JP2015062117A - Entity linking method and entity linking apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an entity linkage method and an entity linkage device.SOLUTION: An entity linkage method includes: a step of acquiring a potential link entity of a mention; a step of acquiring a first text obtained by extending an existing original text of the mention and corresponding to the mention, and calculating a text similarity between the first text and a second text corresponding to the potential link entity; a step of calculating a coincidence between the potential link entity and the mention on the basis of the text similarity; and a step of linking the potential link entity having the coincidence satisfying a predetermined condition to the mention. Accuracy of the link entity can be improved by extending the existing original text of the mention and selecting the potential link entity by the extended text.

Description

  The present invention relates to the field of databases, and in particular to a method and apparatus for linking entities.

  With the rapid development of the Internet vocabulary and knowledge library, and the upper platform of shot texts such as miniblogs, shot message service (SMS), etc., how “mention” in shot messages and Internet vocabulary and knowledge It is a problem in the field of linguistic information processing whether the content of shot text is lexicalized by associating with entities in the library.

  An embodiment of the present invention aims to provide an entity linking method and apparatus capable of attaching a link to an entity matching the entity reference in the content of the shot text.

  In one aspect of the invention, obtaining a potential link entity of an entity reference, extending the original text where the entity reference is located to obtain a first text corresponding to the entity reference; A step of calculating a text similarity between the first text and a second text corresponding to the potential link entity; and a degree of coincidence between the potential link entity and the entity reference based on the text similarity And linking the entity reference to a potential link entity having a degree of coincidence that satisfies a predetermined condition.

  In another aspect of the present invention, acquisition means for acquiring a potential link entity of an entity reference (mention), and an original text in which the entity reference is located are expanded to acquire a first text corresponding to the entity reference. Expansion processing means for calculating a text similarity between the first text and a second text corresponding to the potential link entity, and the potential link entity and the entity reference based on the text similarity. An entity linking apparatus comprising: matching processing means for calculating a degree of coincidence; and linking means for attaching a link to a potential link entity having a degree of coincidence satisfying a predetermined condition to the entity reference. provide.

  According to the embodiment of the present invention, the accuracy of a link entity can be improved by expanding the original text where the entity reference is located and selecting the potential link entity based on the expanded text.

Through the following detailed description of the drawings, the above objects, other objects, features and advantages of the embodiments of the present invention will become clearer. The units in the drawings are merely illustrative of the principles of the present invention. In the drawings, identical or similar technical features or units are indicated by identical or similar symbols.
3 is a flowchart of an entity linking method according to the present invention. It is a figure which shows the acquisition method of the potential link entity in the entity linking method according to the present invention. It is a figure which shows the other acquisition method of the potential link entity in the linking method of the entity based on this invention. It is a figure which shows the structure of the linking apparatus of the substance based on this invention. It is a figure which shows the structure of the computer apparatus for implementing the apparatus and method which concern on this invention.

  Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the drawings. For convenience of explanation, the specification does not show all the features of the actual embodiment. In actual implementation, a specific embodiment may be changed in order to realize a specific goal of a developer. For example, the embodiment may be changed according to a restriction condition related to a system and business. Good. Also, the development work is very complex and time consuming, but for those skilled in the art, this development work is just an example work. For the sake of clarity, the drawings show only the components or processing steps of the apparatus closely related to the present invention, and omit details not related to the present invention. Further, elements and features shown in the drawings or embodiments of the present invention may be combined with elements and features shown in other drawings or embodiments.

  Embodiments of the present invention provide an entity linking method and apparatus. Hereinafter, an entity linking method according to an embodiment of the present invention will be described based on a specific embodiment with reference to the drawings.

  FIG. 1 is a flowchart of an entity linking method according to the present invention.

  As shown in FIG. 1, in step S101, a potential link entity of entity mention is acquired.

  Here, there are a plurality of methods for acquiring the entity-referenced potential link entity, which will be described in detail below.

  After the entity reference in the shot text is determined, an entity corresponding to the entity reference may be extracted as a potential link entity by directly searching the knowledge library based on the entity reference. This knowledge library may include specific knowledge libraries corresponding to the scene applied to the entity linking method, for example, Internet vocabulary and knowledge libraries Wikipedia, Wikipedia, BaiduBaike, etc. Not limited to knowledge libraries. For example, if the entity reference is “apple”, a plurality of potential link entities such as “Apple”, “Apple Inc. of the United States”, etc. can be found from the Internet wording and knowledge library.

