowledge domains (such as simple statements about dogs and mammals) can generate data volumes orders of magnitude larger than their natural language counterparts.

Reliance on Domain Expertise: The underlying theories guiding the practice of KR must be expressed by humans in the routine creation of KR data structures. This is a time-consuming activity that excludes most humans and all machines from the production of these crucial data assets. As a result, most human knowledge to date has remained implicit and outside the realm of computation.

Creating data before use: Knowledge is conventionally modeled as data before it is called upon for a specific use, which is costly and potentially wasteful if the knowledge is not needed. Therefore, the inventors have realized that if knowledge can be created by machines only on demand, data generation and storage requirements can be greatly reduced.

Large-scale data and processing costs: Conventional KR systems must reason over large data structures in order to create new facts or answer queries. This scale burden represents a significant challenge in conventional KR systems, a burden that can be reduced by using more of a just-in-time approach to creating the underlying data structures rather than the conventional data-before-use approach.

Integration and interoperability challenges: Semantic interoperability (the ability of two different KRs to share knowledge) is a massively difficult challenge when various KRs are created under different models and expressed in different ways, often dealing with subjective and ambiguous topics. Precision and the ability to accurately reason are often lost across multiple different KRs. In this regard, the inventors have appreciated that knowledge coordination across different KRs can become a thorny issue if the underlying theory for how knowledge is created is included as part of the KRs.

Thus, some embodiments according to the present disclosure provide a system for encoding knowledge creation rules to automate the process of creating knowledge representations.Some embodiments combine new knowledge representation synthesis approaches with computing systems for creating and managing the resulting data structures derived from such approaches.

Some embodiments combine more compressed (atomic) data sets with generative rule sets that encode fundamental knowledge creation rather than modeling all knowledge in the field as explicit data. Such rules can be applied by the system in some embodiments when it is necessary or desirable to create new knowledge and explicitly express it as data. It will be appreciated from the above discussion that the benefit of such technology can be, in at least some cases, to significantly reduce the amount of data in the system and to provide new capabilities and applications for machine-based new knowledge creation (synthesis). However, it will be appreciated that not every embodiment according to the present invention can solve every identified problem in a conventional manner, and some embodiments may not solve any of these problems. Some embodiments may also solve problems other than those described here. In addition, not every embodiment can provide all or any of the benefits discussed here, and some embodiments may provide other benefits that are not described.

Some embodiments also provide techniques for complex knowledge representations, such as taxonomies, ontologies, and faceted classifications, to interoperate not only at the data level but also at the semantic level (meaning interoperability).

Other benefits that may be conferred in some embodiments and applicable across many new and existing application domains include: lower costs in both the generation and application of knowledge representations enabled by simpler and more economical data structures; the potential for new knowledge creation; a more scalable system enabled by just-in-time, on-demand knowledge; and support for "context" from users and data consumers as input variables. The dynamic nature of some embodiments of the present disclosure, which apply synthetic and analytical knowledge processing rules on a just-in-time basis to create knowledge representation data structures, can provide benefits that are more economical than conventional approaches that analyze and model the entire knowledge domain in advance.

By incorporating a fundamental knowledge creation rule set into KR, the amount of data in the system can be reduced, thereby providing a more economical data management system and providing new applications for knowledge management. Therefore, in some embodiments, the cost of creating and maintaining a KR system can be reduced by reducing the burden of data scalability and not creating data until it is needed. Once created, the data structures that model complex knowledge are, in some embodiments, smaller than in conventional systems because they only contain data relevant to the task at hand. This in turn can reduce the cost of downstream applications (such as inference engines or data mining tools that work on these knowledge models).

The integrated computing approach according to some embodiments of the present disclosure also supports entirely new capabilities in knowledge representation and data management. Some embodiments may provide improved support for "possibilities" (i.e., creating entirely new knowledge representations from existing data). For example, such possibilities may be useful for creative activities such as education, journalism, and art.

FIG1 illustrates an exemplary system 100 according to some embodiments of the present invention, which can be used in some embodiments to implement an atomic knowledge representation model (AKRM) involved in the analysis and synthesis of complex knowledge representations (KRs). In exemplary system 100, the AKRM can be encoded as computer-readable data and stored on one or more tangible, non-transitory computer-readable storage media. For example, the AKRM can be stored in a dataset 110 in non-volatile computer memory using the following data structure, some examples of which are provided below. The data structure is designed to support both basic and complex knowledge representation data structures.

In some embodiments, the AKRM may include one or more base data structures 120 and one or more knowledge processing rules 130. In some embodiments, the rules 130 may be used by the system 100 to deconstruct (analyze) one or more complex KRs to generate the base data structures 120. For example, the system 100 may include one or more computer processors and one or more computer memory hardware components, and the memory may be encoded with computer-executable instructions that, when executed by the one or more processors, cause the one or more processors of the system 100 to use the rules 130 when analyzing the one or more complex KRs to generate the base data structures 120 of the AKRM. The memory may also be encoded with instructions that program the one or more processors to use the rules 130 to synthesize new complex KRs from the base data structures 120. In some embodiments, the computer memory may be implemented as one or more tangible, non-transitory, computer-readable storage media encoded with computer-executable instructions that, when executed, cause the one or more processors to perform any of the functions described herein.

Unlike previous knowledge representation systems, systems according to some embodiments of the present invention (such as system 100) can combine data structures and knowledge processing rules to create a knowledge representation model that is encoded as data. In some embodiments, the rules may not be encoded as knowledge (for example, as rules or axioms that describe the boundaries or constraints of knowledge within a specific domain), but are actually encoded as construction and deconstruction rules for creating data structures that represent new knowledge. In addition to "inference rules" for generating implicit facts such as the following (these implicit facts are logical consequences of explicit concepts given by the original KR), in some embodiments, the knowledge representation model can also be encoded with "knowledge processing rules", which can be applied to create new knowledge that may not be implicit from the original KR data structure.

For example, starting with two explicit knowledge statements "Mary is a person" and "All people are human", inference rules can be applied to determine the explicit knowledge statement "Mary is a human", which is the logical result of the previous two statements. In a different example according to some embodiments of the present invention, starting with two explicit knowledge statements "Mary is Bob's friend" and "Bob is Charlie's friend", exemplary knowledge processing rules that model the meaning of friend relationships can be applied to determine the new knowledge statement "Mary is Charlie's friend". Obviously, the application of such knowledge processing rules can produce new knowledge, which is not necessarily a logical result of the explicit knowledge given in the original input KR. As described above, the knowledge representation model according to some embodiments of the present invention (including knowledge processing rules stored in association with data structures that encode concepts and concept relationships (such as those that are different from or supplement logical reasoning rules)) can model a framework for how new and potentially non-implicit knowledge can be created and/or decomposed.

This focus on knowledge synthesis can bring systems such as system 100 into new application areas. While existing systems focus on deductive reasoning (i.e., where insights are gleaned by precisely deducing existing facts and arguments), systems according to some embodiments of the present invention can support inductive reasoning and other types of theory building (i.e., where existing facts can be used to support probabilistic predictions of new knowledge).

In some embodiments according to the present invention, the system (such as system 100) can be loosely based on a framework based on conceptual semantics, thereby encoding semantic primitives (such as "atoms" or "basic" concepts) and rules (principles), which guide how such atomic structures can be combined to create more complex knowledge. However, it should be appreciated that the system according to an embodiment of the present invention can work within many such frameworks because aspects of the present invention are not limited to any specific theory, model or practice of knowledge representation. In some embodiments, the system (such as system 100) can be designed to dock with a wide range of methods and techniques (such as, implemented as software applications or components) that model these frameworks. For example, the docking analysis component (such as analysis engine 150) can decompose the input complex KR 160 into a basic data structure 120. The synthesis component (such as synthesis engine 170) can use the basic data structure 120 to construct a new output complex KR 190.

In some embodiments, the analysis engine 150 can analyze the input complex KR 160, for example, by executing appropriate computer-readable instructions by one or more processors of the system 100 and applying one or more knowledge processing rules 130 to deconstruct the data structure of the input KR 160 into more basic constructs. In some embodiments, the most basic constructs included within the basic data structure 120 of the AKRM 110 can represent a minimal set of fundamental building blocks of information and information relationships that, collectively, provide the information-carrying capacity used to classify the input data structure. The input KR 160 can be obtained from any suitable source, including direct input from a user or software application interacting with the system 100. In some embodiments, the input KR 160 can be obtained by interfacing with various database technologies, such as a relational database system or a graph-based database system. It should be appreciated that the input KR 160 can be obtained in any suitable form and in any suitable manner, as aspects of the present invention are not limited in this respect.