  Here, retrieving a potential link entity from an existing knowledge library includes (1) making the transfer link in the knowledge library a potential link entity, and (2) the first paragraph in the content found from the knowledge library. Bold text is used as a potential link entity, (3) Obtaining a potential link entity based on an ambiguity avoidance page found from a knowledge library, and (4) Linking an anchor file with an entity Obtaining a potential link entity of an entity reference based on information such as a relationship may include, but is not limited to. For example, as shown in FIG. 2, for the entity reference “IBM”, the content found from the knowledge library is the transfer link “International Business Machines Corporation”, as well as the bold “International Business Machines Corporation” in the first paragraph. Including “Corporation”, “International Business Machines Corporation”, and “Universal Commercial Equipment Company”, any of which may be potential link entities of the entity reference “IBM”. As shown in FIG. 3, an ambiguity avoidance page is found from the knowledge library for the entity reference “Apple”, and “Apple Incorporated”, “Apple Daily Newspaper”, “Apple (Movie)”, etc. are all entities. It may be a potential link entity of the mention “Apple”.

  After the entity reference in the shot text is determined, a potential link entity of the entity reference may be obtained based on the shot text. For example, an entity that includes the entity reference in the shot text and exists in the knowledge library may be a potential link entity. For example, in the shot text “Beijing time March 12, 2013 Asian Champions League group league second round, Guangzhou Hengda football club team away from Jeonbuk Hyundai, Guangzhou Hengda ’s first announcement” If the shot text includes “Guangzhou Hengda Soccer Club Team” and “Guangzhou Hengda Soccer Club Team” is an entity in the knowledge library, ”Is a potential link entity of“ Guangzhou Hengdai ”, and“ Guangzhou Hengda Soccer Club Team ”is not an entity in the knowledge library, no candidate is provided by this method.

  After the entity reference in the shot text is determined, before performing the search for the potential link entity described above, the entity reference is expanded to obtain a temporary entity reference with a longer length, based on the temporary entity reference. The potential link entity may be searched by any one of the above methods. Because temporary entity references contain more information, it is possible to more accurately link to entities that should be linked to entity references.

  Before describing the method for acquiring a temporary entity reference, first, the determination of the type of entity reference used for acquiring the temporary entity reference will be described.

  The entity type body is used to determine the type of entity reference. The entity type body defines the type of entity in the real world and the relationship between the types. For example, the highest type of entity is an object (OBJECT), OBJECT is a person (PERSON) entity, a position (LOCATION) entity, Including an organization (ORGANIZATION) entity, a government entity (GEO-POLITICAL ENTITY), etc., human entities are classified into different types according to job types. Thus, the entity type body may be understood as a classification system from general to detail. There are two methods for obtaining the entity type body. One is using an existing entity type body, for example, an entity type published on the website http://www.dmoz.org/, and the other is The user sets his / her own entity type body according to his / her demand, and the text is not limited to these.

  After the entity type main body is acquired, a machine recognition model that can automatically determine the type of a predetermined entity may be trained by a machine learning technique. Machine learning techniques include two methods: supervised learning and unsupervised learning. Supervised learning requires the training corpus to be marked, for example to recognize a PERSON entity, mark the training corpus that contains the PERSON entity, and machine recognition by a supervised model such as CRF, SVM, and training corpus You may train the model. With this machine recognition model, the PERSON entity in the text can be recognized. For other entity types, a machine recognition model may be trained by a similar method to obtain a machine recognition model that can automatically determine a predetermined entity type. In unsupervised learning, an entity type may be recognized by a specialized dictionary, and for example, a place name may be recognized by character string matching using a place name dictionary, or an entity type may be recognized based on a model. For example, in the case of “<xxx president>”, the type of xxx is recognized as PERSON, and a machine recognition model that can automatically determine the type of a predetermined entity is acquired by a series of unsupervised learning.