For example, Figure 2A illustrates a small complex KR 200 (in this example, a taxonomy) that a user or software application using the system 100 may input into the analysis engine 150. The complex KR 200 includes a set of concepts that are linked by various hierarchical relationships. For example, a concept 210 labeled "animals" is linked in a parent-child relationship to a concept 220 labeled "pets" and a concept 230 labeled "mountain animals." At each level of the hierarchy, the concept entities represent units of meaning that can be combined to create more complex semantics or potentially deconstructed into more basic semantics. For example, the complex meaning of "mountain animals" may include the concepts "mountain" and "animals."

In some embodiments, the system 100 can deconstruct a complex KR (such as complex KR 200) to discover basic concepts, which include the complex concepts of the complex KR, for example, through the analysis engine 150. For example, FIG2B illustrates a basic data structure 300 that can be generated from the analysis and deconstruction of the complex KR 200. In the basic data structure 300, it has been discovered that the complex concept 230 labeled "mountain-dwelling animals" includes a more basic concept 235 labeled "mountain-dwelling" and a more basic concept 240 labeled "animals". In this example, "mountain-dwelling" and "animals" represent more basic concepts than the more complex concept labeled "mountain-dwelling animals" because the concepts "mountain-dwelling" and "animals" can be combined to create the concept labeled "mountain-dwelling animals". Similarly, a complex concept 250 labeled "domestic dog" has been discovered to include a more basic concept 255 labeled "domestic" and a more basic concept 260 labeled "dog," and a complex concept 270 labeled "Siamese cat" has been discovered to include a more basic concept 275 labeled "Siamese" and a more basic concept 280 labeled "cat." Furthermore, each newly discovered basic concept has inherited concept relationships from the complex concept that includes it. Thus, "domestic," "dog," "Siamese," and "cat" are children of "pet"; "mountain dweller" and "animal" (concept 240) are children of "animal" (concept 210); and "mountain dweller" and "animal" (concept 240) are both parents of concept 290 labeled "lion" and concept 295 labeled "goat."

Note that although the label "animal" is attributed to both concept 210 and concept 240 in the basic data structure 300, the two concepts may still represent different abstract meanings that function differently within the knowledge representation hierarchy. In some embodiments, "labels" or "symbols" can be combined into abstract concepts to provide human and/or machine readable terms or labels for concepts and relationships and to provide a basis for various symbol-based processing methods (such as text analysis). Labels can provide knowledge representation entities that are discernible to humans and/or machines and can be derived from a unique vocabulary in the source domain. Therefore, because the labels assigned to each concept element can be extracted from the language and terminology present in the domain, the labels themselves may not fully describe the abstract concepts and concept relationships that they are used to name, as those abstract entities are understood in human knowledge.

Similarly, in some embodiments, it should be appreciated that there is a difference between abstract concepts in a knowledge representation model and the objects that those concepts can be used to describe or classify. An object can be any item in the real physical or virtual world that can be described by a concept (e.g., examples of objects are documents, web pages, characters, etc.). For example, an individual in the real world can be represented by a conceptual abstraction labeled "Bob." The information in the domain to be described, classified, or analyzed can involve virtual or physical objects, processes, and the relationships between such information. In some exemplary embodiments, complex KRs as described herein can be used when classifying content residing within a web page. Other types of domains may, in some embodiments, include document repositories, recommendation systems for music, software code repositories, models of workflows and business processes, and the like.

In some embodiments, the objects in the field to be classified can be referred to as content nodes. Content nodes can include any objects that are responsible for classification, description, analysis, etc. using a knowledge representation model. For example, a content node can be a file, a document, a block of a document (such as an annotation), an image, or a stored string. A content node can reference a physical object or a virtual object. In some embodiments, a content node can be contained in a content container that provides addressable (or locatable) information by which the content node can be retrieved. For example, the content container of a web page addressable by a URL can include content nodes in the form of text and images. Content can be associated with a content node to extract some meanings (such as the description, purpose, usage, or intention of the content node). For example, aspects of a content node in the real world can be described by concepts in an abstract knowledge representation.

Concepts are defined at a complex level of abstraction through their relationships to other entities and structurally in terms of other more basic knowledge representation entities (e.g., keywords and morphemes). Such structures are referred to here as concept definitions. In some embodiments, concepts can be related by two basic types of conceptual relationships: an intrinsic type, which refers to the connection between basic concepts used to create more complex concepts (e.g., the relationship between "mountain dwelling," "animals," and "mountain dwelling animals" in the basic data structure 300); and an extrinsic type, which refers to the connection between complex relationships. Extrinsic relationships can describe features between pairs of concepts, such as equivalence, hierarchy (e.g., the relationship between "animals" and "pets"), and association. In addition, in some embodiments, extrinsic and intrinsic concept relationships themselves can also be described as concept types, and they can be classified into more complex relationships. For example, the association relationship "marrying..." can include the relationship concepts "marrying" and "with."

In some embodiments, the overall organization of the AKRM data model, stored as the base data structure 120 in the system 100, can be encoded as a faceted data structure, in which conceptual entities are explicitly related in a hierarchy (external relationships) and joined in sets to create complex concepts (internal relationships). Additionally, as discussed above, these external and internal relationships themselves can be categorized using concepts. However, it should be appreciated that any suitable type of knowledge representation model or theoretical construct (including any suitable type of concept relationships) can be utilized in representing the AKRM, as aspects of the present invention are not limited in this respect.

As an example, FIG3 provides an exemplary data plan 350 that can be used in the data set 110 of the system 100 according to some embodiments of the present invention. Such a data plan can be designed to be able to encode complex knowledge representation data structures (complex KR) (such as ontologies and taxonomies) and the atomic knowledge representation data structures (e.g., basic data structures 120) into which the complex KR is decomposed. In the plan 350, concepts can be joined to form more complex types (with types) using many-to-many relationships. In this way, the core concept entities in the model can represent a wide variety of simplicity or complexity based on the properties of the complex knowledge representation modeled by the data. By joining symbols, rules, and objects to these concepts using many-to-many relationships, such a plan can manage data for modeling a wide range of knowledge representations.

In the schema 350 shown in FIG3 , the rectangular boxes represent entity sets (e.g., real-world objects that can be encoded as primary objects in a database), as well as abstract concepts, abstract human- and/or machine-readable symbols that reference the concepts, and rules that apply to the concepts in the knowledge representation. Each solid connector represents a relationship between two entity sets, of a relationship type such as that represented by the diamond. “N” represents the cardinality of the relationship; here, the relationship is many-to-many, indicating that many entities of each entity set can participate in a relationship with entities of the other entity set that participate in the relationship, and vice versa. In contrast, a relationship labeled “1” on both sides of the diamond would represent a one-to-one relationship; a relationship labeled “1” on one side and “N” on the other side would represent a one-to-many relationship, where one entity of the first type can participate in a relationship with many entities of the second type, while each entity of the second type can participate in the relationship with only one entity of the first type; and so on.

In some embodiments, the data structure of the knowledge representation may be encoded in one or more database tables using any suitable database and/or other data encoding techniques according to schema 350. For example, in some embodiments, a data set for a KR data structure may be structured as a computer-readable representation of a table in which each row represents a relationship between a pair of concepts. For example, an example of a data table may have four attribute columns, including a "Concept 1" attribute, a "Concept 2" attribute, a "Relationship" attribute, and a "Type" attribute, thereby modeling a three-way relationship for each row of the table as: "Concept 1 is related to Concept 2 by a relationship concept of type (e.g., extrinsic or intrinsic). For example, a row of such a table with the attributes (column entities) {Concept 1: "hammer"; Concept 2: "nail"; Relationship "Tool"; Type: "External"} may represent the relationship: "Hammer" is related to "Nail" as a "Tool", and the relationship is "External". In many exemplary data structures, each concept may appear in one or more rows of the database table, e.g., appearing in multiple rows to represent relationships with multiple other concepts. Furthermore, a particular pair of concepts may appear in multiple rows, for example, if the pair is related by multiple types of relationships. However, it should be appreciated that the foregoing description is by way of example only and that the data structure may be implemented and/or encoded and stored in any suitable manner, as aspects of the present invention are not limited in this respect.

In some embodiments, various metadata can be associated with each entity (e.g., concepts and concept relationships) within the AKRM to support rule-based programming. For example, since many rules will require a ranked set of concepts, the priority of concepts within concept relationships (internal or external) can be added to this schema. These details are omitted here only to simplify the presentation of the data model.

While the exemplary data plan of FIG3 may be relatively simple, when it is coupled with processing rules for constructing and deconstructing knowledge representations, it can become capable of managing a wide range of complex knowledge (as described in various examples below). Benefits may include real-time knowledge engineering for improving data economy and reducing the need to build complexity into large knowledge representation data structures. Additionally, since the scope of the knowledge representation data structure is reduced, it may also have a beneficial impact on integrated knowledge engineering processes such as reasoning, analysis, data mining, and search.