  Except for automatically recognizing the type of the entity reference by the device, the user may artificially specify the type of the entity reference. Here, the type of the entity reference mention corresponds to the type in the entity type main body. That is, the user may directly determine the type of entity reference without using mechanical determination. The advantage of this is that the user may link to the type of entity of interest, eg only the PERSON or ORGANIZATION type of entity in the shot text.

  The determination of the type of entity reference does not have to depend on the entity type main body. That is, it is not necessary to predetermine the correspondence between the entity and the type. For example, when the user determines the type of entity reference, the type of entity reference may be specified at random.

After the type of entity reference is acquired, the provisional entity reference may be determined by the following method. When the type of the entity reference is human, another expression obtained by extending the entity reference may be extracted as the temporary entity reference from the original text by coreference resolution. For example, if the text is “LeBron Raymone James, nicknamed“ King James ”, is an American
professional basketball player for the Miami Heat of the National Basketball
Association (NBA). Lebron has played the small forward and power forward
If the position is "Lebron", and the "LeBron Raymone James" co-reference to "Lebron" is found in the original text, then "LeBron Raymone James" instead of the original mention "Lebron" That is, “LeBron Raymone James” is a temporary entity reference for “Lebron”.

  If the type of entity reference is an administrative agency, a place name is searched from the original text, and the entity reference and the searched place name are combined to make a provisional entity reference. For example, using a naming entity recognition tool, an entity to which the mention belongs is found from the context of the mention, and a combination of these entities and the mention is used as a temporary entity reference. For example, the text is "The Supreme Court was established on October 22, 1949 and is the highest refereeing agency in the People's Republic of China and oversees the refereeing work of each level court and professional court" Is "Supreme Court", find the place name "People's Republic of China" to which the mention belongs from the original text, and combine the "People's Republic of China" and "Supreme Court" to refer to the provisional entity "People's Republic of China" Acquired the Supreme Court.

If the type of entity reference is an institution, text corresponding to the entity reference is extracted from the original text as a temporary entity reference by pattern matching. For example, if the text is “Agricultural Bank of China (ABC), also known as AgBank or agriculture, is one of the“ Big Four ”banks in the People's Republic of China” and the mention is “ABC”, the pattern By matching, from the original text "Agricultural
"Bank of China" is extracted and "Agricultural" is substituted for "ABC"
“Bank of China” may be used directly, that is, “Agricultural Bank of China” may be used as a temporary entity reference of “ABC”.

The provisional entity reference may be determined by the following method. If the entity reference is an abbreviation, the text corresponding to the entity reference may be extracted from the original text as a temporary entity reference by pattern matching. For example, if the text is `` Michael Jordan (MJ) has been the best basketball player, but MJ is
If it is "now too old" and the entity mention is "MJ", "Michael Jordan" is extracted from the original text by pattern matching, and "Michael Jordan" is used as a temporary entity reference for "MJ" Also good.

  When the entity reference is not an abbreviation, the temporary entity reference may be determined based on the type of the entity reference in the same manner as the method for determining the temporary entity reference. For example, if the type of entity reference is recognized based on the entity type body, and the type of entity reference is human, other expressions obtained by extending the entity reference from the original text by coreference analysis are extracted as temporary entity references, If the type of entity reference is an administrative agency, a place name is searched from the original text, and the entity reference and the searched place name are combined to make a provisional entity reference. The example may refer to the method for determining the provisional entity reference, and the description thereof is omitted here.

  After acquiring the potential link entity in step S101, step S102 is executed.

  In step S102, a potential link entity is selected based on the type of entity, and a potential link entity whose entity type does not satisfy a predetermined condition is removed. The predetermined condition may be the same as the entity reference type, or may be the same as the entity type preset by the user. For example, if the user presets that the type of the link entity of the entity reference “MJ” is a company, for example, the found potential link entity “Michael Jordan” is excluded. Here, the entity type determination method will be described later in the detailed description of step S101.

  In step S103, the original text where the entity reference is located is expanded to obtain the first text corresponding to the entity reference, and the text similarity between the first text and the second text corresponding to the potential link entity Is calculated. For example, the original text “Hongdae won” of the entity reference “Hengdai” will be expanded to the first text “Guangzhou Hengdao football club team has won the second place in the China Football Premium League” Can do.