1 , in some embodiments, knowledge processing rules 130 may be encoded and persisted in the system 100 (e.g., in the data set 110) and may be coupled to the input KR 160 and/or concepts within the base data structure 120. Rules may be coupled to concepts such that, given a specific concept, the rules may be applied by executing programming code by one or more processors of the system 100 to generate new semantic entities (concepts and relations) from the base data structure 120 and/or deconstruct the input KR 160 into base entities to be included in the base data structure 120. Examples of such rules are described in more detail below.

For example, a developer of the system 100 and/or an end user of the system 100 may introduce rules 130 as input rules 140 to the data set 110 based on their individual knowledge processing needs or preferences. It should be appreciated that the input rules 140 may be obtained from any appropriate source at any appropriate time, and that any appropriate user may update and/or change the rules 130 stored as part of the AKRM at any appropriate time before or during operation of the system 100, and that different stored rules 130 may be maintained for different users or applications interacting with the system 100, as aspects of the present invention are not limited in this respect. Furthermore, in some embodiments, different subsets of the stored rules 130 may be applied to analyzing the input KR 160 rather than synthesizing the output KR 190, while in other embodiments, the same rules 130 may be applied in both analysis and synthesis operations, and different subsets of the stored rules 130 may be applied to different types of knowledge representations.

The rules 130, when applied to concepts in the analysis and synthesis of KR, can provide construction and deconstruction logic for a system (such as the system 100). The method of how to create (synthesize) or deconstruct (analyze) knowledge can be encoded in the rule set 130. The rules 130 can be designed to work systematically (a single rule operates in both analysis and synthesis) or asymmetrically (where a single rule is designed to work only in synthesis or analysis). In some embodiments, the rules 130 may not be encoded as entities within the conceptual data structure of the knowledge model, but rather as rules within the knowledge representation model that operate on the conceptual data structure in a generative capacity. In some embodiments, the rules 130 can be encoded as data in a machine-readable encoding of the AKRM that includes the rules and stored together with the knowledge representation data structure (such as the base data structure 120). The rules 130 may be applied using a rules engine software component implemented, for example, by program instructions encoded in one or more tangible, non-transitory computer-readable storage media included in or accessible by the system 100, the programming instructions being executed by one or more processors of the system 100 to provide a rules engine.

Given the probabilistic nature of a system (such as system 100 according to some embodiments of the present invention), a method for verifying semantic consistency of a knowledge representation data structure generated from the application of rules 130 can be performed. In some embodiments, system 100 can be programmed to collect evidence about whether the resulting data structure exists in existing knowledge models. These existing knowledge models can be internal to the system (as complex knowledge representation data structures) or external (such as knowledge models encoded on the semantic web). In some embodiments, a search engine can be used to investigate whether terms (symbols or tags) associated with the concepts of the resulting data structure exist in external knowledge representations (such as documents). Term-document frequency (e.g., the number of search engine hits) can provide an exemplary measure of semantic consistency for the resulting knowledge representation data structure. However, it should be appreciated that any appropriate measure of semantic consistency for such data structures can be used, as aspects of the present invention are not limited in this respect.

The analysis engine 150 and the synthesis engine 170 may use any of a variety of semantic analysis and synthesis methods to support the construction and deconstruction of knowledge representation data structures, as aspects of the present invention are not limited in this regard. Examples of analysis methods that can be used by the analysis engine 150 in conjunction with the application of rules 130 when deconstructing the input complex KR 160 include text analysis, entity and information extraction, information retrieval, data mining, classification, statistical clustering, linguistic analysis, facet analysis, natural language processing, and semantic knowledge bases (e.g., dictionaries, ontologies, etc.). Examples of synthesis methods that can be used by the synthesis engine 170 in conjunction with the application of rules 130 when constructing the complex KR 190 include formal concept analysis, facet classification synthesis, semantic synthesis, and dynamic taxonomy.

It should be appreciated that the exemplary analysis and synthesis methods of complex KRs may be performed by the analysis engine 150 and the synthesis engine 170 operating individually and/or in conjunction with any suitable external software applications that may interface with the analysis engine 150 and the synthesis engine 170 and/or the system 100. Such external software applications may be implemented within the same physical device or set of devices as the other components of the system 100 or portions or all of such software applications may be implemented in a distributed manner in communication with other separate devices, as aspects of the present invention are not limited in this respect.

FIG4 illustrates an exemplary method 400 of semantic analysis that can be used by the analysis engine 150 when deconstructing the input complex KR 160. It should be appreciated that the method shown in FIG4 is merely an example and, as discussed above, many other analysis methods are possible, as aspects of the present invention are not limited in this regard. The exemplary method 400 begins by extracting source concepts 410 having text concept labels that are explicitly present in a source data structure. A plurality of source concepts 410 can be extracted from the source data structure along with source concept relationships between the source concepts 410 that can be explicitly present in the source data structure.

A series of keyword delineators can be identified in the concept tags for the source concepts 410. A preliminary keyword scope can be parsed from the concept tags based on common structural text delineators for keywords (such as brackets, quotes, and commas). Complete words can then be parsed from the preliminary keyword scope using common word delineators (such as spaces and grammatical symbols). A check for single word independence can then be performed to ensure that the parsed candidate keywords are valid. In some embodiments, the check for word independence can be based on a method of stem (or root) matching referred to as "stemming" below. Once valid, a word can define a keyword if it exists in a concept tag with other words and exists in related concept tags without those other words.

Once a preliminary keyword tag set has been generated, all preliminary keyword tags can be aggregated to identify the composite keywords that exist within multiple valid keyword tags within a single concept tag. In some embodiments, recursion can be used to exhaustively split the composite keyword set into the most basic keyword set supported by the source data. The process of refining, taking effect, and splitting the candidate keywords can be repeated until no more atomic keywords can be found.

In some embodiments, the last round of merging methods can be used to disambiguate keyword tags across the entire domain. Such disambiguation can be used to eliminate the ambiguity that occurs when entities share the same tag. In some embodiments, disambiguation can be provided by merging keywords into a single structural entity that shares the same tag. The result can be a set of keyword concepts, each of which is included in the source concept from which it is derived. For example, source concept 410 can be decomposed into keywords 420, 440 and 460 parsed from its concept tags, and keywords 420, 440 and 460 can form the concept definition for source concept 410. For example, in the example basic data structure 300 of Figure 2 B, a more basic concept 255 labeled as "domestic" can be deconstructed from a more complex concept 250 labeled as "domestic dog" as a keyword parsed from the concept tag.

In certain embodiments, the concept definition that comprises keyword concept can be expanded to comprise morpheme concept entity as deeper and more basic refinement level in their structure by further deconstruction.In certain embodiments, morpheme can represent the basic irreducible attribute of more complex concept and relation thereof.At the morpheme refinement level, many attributes will not be identified as concept by human classifier.Yet, when being combined into relational data structure across whole domain, morpheme can in some embodiments, can use less information to carry the meaning implication of more complex concept.

In some embodiments, the morpheme extraction method may have common elements with the keyword extraction method discussed above. Patterns may be defined to serve as criteria for identifying morpheme candidates. These patterns may establish parameters for stemming and may include patterns for whole words as well as partial word matching. As with keyword extraction, the source concept relationship set may provide context for morpheme pattern matching. The pattern may be applied to a pool of keywords within the source concept relationship set in which the keyword appears. A shared root set based on the stemming pattern may be identified. The shared root set may include a set of candidate morpheme roots for each keyword.

In certain embodiments, the candidate morpheme root set for each keyword can be compared to ensure that they are consistent with each other. It can be assumed that the resident root and the source concept relationship set in which the keyword appears in the context of the same keyword have overlapping roots. In addition, it can be assumed that the basic root derived from the intersection of those overlapping roots will remain in the parameter used to identify the effective morpheme. Such entry into force can constrain the splitting of excessive morphemes and provide a meaningful and basic level of refinement of the context. In certain embodiments, any inconsistent candidate morpheme root can be removed from the keyword set. The pattern matching process for identifying morpheme candidates can be repeated until all inconsistent candidates have been removed.

In some embodiments, by checking potential root groups, one or more morpheme delimiters can be identified for each keyword. Morphemes can be extracted based on the position of the delimiter in each keyword tag. Then the keyword concept definition can be constructed by correlating (or mapping) the extracted morphemes to the keywords from which they are derived. For example, morpheme concepts 425 and 430 can be included in the concept definition for keyword concept 420, morpheme concepts 445 and 450 can be included in the concept definition for keyword concept 440, and morpheme concepts 465 and 470 can be included in the concept definition for keyword concept 460. Therefore, the original source concept 410 can be deconstructed into the keyword concept level and further deconstructed into the most basic morpheme concept level included in the basic data structure of AKRM by semantic analysis.

However, it should be appreciated that any appropriate level of abstraction can be employed when generating the base data structures, and any appropriate analysis methods (including methods that are not keyword- or morpheme-centric) can be used, as aspects of the present invention are not limited in this respect. In some embodiments, the base data structures included in the AKRM for use in analyzing and/or synthesizing more complex KRs can include and encode concepts and relationships that are more basic than the concepts and relationships included in the complex KRs that are deconstructed to populate the base data structures and/or synthesized from the base data structures. For example, the abstract meanings of complex concepts encoded in the complex KR can be formed by combining the abstract meanings of the base concepts encoded in the base data structures of the AKRM.

In some embodiments, concepts stored in a basic data structure as part of a centralized AKRM may have been deconstructed from more complex concepts to the level of single complete words (such as keywords). The example of Figure 2B illustrates such a basic data structure encoding a single complete word. In some embodiments, concepts in a basic data structure may have been deconstructed to a more basic level representing parts of words. In some embodiments, concepts in a basic data structure may have been deconstructed to a more basic semantic level represented by morphemes, which are the smallest linguistic units that can still carry semantic meaning. For example, the whole word concept "Siamese tribe" can be deconstructed to create two morpheme concepts "Siamese" and "tribe" with "Siamese" representing a free morpheme and "tribe" representing an affix. In some embodiments, the basic data structure of an AKRM may include only concepts at a specified basicity level; for example, the basic data structure may in some embodiments be formed entirely of morphemes or entirely of single-word concepts. In other embodiments, the basic data structure may include concepts at various levels of basicity (e.g., including morpheme concepts, keyword concepts, and/or other concepts at other levels of basicity), and at least some of the concepts in the basic data structure are more basic than the complex concepts in the input KR from which they are deconstructed and/or the complex concepts in the output KR created by combining them with other basic concepts. It should be appreciated that any suitable basis for deconstructing complex KRs into more basic data structures (including basis connected to paradigms other than language and semantics) may be utilized, as aspects of the present invention are not limited in this respect.

1 , data consumers 195 can represent one or more human users of system 100 and/or one or more machine-implemented software applications that interact with system 100. In some embodiments, data consumers 195 can make requests and/or receive output from system 100 in various data forms. For example, data consumers 195 can input complex KRs 160 to system 100 for deconstruction into basic concepts and concept relationships to generate and/or update basic data structures 120. Data consumers 195 (the same or different data consumers) can also receive output complex KRs 190 from system 100 that are synthesized by applying one or more knowledge processing rules 130 to part or all of basic data structures 120.

In some embodiments, data consumer 195 may also provide context 180 for guiding synthesis and analysis operations. For example, by inputting a specific context 180 along with a request for output KRs, data consumer 195 can direct system 100 to generate output KRs 190 with appropriate characteristics for the desired information or current task being performed by the data consumer. For example, data consumer 195 may input specific context 180 as a search term that can be mapped to a specific concept about which consumer 195 requires or wants to receive relevant information. Synthesis engine 170 may, for example, apply rules 130 only to those portions of base data structure 120 that are conceptually related (i.e., connected in the data structure) to the concept corresponding to context 180. In another example, input context 180 may indicate that data consumer 195 desires a specific type of knowledge representation model, such as a taxonomy, to which output KRs 190 conform. Thus, synthesis engine 170 may apply only those rules of rule set 130 that are suitable for synthesizing a taxonomy from base data structure 120.

It should be appreciated that the input context 180 can include any number of requests and/or constraints that apply to synthesizing the output KR 190, and that the composition of the input context 180 can be of any suitable type encoded in any suitable form of data or programming language, as aspects of the invention are not limited in this respect. Examples of suitable input contexts include, but are not limited to, free text queries and submissions, and structured inputs (such as terminology or tag sets consistent with various Web 2.0 systems, such as those mediated by natural language processing (NLP) techniques. In some embodiments, generating the output KR 190 based on a particular context 180 can enable a more fluid and dynamic knowledge exchange with data consumers. However, it should be appreciated that the input context 180 is not required, and that the system 100 can, in some embodiments, generate the output KR 190 without an input context, as aspects of the invention are not limited in this respect.

The data consumer 195 may also provide any suitable type of input KR 160 to the system 100 in any suitable form using any suitable data encoding and/or programming language, as aspects of the present invention are not limited in this respect. Examples of suitable forms of input KR include, but are not limited to, semi-structured or unstructured documents and structured knowledge representations (such as taxonomies, controlled vocabularies, faceted classifications, and ontologies) also used with various forms of NLP and text analysis.

In some embodiments according to the present disclosure, a system for analyzing and synthesizing complex KRs using AKRMs (such as system 100) can be implemented on the server side of a distributed computing system in network communication with one or more client devices, machines, and/or computers. FIG5 illustrates such a distributed computing environment 500, in which system 100 can operate as a server-side transformation engine for KR data structures. The transformation engine can take as input one or more source complex KR data structures 520 provided by a client 510, for example, through actions of a human user of the client 510 or a software application from one or more domains. In some embodiments, the input complex KRs 520 can be encoded into one or more XML files 530 that can be distributed via a network (such as the Internet 550) via a web service (or API or other distribution channel) to the computing system on which system 100 is implemented. Similarly, system 100 can return the requested output KRs as XML files 540 to various clients 510 via the network. However, it should be appreciated that data may be communicated between server system 100 and client system 510 in any suitable manner and in any suitable form, as aspects of the invention are not limited in this respect.

Through this and/or other distribution and decentralization models, in some embodiments, a wide range of developers and/or publishers can use the analysis engine 150 and the synthesis engine 170 to deconstruct and create complex KR data structures. Exemplary applications include, but are not limited to, websites, knowledge bases, e-commerce stores, search services, client software, management information systems, analytics, etc.

In some embodiments, an advantage of such a distributed system can be a clear separation between private domain data and shared data used by the system to process a domain. This data separation can facilitate hosted processing models (such as a Software as a Service (SaaS) model), whereby a third party can provide transformation engine services to a domain owner. A domain owner's domain-specific data can be securely hosted by SaaS because it can be separated from other domain owners' shared data (e.g., AKRM dataset 110) and private data. Alternatively, the domain-specific data can be hosted by the domain owner, physically removed from the shared data. In some embodiments, domain owners can build on the shared knowledge (e.g., AKRM) of the entire user community without compromising their unique knowledge.

As should be appreciated from the foregoing discussion, some embodiments according to the present disclosure relate to techniques for analyzing existing complex knowledge representations to deconstruct complex KRs and generate or update the basic data structures of atomic knowledge representation models. FIG6 illustrates one such technique as an exemplary process 600. Process 600 begins at action 610, where, for example, an analysis/synthesis system (such as system 100) may receive an input complex KR from a data consumer.

In action 620, one or more knowledge processing rules encoded in the system 100 as part of the AKRM can be applied to deconstruct the input complex KR into one or more basic concepts and/or one or more basic concept relationships. Examples of knowledge processing rules applicable to various types of input KRs are provided below. However, it should be appreciated that aspects of the present invention are not limited to any particular example of knowledge processing rules and any appropriate rules encoded in association with the atomic knowledge representation model can be utilized. As discussed above, such rules can be provided at any appropriate time by the developer of the analysis system and/or by one or more end users of the analysis system.

In act 630, one or more of the basic concepts and/or basic concept relationships discovered and/or derived in act 620 may be included in a basic data structure encoded and stored as part of the system's AKRM. In some embodiments, some or all of the basic concepts and relationships derived from a single input complex KR may be used to populate a new basic data structure of the AKRM. In some embodiments, once the stored basic data structure has been populated, new basic concepts and/or relationships discovered from subsequent input KRs may be included in the stored basic data structure to update and/or expand the centralized AKRM. In some embodiments, process 600 may continue to loop back to the beginning to further update the stored basic data structure and/or generate new basic data structures as new input KRs become available. In other embodiments, process 600 may terminate after a single pass or another predetermined number of passes through the process, after the stored basic data structure has reached a predetermined size or complexity, or after any other suitable stopping criteria are met.

As should be appreciated from the foregoing discussion, further embodiments according to the present disclosure relate to techniques for generating (synthesizing) complex knowledge representations using an atomic knowledge representation model. FIG7 illustrates such a technique as an exemplary process 700. Process 700 begins with action 710, where an input context may be received, for example, from a data consumer (such as a human user or a software application). As discussed above, such context may include a text query or request, one or more search terms, identification of one or more seed concepts, and the like. In addition, the context may indicate a request for a particular form of complex KR. However, in some embodiments, a request for a complex KR may be received without further context for limiting the concepts and/or concept relationships to be included in the complex KR, as aspects of the present invention are not limited in this regard. Additionally, in some embodiments, a received context may be interpreted as a request for a complex KR without an explicit request to accompany the context.