  Here, the original text expansion method may include the following methods, but is not limited thereto. Select text within a predetermined length L around the entity reference, divide the text into words, calculate the degree of association between each word and the entity reference, and n words with the highest degree of association And search the word library (corpus) by n words and entity references to obtain a text segment containing the entity references and n keywords, and combine the text segment with the original text where the entity reference is located. To obtain the first text.

In addition, although calculation of a relevance may be calculated | required with the following formula | equation, it is not limited to this.

  Here, x represents an entity reference, y represents a divided word, NGD (x, y) represents the degree of association between x and y, and M represents the total number of potential link entities obtained by searching. F (x) represents the number of times x appears in the potential link entity, f (y) represents the number of times y appears in the potential link entity, and f (x, y) represents the potential of x and y Represents the number of times a link entity appears simultaneously

  The calculation of the degree of association may be realized in other ways, for example, by determining the degree of association between each divided word and the entity reference based on the frequency with which each divided word and entity reference appears simultaneously in the potential link entity. It may be determined directly.

  Note that the text segment including the entity reference and the n keywords obtained by searching the phrase library by n keywords and the entity reference may not be unique. In this case, a plurality of texts are generated based on the generation time. Segments may be sorted out. For example, when there are m obtained text segments, k search results whose difference from the original text time is t or less are selected. Assuming that the generation time of the original text is T and setting the time window to t, the search result with the generation time [T−t, T + t] is selected. For example, if the entity reference is “James”, the obtained keywords are “NBA” and “Final”, and the original text generation time T = “June 18, 2013”, t = 10 days Assuming that, a searched text segment having a generation time of “June 8, 2013” to “June 28, 2013” may be selected.

If there are still a plurality of text segments obtained by sorting based on time, further sorting may be performed. For example, construct a vector for each text segment and each original text, calculate the cosine distance between each text segment vector and the original text vector, and obtain the text segment whose cosine distance satisfies the threshold condition Segments may be sorted out. Assuming that the number of text segments selected by time is k, for each of the k search results, construct a vector v _i (i = 1,2,... K) and a vector for the original text. v _o is constructed, the cosine distance between v _i and v _o is calculated, and the top w search results with the smallest cosine distance are selected. Preferably, w may be set to 1.

  In addition, as an extension method of the original text, a text within a certain length around the entity reference may be the first text, and the certain length may be, for example, an experience value.

  The text similarity between the first text and the second text may be calculated by various similarity calculation methods in the prior art, but is not limited to the text.

  After calculating the text similarity between the first text and the second text in step S103, step S104 is executed based on the similarity.

  In step S104, the degree of coincidence of potential link entities is determined based on the text similarity between the first text and the second text.

  Here, the step of determining the matching degree of the potential link entity based on the text similarity between the first text and the second text is a step of directly setting the text similarity as the matching degree, or is set in advance A step of determining the degree of coincidence based on a relationship between the text similarity and the degree of coincidence, for example, a proportional coefficient, may be included.

  The degree of coincidence may be determined based on a combination of the text similarity and other factors. Here, other factors include the similarity between the character string of the entity reference and the character string of the entity name of the potential link entity, the popularity of the link between the entity reference and the potential link entity, and the potential link entity It may include, but is not limited to, the probability of cross-referencing with entity references. Other factors may be used alone or in combination with each other.

  Here, the similarity between the character string of the entity reference and the character string of the entity name of the potential link entity may use various similarity calculation methods in the prior art, but the text is not limited.

The popularity of linking entity references to potential link entities is determined by calculating the frequency at which each potential link entity appears in all potential link entities retrieved based on the entity reference. For example, for the entity reference “OA”, 100 pages are found from the net knowledge library and the potential link entity “Office” is found.
If "office automation" appears on 90 pages and "Office Action" appears on 10 pages, including "Action" and "office automation", the potential link entities "Office Action" and "office automation" The popularity of linking “automation” is 0.1 and 0.9, for example. This example is for explanation, and does not limit the specific method for determining the popularity of linking.