In action 720, in response to the input request and/or context, one or more appropriate knowledge processing rules encoded in the AKRM may be applied to the basic data structure of the AKRM to synthesize one or more additional concepts and/or concept relationships that are not explicitly encoded in the basic data structure. Examples of knowledge processing rules that may be applicable to synthesizing various types of output KRs are provided below. As discussed above, in some embodiments, rules may be applied bidirectionally to achieve both analysis and synthesis of complex KRs using the same knowledge processing rules, while in other embodiments, one rule set may be applied to analysis and a different rule set may be applied to synthesis. However, it should be appreciated that aspects of the present invention are not limited to any particular example of knowledge processing rules, and any appropriate rules encoded in association with an atomic knowledge representation model may be utilized. As discussed above, such rules may be provided at any appropriate time by the developer of the analysis system and/or by one or more end users of the analysis system.

In some embodiments, appropriate rules can be applied to appropriate parts of the basic data structure based on the received input request and/or context. For example, if the input request specifies a specific type of complex KR to be output, then in some embodiments, only those rules encoded in the AKRM that are applied to synthesizing complex KRs of that type can be applied to the basic data structure. In some embodiments, if a specific type of complex KR is not specified, a default type of complex KR (such as a taxonomy) can be synthesized, or a random type of complex KR can be selected. For example, if the input context specifies one or more specific seed concepts of interest, only those parts of the basic data structure that are related to those seed concepts (i.e., connected by concept relationships) can be selected and rules applied to them to synthesize new complex KRs. In some embodiments, for example, the developer or end user of the synthesis system can set a predetermined limit on the size and/or complexity of the output complex KR, such as the number of concepts included, the hierarchical distance between the seed concepts and the selected related concepts in the basic data structure, the size of the encoded data of the resulting output complex KR, processing requirements, etc.

In act 730, a new complex KR can be synthesized from the additional concepts and relationships synthesized in act 720 and the selected appropriate portion of the base data structure and encoded according to any specified type of KR indicated in the received input. In act 740, the resulting synthesized complex KR can be provided to the data consumer from whom the request was received. As discussed above, this can be, for example, a software application or a human user who can view and/or utilize the provided complex KR through a software user interface. Process 700 can then conclude by providing the newly synthesized complex KR that encodes the new knowledge.

The following pseudo-code segment may be used as a further illustration of the method described above.

Knowledge creation (KR _in , rules _in , context, analysis, and synthesis)

enter:

-CONTEXT: User/application context (e.g., request, seed concept, domain constraints)

-KR _in : Knowledge Representation (e.g., taxonomy)

-RULES: relevant knowledge processing rules

-ANALYSIS: Flag to enable profiling events

-SYNTHESIS: Flag to enable synthetic events

Output:

-Concepts and relationships to be stored in AKRM

- Complex KR _out for presentation to users/applications

process:

As should be appreciated from the foregoing discussion, some embodiments according to the present disclosure relate to techniques for supporting semantic interoperability between knowledge representations using an atomic knowledge representation model. As discussed above, maintaining a shared centralized AKRM with a stored base data structure can, in some embodiments, allow multiple different input complex KRs (in some cases of different types of knowledge representation models) to be deconstructed into base concepts and/or concept relationships used in generating and/or updating a single shared base data structure that is semantically compatible with all types of complex KRs. Furthermore, by deconstructing into base data structures and subsequently synthesizing into new complex KRs, one type of input KR can, in some embodiments, be transformed into different types of output KRs based on the same source data.

The following pseudo-code may serve as a further illustration of a method for integrating multiple different KRs under the AKRM described herein to provide the benefits of semantic interoperability.

enter:

-KR ₁ , KR ₂ , ..., KR _n : /*n possible different KRs*/

-RULES ₁ , RULES ₂ , ..., RULES _n /*Related knowledge processing rules*/

- User/Application Context

Output:

-Concepts and relationships to be stored in AKRM

- Complex KRs for presentation to users/applications

process:

Figure 8 provides the following table, which illustrates six exemplary knowledge processing rules that can be used when analyzing and/or synthesizing five exemplary types of complex knowledge representations (i.e., taxonomies, synonym rings, dictionaries, faceted classifications, and ontologies) in some embodiments according to the present disclosure. However, as discussed above, it should be appreciated that these examples are provided for illustration purposes only, and aspects of the present invention are not limited to any particular rule or KR type or model set. In addition, in some embodiments, for example, a developer of a system can seed an analysis/synthesis system with an initial set of knowledge processing rules, which can be expanded with additional rules and/or updated at a later time with changed and/or deleted rules, for example, by an end user of the system. Different rule sets applicable to different types of KRs can also be stored, for example, in user accounts, for different end users or applications. In addition, in some embodiments, knowledge processing rules can be reused and combined in novel ways to address requirements for specific KRs.

Below with reference to being related to the concrete example of the exemplary KR type that provides in the accompanying drawings, discuss the exemplary rule that presents among Fig. 8.Should be appreciated that any method in the generalized method described above can be applied to any example in the following example and relate to different inputs, outputs and knowledge processing rules.Also should be appreciated that, although can model many different aspects of knowledge creation theory by the exemplary rule discussed here, the rule of various other types is possible.Following example is mainly driven by the topology of knowledge representation data structure.Other basis that is used for rule can comprise semantic morphology and grammar, phonetic system, metaphor, symbol and sensory perception and other basis.

In some embodiments, encoding a set of knowledge processing rules within an atomic knowledge representation model (such as the exemplary rules given in Figure 8) can allow analysis and/or synthesis of any complex KR within a supported set of KR types (such as those represented in Figure 8). In the example of Figure 8, the "X" indicates which rules of the exemplary set of six rules are applied to which KR types of the exemplary set of five KR types. In these examples, the rules can be applied bidirectionally when analyzing or synthesizing complex KRs of the type to which each rule is applied. For example, when an input dictionary KR is given, Figure 8 shows that rules 1, 2, 3, and 4 can be applied to the input dictionary to deconstruct it into basic concepts and concept relationships to be included in the basic data structure. In another example, applying rules 1, 2, and 3 to the basic data structure results in outputting a synonym ring KR. The following describes the use of each of these exemplary rules to perform analysis and/or synthesis of appropriately complex KRs with reference to examples.

Classification rules

The following import/export and knowledge processing rules provide the characteristics of the taxonomy as a hierarchical classification of concepts.

Input/Output

Concept Set C

Hierarchical relation sets (acyclic)

R={r(c _i , c _j ): c _i , c _j ∈C and c _i Is-a c _j }

Definition 1 (Consistent Concepts): Two concepts c _i , c _j are considered congruent if, according to some distance metric M , M(c _i , c _j ) < T , where T is a preselected threshold. Possible metrics include the co-occurrence frequency of the two concepts in the input corpus, or a tree distance function applied to the taxonomy hierarchy.

Rule 1 (Consistent Concept Composition): Create a new concept c = { _ci , _cj }. If and only if _ci and _cj are consistent with respect to Definition 1, then c is considered to include _ci and _cj .

Rule 2 (One-Concept Composition): Let c ₁ = {c ₁₁ , c ₂₂ , ... c _1n } be a concept consisting of n concepts c ₁₁ to c _1n . Similarly, let c ₂ = {c ₂₁ , c ₂₂ , ... c _2m } be a concept consisting of m concepts c ₂₁ to c _2m . If and only if for each c _1i there exists a relation r(c _1i , c _2j ) for some concept c _2j , then create a new hierarchical relation r(c ₁ , c ₂ ).

Note that the if-and-only-if portion of each exemplary rule (e.g., Rule 1 and Rule 2) reflects the bidirectional analysis/synthesis nature of the rule. For example, analysis will enforce the "if" portion (forcing the explicit hierarchical relationship to be present in the AKRM to satisfy the condition). On the other hand, synthesis will discover the "only if" portion (discovering the hierarchical relationship if the condition applies).

An example of applying these exemplary rules to analyze and deconstruct an input taxonomy 200 into a more basic data structure 300 is provided in Figures 2A and 2B. In this example, complex concepts 230, 250, and 270 are deconstructed to generate new more basic concepts 235, 240, 255, 260, 275, and 280 by applying Rule 1, and their relationships are generated by applying Rule 2. Furthermore, new complex concepts are synthesized by applying Rule 1, for example, using an external corpus as evidence: {domestic, lion}, {mountain, dog}, {mountain, cat}, {domestic, goat}, {domestic, pet}, {domestic, cat}. Applying Rule 2 in the synthesis can generate new concept relationships; for example, since hierarchical relationships exist between "animal" and "dog" and between "animal" and "mountain," a new hierarchical relationship between "animal" and "mountain dog" can be synthesized.

Synonym Ring Rule

The following input/output and knowledge processing rules provide for the characterization of synonym rings as defined by proximity of meaning across terms or concepts or logically internal substitutability of terms for truth-preserving terms.

Input/Output:

Concept set C (may have an "include" relationship)

Synonym list: synonyms ( _ci , _cj )

Definition 2 (Semantic Similarity): Let c ₁ = {c ₁₁ , c ₂₂ , ... c _1n } be a concept consisting of n concepts c ₁₁ to c _1n . Similarly, let c ₂ = {c ₂₁ , c ₂₂ , ... c _2m }. The similarity function S, S(c ₁ , c ₂ ) describes the semantic similarity between two concepts. An example function is as follows:

(if: if; Synonym: synonym; otherwise: otherwise)

Definition 3 (Concept Intersection): Let c ₁ ={c ₁₁ , c ₂₂ , ...c _1n } be a concept including n concepts c ₁₁ to c _1n . Similarly, let c ₂ ={c ₂₁ , c ₂₂ , ...c _2m }.

(if: if)

Definition 3 (Synonym Concept Composition): Let c ₁ = {c ₁₁ , c ₂₂ , ... c _1n } and c ₂ = {c ₂₁ , c ₂₂ , ... c _2m } be two synonym concepts according to Definition 2. Concept c ₃ = c ₁ ∩ c ₂ and hierarchical relations r(c 1 , c 3 ) and r(c 2 , c 3 ) exist if and only if _{S(c 1 , c 2 ) >} T _synonym (T _synonym ), where T _synonym is the threshold of semantic similarity for the assertion “synonym” to hold:

Synonym：：=c ₃ ＝c ₁ ∩c ₂ ≠φ∧r(c ₁ , c ₃ )∧r(c ₂ , c ₃ )

S(c ₁ ，c ₂ )＞T _synonym

(Synonym: synonym; T _synonym : T _synonym )

Examples of synonym rings are as follows:

Pets: Domestic Animals: Household Beasts: Cats

Analysis according to Rule 3 derives hierarchical relationships by which all four concepts are children of "household animals." Analysis according to Rule 1 derives the following new concepts:

residence, domestic, household, animal, beast, mammal

Analysis based on Rule 2 reveals a hierarchy in which "domestic" and "home" are children of "residence," and "pet," "mammal," "beast," and "cat" are children of "animal." These hierarchical relationships can be created based on relationships between complex concepts from which simpler concepts are extracted. Thus, the following new synonym ring can be synthesized by applying Rule 3:

Cat: Pet: Mammal: Wild Beast

Domestication: Home

Dictionary rules

The following input/output and knowledge processing rules provide the characteristics of the dictionary (including the characteristics of KR described above) and the relationship (related terms)

Input/Output:

Concept set C (may have an "include relationship")

A list of association relationships, such as synonyms ( _ci , _cj ), related terms ( _ci , _cj ).

Hierarchical relation set (acyclic) R = {r( _ci , _cj ): _ci , _cj∈C and _{ciNTcj }}

Rule 1 (consistent concept synthesis) is applied to the dictionary.

Rule 2 (hierarchical concept synthesis) is applied to the dictionary.

Rule 4 (Association Relation Synthesis): Let c ₁ = {c ₁₁ , c ₂₂ , ... c _{1n }} and c ₂ = {c ₂₁ , c ₂₂ , ... c _2m } be two related concepts according to some association relation AR. If and only if S(c ₁ , c ₂ ) > T _AR , then concept c ₃ = c ₁ ∩ c ₂ , c ₄ = {AR} and three hierarchical relations r(c ₁ , c ₃ ), r(c ₂ , c ₃ ) and r(c ₄ , c ₃ ) exist, where T _AR is the threshold of semantic similarity for the assertion of an “AR” relation between two concepts:

Association

Note that _TAR can be set to zero if semantic similarity is not required and the association via _c3 is sufficient to capture the relationship.

The example dictionary may include an association relationship: {cat, recipe} is associated with {fish, food}. Based on the analysis of Rule 1, the following new concepts can be derived:

cat, recipe, fish, food

Given appropriate patterns in the presented hierarchical relationships, new associations can be synthesized by applying Rule 4, e.g., "cat" is associated with "fish" and "recipe" is associated with "food." Similarly, associations can be created based on relationships between complex concepts from which simpler concepts are extracted.

Classification rules by surface

The following input/output and knowledge processing rules provide features for facet classification (including faces and face attributes as concepts) and faces as concept categories organized in a class hierarchy. Furthermore, the following example adds features for mutually exclusive facet hierarchies (face attributes constrained to a strict/single hierarchy, single inheritance), and assigns face attributes to objects (or nodes) that are to be classified as concept sets. Furthermore, faces are topologically identified as root nodes in the facet hierarchy.

Input/Output:

Face classification (classification of value nodes for each root face)

Terms/concepts regarding face value marking

Definition 4 (mutually exclusive facet classification): Any concept can be classified by picking one and only one node label/value/attribute from each facet classification. That is, the semantics of the concepts representing the nodes in any facet classification do not overlap.

Rules 1, 2, and 4 are applied to face classification.

Rule 5 (Aspect Attribute Assignment): Every node/value/attribute in the aspect hierarchy corresponds to a concept c. A relation r(ci, cj ₎ exists if and only if _ci appears as a child of only one parent _cj in some aspect hierarchy, and _{if c1∩c2} = {} for any two concepts _c1 , _{c2 in} the aspect hierarchy.

Rule 6 (Concept assignment of annotations): Each annotated term in the faceted classification corresponds to a concept c _i ={c _i1 , c _i2 , . . . c _in }, where c _ij is a label concept according to Rule 5.

The example input is distributed by face as follows:

Object with face attributes/nodes/value assignments

"domestic dog" {North American, domesticated, dog}

"Mountain Lion" {American, wild, cat, mountain dwelling}

"Siamese Cat" {World, Domesticated, Cat}

"lion" {Africa, wild, lion, grassland}

As shown in the example above, analysis according to rules 2 and 5 can be used to decompose the input facet classification into a broader facet hierarchy (eg, using facet analysis or statistical clustering methods).

Surface: "Pet" /*Crafting Tag*/

- "Common Pets" /* derived from cluster {domesticated, animals} */

- "rare pets" /* derived from cluster {wild,animals} */

Since both “dog” and “cat” are “animals” (derived from the facet classification “animal”), it can be found that the new concept “domesticated, animal” is clearly consistent in sets such as “domesticated, dog” and “domesticated, cat”.

Similarly, new objects with facet attributes/nodes/value assignments can be created according to rules 1 and 6. For example, using the rules for concept synthesis described above, new concepts such as "lion pet" {artificial, lion, domesticated} can also be synthesized. Although this may not exist in real life, it can be confirmed as possible new knowledge given the evidence in the input KR and later evaluated (for example) through user interaction with the data.

Ontological rules

Rules 1, 2, 4, 5, and 6 are applied to provide the characteristics of the topology (including faces as concepts and face attributes) and faces as categories of concepts organized in a class hierarchy.

Consider the example complex relationship Cohabitation (COH):

Wild cat ← COH → lion

Domestic dog ← COH → Domestic cat

Analyzing COH relations allows breaking them down into more atomic relations and concepts. The following atomic constructs are possibilities:

Wild cats, lions, domestic dogs, domestic cats, cohabitation

The above-described rules for knowledge creation can be applied in a complex manner to represent richer relationships (e.g., _C1 and _C2 ), where the relationship is a general associative relationship. For complex relationships that are associative relationships (bidirectional), the intersection property of the meanings between the paired concepts in the relationship can be utilized. For hierarchical (unidirectional) relationships, the inclusion property of the meanings between the paired concepts in the relationship can be utilized. The labels derived for the synthesized complex relationships can conform to conventional presentation, such as "C1 and C2 are related because they share C3."

Applying Rule 1 (consistent concept composition) and Rule 4 (association relationship composition) can produce the following more atomic concepts:

Wild, cat, dog, domestic, habitat, wild habitat, domestic habitat, "wild habitat" belongs to habitat, "domestic habitat" belongs to habitat

If agreement is found, synthesis can construct the following concepts and relationships:

"Wild dogs" includes {wild, dogs, wild habitats}

Therefore, the following higher-order relations can be deduced:

Wild Dog ←COH→Lion

Feral Dog ← COH → Feral Cat

Thus, both "wild dog" and the relationship with "lion" and "wild cat" are newly synthesized constructs.