  The probability of cross-referencing between a potential link entity and an entity reference may be determined based on the source of the potential link entity. For example, if the source is a forward link, the cross reference probability corresponding to the potential link entity is determined to be a first predetermined value, for example 1, and if the source is bold, it corresponds to the potential link entity. If the cross reference probability is determined to be a second predetermined value, for example 0.8, and the source is an ambiguity avoidance page, the cross reference probability corresponding to the potential link entity is determined to be 1 / k, and k Is the total number of ambiguous entities.

  Preferably, the cross-reference probability between the potential link entity and the entity reference is set directly in the process of obtaining the potential link entity, and the preset cross-reference probability is directly used in the subsequent process. Also good.

As described above, the degree of coincidence may be determined based on a plurality of factors. For example, the text similarity is expressed as ContextSimilarity (mention, entity), the similarity between the entity reference character string and the entity name character string of the potential link entity is expressed as LexicalSimilarity (mention, entity), and the entity reference and potential The popularity of linking with link entities is expressed as PopularProbability (mention, entity), the probability of cross-reference between potential link entities and entity references is expressed as p (m, e), and entity references and potential link entities FinalScore (m,
When expressed as e), FinalScore (m, e) is calculated by the following equation. Here, m represents an entity mention, and e represents an entity.

  Here, α, β, and γ are weighting factors, and their values may be estimated by training samples. Preferably, because the characteristics of different types of entity references are different, α, β and γ may be estimated independently for different types of entity references, ie different types of entity references (eg, the above mentioned human, institution and government offices). ) Different training samples may be constructed and trained to obtain different weights α, β and γ.

  In step S105, it is determined whether or not there is a potential link entity having a matching degree satisfying a predetermined condition. If it is determined that there exists, step S106 is executed, and if not, step S107 is executed. Run. The predetermined condition is, for example, a threshold, that is, a potential link entity having a matching degree equal to or higher than the threshold is a potential link entity that satisfies the predetermined condition. Preferably, different types of entity references may have different predetermined conditions, such as different thresholds of coincidence.

  In step S106, a link to a potential link entity having a matching degree that satisfies a predetermined condition is attached to the entity reference.

  In step S107, all potential link entities corresponding to entity references are stored in the knowledge library. If there is no potential link entity with the matching degree that satisfies the predetermined condition, it is considered that there is no entity corresponding to the entity reference in the current knowledge library, and all the potential links corresponding to the entity reference The entity may be stored in a database. The knowledge library may be various knowledge libraries described in step S101. If there is no potential link entity having a matching degree that satisfies a predetermined condition, another process may be executed. For example, a presentation message indicating that there is no potential link entity may be fed back, and the presentation message may be in the form of text or speech.

  According to the embodiment of the present invention, the accuracy of a link entity can be improved by expanding the original text where the entity reference is located and selecting the potential link entity based on the expanded text.

  Based on the same technical idea as the above technical solution, the present invention further provides an entity linking device. As shown in FIG. 4, the entity linking device includes an acquisition unit 41 that acquires a potential link entity of an entity mention, and a first corresponding to the entity reference by extending the original text where the entity reference is located. An extension processing unit 42 that obtains the text of the first text and calculates the text similarity between the first text and the second text corresponding to the potential link entity, and the potential link entity and entity reference based on the text similarity A matching processing unit 43 that calculates the degree of coincidence with the data, and a linking unit 44 that attaches a link to a potential link entity having a degree of coincidence satisfying a predetermined condition in the entity reference.

  The entity linking device sorts out potential link entities based on the entity types before the extension processing unit 42 executes processing, and removes the potential link entities whose entity types do not satisfy the predetermined condition. The sorting unit 45 may be further included.

  The entity linking apparatus includes a determination unit 46 that determines whether or not there is a potential link entity having a matching degree satisfying a predetermined condition, and a potential link entity having a matching degree that satisfies the predetermined condition. May further include a storage unit 47 for storing the entity reference and all its potential link entities in a knowledge library.

  In order to correspond to the technical proposal of the above method, the acquisition unit 41 can execute the process of the above step S101 and refer to the above step S101 for the specific process, and the description thereof is omitted here. . Similarly, the selection unit 45 executes the process of step S102, the expansion processing unit 42 executes the process of step S103, the matching processing unit 43 executes the process of step S104, and the determination unit 46 performs the above step. The process of S105 is executed, the linking unit 44 executes the process of step S106, and the storage unit 47 executes the process of step S107.