Free text (natural language) examples

The following is an example of natural language text that can be transformed into a structured semantic representation using approaches such as natural language processing, entity extraction, and statistical clustering.Once transformed, the exemplary rules described above can be applied to process the data.

The cat (Felis silvestris catus) (also known as the domestic cat or house cat to distinguish it from other cats and felines) is a small carnivorous mammal cherished by humans for its companionship and its ability to hunt parasites and household pests. Cats have been associated with humans for at least 9,500 years and are currently the world's most popular pets. Cats are now nearly ubiquitous on Earth due to their close association with humans.

From this natural language text, a structured knowledge representation as shown in FIG9 can be derived. This knowledge representation can be processed using the rules described under each exemplary knowledge representation type as follows:

Classification: C1 belongs to C5 (grading)

Synonym ring: C1:C2:C3

Dictionary: C1 is associated with C7.

Ontology: C1 hunts C6; C1 is found on C7

Applying composition to this example, additional structured data can be derived. For example, applying Rule 1 (consistent concept composition) can derive additional concepts:

C8: Domestic

C9: Residential

Then, the new relation can be synthesized, for example, by applying Rule 3 (Synonym Concept Synthesis):

C8::C9 ("domesticated" is a synonym for "residential")

Semantic Interoperability Example

The following example illustrates semantic interoperability, where input in one KR can be transformed into a different KR as output.The exemplary process described below can be implemented, for example, according to the general data flow of the pseudocode presented above for the semantic interoperability process.

Input (Input KR is a dictionary; :: indicates a synonym for ...; |- indicates a narrower meaning.)

The basic data structure that can be analyzed from the above-mentioned input KR is illustrated in Figure 10. In this figure, the solid arrows represent "belong to" relationships, and the dashed arrows represent "include" relationships.

Output (Output KR is the face classification of the concept "Red-headed Woodpecker").

Annotation

"Red-headed Woodpecker" is {bird: Woodpecker, coloration: Red, eponymous anatomy: Head}

Note that in the above example, the atomic semantics in the AKRM representation can be used to explore the intersection of meanings across each KR (semantic interoperability). For example, the atomic concepts "crown" and "head" can provide connections of meaning across the formally disjoint concepts "sparrow" and "woodpecker".

It should be appreciated from the foregoing discussion and examples that aspects of the present invention can address some of the most pressing and challenging applications in knowledge representation, including tools for brainstorming and cognitive augmentation, supporting dynamic and emerging knowledge, and providing semantic interoperability by converting between various complex knowledge representations into a common semantic vocabulary.

The various inventive aspects described herein can be used with any of one or more computers and/or devices, each having one or more processors that can be programmed to perform any of the actions described above for using an atomic knowledge representation model when analyzing and synthesizing complex knowledge representations. For example, both the server and client computing systems can be implemented as one or more computers as described above. Figure 11 schematically illustrates an exemplary computer 1100 on which the various inventive aspects of the present disclosure can be implemented. The computer 1100 includes a processor or processing unit 1101 and a memory 1102, which can include volatile and/or non-volatile memory. The computer 1100 can also include storage 1105 (e.g., one or more disk drives) in addition to the system memory 1102.

Memory 1102 and/or storage 1105 may store one or more computer-executable instructions for programming processing unit 1101 to perform any of the functions described herein. Storage 1105 may also optionally store one or more datasets as needed. For example, a computer used to implement server system 100 may, in some embodiments, store AKRM dataset 110 in storage 1105. Alternatively, such a dataset may be implemented separately from the computer used to implement server system 100.

References here to a computer may include any device having a programmed processor, including a rack-mounted computer, a desktop computer, a laptop computer, a tablet computer, or any of the many devices that include a programmed processor that may not generally be considered a computer (e.g., a PDA, an MP3 player, a mobile phone, a wireless headset).

Exemplary computer 1100 can have one or more input devices and/or output devices, such as the equipment 1106 and 1107 shown in Figure 11.These devices can be used for presenting user interface and other functions.The example of the output device that can be used to provide user interface comprises the printer or the display screen that is used for visual presentation output and the loudspeaker or other sound generating device that is used for audible presentation output.The example of the input device that can be used for user interface comprises keyboard and pointing device, such as mouse, touch pad and digital panel.As another example, computer can receive input information by voice recognition or with other audible formats.

As shown in FIG11 , computer 1100 may also include one or more network interfaces (e.g., network interface 1110) for enabling communication via various networks (e.g., network 1120). Examples of networks include local area networks or wide area networks, such as an intranet or the Internet. Such networks may be based on any suitable technology and may operate according to any suitable protocol and may include wireless networks, wired networks, or fiber optic networks.

Having thus described several aspects of at least one embodiment of the present invention, it will be appreciated that various alterations, modifications, and improvements will readily occur to those skilled in the art.Accordingly, the foregoing description and drawings are by way of example only.

The embodiments described above of the present invention may be implemented in any of a number of ways. For example, the embodiments may be implemented using hardware, software, or a combination thereof. When implemented using software, the software code may be executed on any suitable processor or on a collection of processors, whether provided in a single computer or distributed among multiple processors. Such a processor may be implemented as an integrated circuit, wherein one or more processors are included in an integrated circuit component. However, the processor may be implemented using circuits of any suitable format.

Additionally, it will be appreciated that the computer may be implemented in any of a number of forms, such as a rack-mounted computer, a desktop computer, a laptop computer, or a tablet computer. Furthermore, the computer may be embedded in a device not generally regarded as a computer but with suitable processing capabilities, including a personal digital assistant (PDA), a smart phone, or any other suitable portable or fixed electronic device.

Computers may also have one or more input and output devices. These devices may be used to present user interfaces and other functions. Examples of output devices that may be used to provide user interfaces include printers or display screens for visual presentation output and loudspeakers or other sound generating devices for audible presentation output. Examples of input devices that may be used for user interfaces include keyboards and pointing devices, such as mice, touchpads, and digital panels. As another example, computers may receive input information via voice recognition or in other audible formats.

Such computers may be interconnected in any suitable manner by one or more networks (including as local area networks or wide area networks, such as an enterprise network or the Internet). Such networks may be based on any suitable technology and may operate according to any suitable protocol and may include wireless networks, wired networks, or fiber optic networks.

The various methods or processes summarized herein may also be encoded as software executable on one or more processors utilizing any of a variety of operating systems or platforms. Additionally, such software may be written using any of a plurality of appropriate programming languages and/or programming or scripting tools, and may also be compiled into executable machine language code or intermediate code that is executed on a framework or virtual machine.

In this regard, the present invention may be implemented as a tangible, non-transitory computer-readable storage medium (or multiple computer-readable media) (e.g., computer memory, one or more floppy disks, compact disks (CDs), optical disks, digital versatile disks (DVDs), magnetic tape, flash memory, a circuit configuration in a field programmable gate array or other semiconductor device, or other non-transitory, tangible computer-readable storage medium) encoded with one or more programs that, when executed on one or more computers or other processors, perform methods that implement the various embodiments of the present invention discussed above. The one or more computer-readable media may be portable, such that the one or more programs stored thereon may be loaded onto one or more different computers or other processors to implement various aspects of the present invention as discussed above. As used herein, the term "non-transitory computer-readable storage medium" encompasses only computer-readable media that may be considered an article of manufacture (i.e., an article of manufacture) or a machine.

The terms "program" or "software" are used herein in a generic sense to refer to any type of computer code or set of computer-executable instructions that can be used to program a computer or other processor to implement the various aspects of the present invention as discussed above. Furthermore, it should be appreciated that according to one aspect of the present invention, the one or more computer programs that, when executed, perform the methods of the present invention need not reside on a single computer or processor, but can be distributed in a modular manner among multiple different computers or processors to implement the various aspects of the present invention.

Computer-executable instructions can be in many forms, such as program modules, that are executed by one or more computers or other devices. Generally speaking, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The functionality of the program modules can often be combined or distributed as desired in various embodiments.

The data structure may also be stored in a computer-readable medium in any suitable form. For simplicity of description, the data structure may be illustrated as having fields that are related by their position in the data structure. Similar relationships may be achieved by assigning locations to the memory used for the fields in the computer-readable medium that transmits the relationships between the fields. However, any suitable mechanism may be used to establish relationships between the information in the fields of the data structure, including by using pointers, tags, or other mechanisms that establish relationships between data elements.

The various aspects of the present invention may be used alone, in combination, or in a variety of arrangements not specifically discussed in the embodiments described above, and thus the present invention is not limited in its application to the details and arrangements of components set forth in the foregoing description or illustrated in the accompanying drawings. For example, aspects described in one embodiment may be combined in any manner with aspects described in other embodiments.

The present invention may also be implemented as a method, for which examples have been provided. The actions performed as part of the method may be ordered in any suitable manner. Thus, embodiments may be constructed in which the actions are performed in an order different from that shown, which may include performing some actions simultaneously even though they are shown as sequential actions in the example embodiments.

The use of ordinal terms such as "first", "second", "third", etc. in the claims to modify claim elements does not in itself mean any priority, precedence or order of one claim element over another claim element or mean the temporal order of the actions of performing the method, but is merely used as a label to distinguish one claim element with a certain name from another claim element with the same name (but is used as ordinal terms) to distinguish claim elements.

All definitions, as defined and used herein, should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and/or ordinary meanings of the defined terms.

The indefinite articles "a" and "an" as used herein should be understood to mean "at least one" unless expressly indicated otherwise.

As used herein, the phrase "at least one" when referring to a list of one or more elements should be understood to mean at least one element selected from any one or more elements in the list of elements, but not necessarily including at least one element from each element specifically enumerated in the list of elements, and not excluding any combination of elements in the list of elements. This definition also allows for the optional presence of elements other than the elements specifically identified in the list of elements to which the phrase "at least one" refers, whether related or unrelated to those specifically identified elements. Thus, as a non-limiting example, "at least one of A and B" (or equivalently, "at least one of A or B" or equivalently, "at least one of A and/or B") can refer to at least one A, optionally including multiple A's without B's presence (and optionally including elements other than B) in one embodiment; at least one B, optionally including multiple B's without A's presence (and optionally including elements other than A) in another embodiment; at least one A, optionally including multiple A's and at least one B, optionally including multiple B's (and optionally including other elements) in yet another embodiment; and so on.

The phrase "and/or" as used herein should be understood to mean "either or both" of the elements so united, i.e., elements that exist in conjunction in some cases and in or in other cases. Multiple elements listed with "and/or," i.e., "one or more" of the elements so united, should be understood in the same manner. Other elements may optionally be present in addition to the elements specifically identified by the "and/or" clause, whether related or unrelated to those specifically identified elements. Thus, as a non-limiting example, a reference to "A and/or B," when used in conjunction with open language such as "comprising," may refer to only A (optionally including elements in addition to B) in one embodiment; to only B (optionally including elements in addition to A) in another embodiment; to both A and B (optionally including other elements) in yet another embodiment; and so on.

As used herein, "or" should be understood to have the same meaning as "and/or" as defined above. For example, when separating items in a list, "or" or "and/or" should be interpreted as inclusive, i.e., including a plurality of elements or at least one of a list of elements, also including a plurality and optionally including additional unlisted items.

The phrases and terms used herein are also for description and should not be considered as limiting. The use of "include" or "have", "comprises", "involves" and variations thereof herein means including the items listed thereafter and equivalents thereof as well as additional items.

Having specifically described several embodiments of the present invention, various modifications and improvements will readily occur to those skilled in the art. Therefore, the foregoing description is by way of example only and is not intended to be limiting.

Claims (20)

1. A computer-implemented method for synthesizing complex knowledge representations, the method comprising: Receive context from the data consumer; Based on the context identifier, the basic knowledge representation includes one or more basic components including a first concept; and Complex knowledge representations are generated by applying one or more rules to the one or more basic components through the execution of stored instructions via at least one processor. The generation of the complex knowledge representation includes synthesizing complex concepts that do not exist in the basic knowledge representation, and including the synthesized complex concepts in the complex knowledge representation. The synthesis of the complex concept includes combining the first concept and basic concepts that are not hierarchically related to the first concept to form the synthesized complex concept, and Generating the complex knowledge representation includes encoding the inherent relationship between the first concept and the complex concept in the complex knowledge representation. The generation of the complex knowledge representation includes: synthesizing the basic knowledge representation based on the distance metric between the basic knowledge representations. 2. The computer-implemented method of claim 1, wherein generating the complex knowledge representation includes including intrinsic relations in the complex knowledge representation, the intrinsic relations joining concepts in a set to create complex concepts. 3. The computer-implemented method of claim 2, wherein generating the complex knowledge representation includes encoding in the complex knowledge representation an inherent relationship between the first concept and the complex concept synthesized by combining the first concept with the basic concept. 4. The computer-implemented method of claim 3, wherein the first concept and the basic concept combined with the first concept to synthesize the complex concept form a concept definition for the complex concept. 5. The computer-implemented method of claim 3, wherein generating the complex knowledge representation includes including external relations in the complex knowledge representation, the external relations describing features between concept pairs. 6. The computer-implemented method of claim 5, wherein generating the complex knowledge representation includes encoding both the intrinsic relationship between the first concept and the complex concept and the extrinsic relationship between the first concept and the second concept in the complex knowledge representation. 7. The computer-implemented method of claim 6, wherein the external relationship between the first concept and the second concept encodes the hierarchical relationship between the first concept and the second concept. 8. At least one non-transient computer-readable storage medium storing computer-executable instructions, which, when executed, perform a method for synthesizing complex knowledge representations, the method comprising: Receive context from the data consumer; Based on the context identifier, the basic knowledge representation includes one or more basic components including a first concept; and Complex knowledge representations are generated by applying one or more rules to the one or more basic components through the execution of stored instructions via at least one processor. The generation of the complex knowledge representation includes synthesizing complex concepts that do not exist in the basic knowledge representation, and including the synthesized complex concepts in the complex knowledge representation. The synthesis of the complex concept includes combining the first concept and basic concepts that are not hierarchically related to the first concept to form the synthesized complex concept, and Generating the complex knowledge representation includes encoding the inherent relationship between the first concept and the complex concept in the complex knowledge representation. The generation of the complex knowledge representation includes: synthesizing the basic knowledge representation based on the distance metric between the basic knowledge representations. 9. The at least one non-transient computer-readable storage medium of claim 8, wherein generating the complex knowledge representation includes including intrinsic relations in the complex knowledge representation, the intrinsic relations joining concepts in a set to create complex concepts. 10. At least one non-transient computer-readable storage medium according to claim 9, wherein generating the complex knowledge representation includes encoding in the complex knowledge representation an inherent relationship between the first concept and the complex concept synthesized by combining the first concept with the basic concept. 11. The at least one non-transient computer-readable storage medium of claim 10, wherein the first concept and the basic concept combined with the first concept to synthesize the complex concept form a concept definition for the complex concept. 12. The at least one non-transient computer-readable storage medium of claim 10, wherein generating the complex knowledge representation includes including external relations in the complex knowledge representation, the external relations describing features between concept pairs. 13. The at least one non-transient computer-readable storage medium of claim 12, wherein generating the complex knowledge representation includes encoding both the intrinsic relationship between the first concept and the complex concept and the extrinsic relationship between the first concept and the second concept in the complex knowledge representation. 14. At least one non-transient computer-readable storage medium according to claim 13, wherein the external relationship between the first concept and the second concept encodes the hierarchical relationship between the first concept and the second concept. 15. An apparatus comprising: At least one processor; and At least one storage medium storing processor-executable instructions, which, when executed by the at least one processor, perform a method for synthesizing complex knowledge representations, the method comprising: Receive context from the data consumer; Based on the context identifier, the basic knowledge representation includes one or more basic components including a first concept; and Complex knowledge representations are generated by applying one or more rules to the one or more basic components. The generation of the complex knowledge representation includes synthesizing complex concepts that do not exist in the basic knowledge representation, and including the synthesized complex concepts in the complex knowledge representation. The synthesis of the complex concept includes combining the first concept and basic concepts that are not hierarchically related to the first concept to form the synthesized complex concept, and Generating the complex knowledge representation includes encoding the inherent relationship between the first concept and the complex concept in the complex knowledge representation. The generation of the complex knowledge representation includes: synthesizing the basic knowledge representation based on the distance metric between the basic knowledge representations. 16. The apparatus of claim 15, wherein generating the complex knowledge representation includes including intrinsic relations in the complex knowledge representation, the intrinsic relations joining concepts in a set to create complex concepts. 17. The apparatus of claim 16, wherein generating the complex knowledge representation includes encoding in the complex knowledge representation an inherent relationship between the first concept and the complex concept synthesized by combining the first concept with the basic concept. 18. The apparatus of claim 17, wherein the first concept and the basic concept combined with the first concept to synthesize the complex concept form a concept definition for the complex concept. 19. The apparatus of claim 17, wherein generating the complex knowledge representation includes including external relations in the complex knowledge representation, the external relations describing features between concept pairs. 20. The apparatus of claim 19, wherein generating the complex knowledge representation includes encoding both the intrinsic relationship between the first concept and the complex concept and the extrinsic relationship between the first concept and the second concept in the complex knowledge representation.