  The following describes an exemplary configuration of a computer of a data processing device for carrying out the present invention with reference to FIG. FIG. 5 is a block diagram showing an exemplary configuration of a computer apparatus for carrying out the present invention.

  Each component and unit in the computer may be realized by software, firmware, hardware, or a combination thereof. When implemented by software or firmware, a program constituting the software for performing the above method is installed from a recording medium or a network to a computer having a dedicated hardware configuration (for example, the general-purpose computer 1100 shown in FIG. 5). The computer may execute various functions when various programs are installed.

  In FIG. 5, a central processing unit (ie, CPU) 1101 performs various processes by a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage unit 1108 to a random access memory (RAM) 1103. Run. The RAM 1103 stores data necessary for the CPU 1101 to execute various processes as necessary. The CPU 1101, ROM 1102, and RAM 1103 are connected to each other via a bus 1104. An input / output interface 1105 is also connected to the bus 1104.

  Input unit 1106 (including a keyboard, mouse, etc.), output unit 1107 (display, including a cathode ray tube (CRT), liquid crystal display (LCD), etc., speaker, etc.), storage unit 1108 (including a hard disk, etc.), communication The unit 1109 (for example, including a network interface card such as a LAN card or a modem) is connected to the input / output interface 1105. The communication unit 1109 executes communication processing via a network, for example, the Internet. The drive unit 1110 may be connected to the input / output interface 1105 as needed. The removable medium 1111 is, for example, a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like, and is set up in the drive unit 1110 as necessary, and the computer program read from the memory is stored as necessary. Part 1108 is installed.

  When the above processing is performed by software, a program constituting the software is installed via a network, for example, the Internet, or a storage medium, for example, a removable medium 1111.

  Further, these storage media are not limited to the removable media 1111 shown in FIG. 5 that stores the program and provides the program to the user separately from the device. The removable medium 1111 includes, for example, a magnetic disk (including a floppy (registered trademark) disk), an optical disk (including an optical disk-read only memory (CDROM), and a digital multipurpose disk (DVD)), and a magneto-optical disk (mini disk ( MD) (registered trademark)) and semiconductor memory. Alternatively, the storage medium may be a ROM 1102, a hard disk included in the storage unit 1108, etc., and stores a program and is provided to the user together with a device including them.

  The present invention further provides a program product for storing a command code readable by a device. The command code is read by the device, and the method according to the embodiment of the present invention described above can be executed.

  Accordingly, the present invention further includes a storage medium in which a product of a program that stores a device-readable command code is recorded. The storage medium includes, but is not limited to, a floppy disk, an optical disk, a magneto-optical disk, a memory card, and a memory stick.

  The terms used in the text are for describing specific embodiments and do not limit the present invention. As used herein, the singular forms “a” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. As used herein, the term “inclusive / include” means the presence of a feature, element, step or component, and the presence or absence of one or more other features, elements, steps or components. It does not exclude the addition.

  All equivalents of the elements defined by the corresponding structure, material, operation and “apparatus or step plus function” in the claims are intended to be any structure, material or function that performs the function in combination with other protected elements. It means to include an action. Certain embodiments of the present invention are disclosed in detail and illustrate the manner in which the principles of the present invention can be employed, as illustrated in the following description and drawings. The embodiment of the present invention is not limited in scope. The embodiments of the present invention include various alterations, modifications, and equivalents within the scope and spirit of the appended claims.

  Moreover, the following additional remarks are disclosed regarding the embodiment including each of the above-described examples.

(Appendix 1)
Obtaining a potential link entity for an entity mention;
The original text where the entity reference is located is expanded to obtain a first text corresponding to the entity reference, and the text similarity between the first text and the second text corresponding to the potential link entity is obtained. A calculating step;
Calculating a degree of coincidence between the potential link entity and the entity reference based on the text similarity;
Attaching the entity reference to a link to a potential link entity having a matching degree that satisfies a predetermined condition.

(Appendix 2)
Before calculating the degree of agreement between the potential link entity and the entity reference,
2. The entity linking method according to appendix 1, wherein a potential link entity that does not match the entity reference type or a preset type is removed.

(Appendix 3)
The entity linking method according to claim 2, further comprising the step of recognizing a type of the body reference based on the entity type body.

(Appendix 4)
Obtaining a potential link entity for the entity reference;
Searching for an entity corresponding to the entity reference from a knowledge library to be a latent link entity, and / or, from a text where the entity reference is located, identifying an entity that includes the entity reference and exists in the knowledge library The entity linking method according to appendix 1, comprising the step of acquiring as a static link entity.

(Appendix 5)
Calculating the degree of coincidence between the potential link entity and the entity reference based on the text similarity;
Similarity between the character string of the entity reference and the character string of the entity name of the potential link entity, the popularity of the link between the entity reference and the potential link entity, and the potential link entity and the entity reference The entity link according to claim 1, further comprising: calculating a degree of coincidence between the potential link entity and the entity reference based on at least one of cross-probability probabilities with the text similarity. Attaching method.

(Appendix 6)
Extending the original text in which the entity reference is located to obtain a first text corresponding to the entity reference;
Selecting text within a predetermined length around the entity reference and dividing the text into words;
Calculating the relevance between each word and the entity reference, and using the n words with the highest relevance as keywords,
Searching a phrase library by the n words and the entity reference to obtain a text segment including the entity reference and the n keywords;
The entity linking method according to claim 5, further comprising: obtaining the first text by combining the text segment and the original text where the entity reference is located.

(Appendix 7)
The entity linking method according to appendix 6, wherein when a plurality of the text segments are acquired, the text segments are selected so as to acquire a text segment whose generation time satisfies a predetermined condition.

(Appendix 8)
If there are two or more text segments obtained by sorting, a vector is constructed for each text segment and the original text, and a cosine distance between each text segment vector and the original text vector is calculated. The entity linking method according to appendix 7, wherein the text segment is selected so that the text segment whose distance satisfies the threshold condition is acquired.

(Appendix 9)
In calculating the cross-reference probability between the potential link entity and the entity reference,
6. The entity linking method according to appendix 5, wherein the source of the potential link entity in a knowledge library is acquired, and the probability of cross-reference corresponding to the potential link entity is determined based on the source.

(Appendix 10)
Determining the probability of a cross-reference corresponding to the potential link entity based on the source;
If the source is a forwarding link, determining a cross-reference probability corresponding to the potential link entity as a first predetermined value;
If the source is bold, determining a cross-reference probability corresponding to the potential link entity as a second predetermined value;
If the source is an ambiguity avoidance page, determining the probability of a cross reference corresponding to the potential link entity as 1 / k, where k is the total number of ambiguous entities The entity linking method according to appendix 9, including:

(Appendix 11)
Obtaining a potential link entity for the entity reference;
The entity linking method according to claim 1, further comprising the step of expanding the entity reference to obtain a temporary entity reference and obtaining the potential link entity based on the temporary entity reference.

(Appendix 12)
Extending the entity reference to obtain a provisional entity reference comprises:
12. The entity linking method according to appendix 11, including a step of extracting text corresponding to the entity reference from the original text as the temporary entity reference by pattern matching when the entity reference is an abbreviation.

(Appendix 13)
Extending the entity reference to obtain a provisional entity reference comprises:
The entity linking method according to claim 11, further comprising: expanding the entity reference to obtain the temporary entity reference based on the type of the entity reference.

(Appendix 14)
Extending the entity reference based on the entity reference type to obtain the temporary entity reference comprises:
If the type of the entity reference is human, extracting from the original text another representation as an extension of the entity reference as the provisional entity reference by coreference resolution;
If the type of the entity reference is an administrative agency, a step of searching for a place name from the original text and combining the entity reference with the searched place name to form the temporary entity reference;
If the type of the entity reference is an institution, the step of extracting the text corresponding to the entity reference as the temporary entity reference from the original text by pattern matching is included. Method.

(Appendix 15)
15. The entity linking method according to appendix 14, wherein the entity reference type is recognized based on the entity type main body.

(Appendix 16)
The entity linking method according to appendix 1, wherein when there is no potential link entity having a matching degree that satisfies a predetermined condition, all potential link entities of the entity reference are stored in a corresponding database.

(Appendix 17)
An acquisition means for acquiring a potential link entity of an entity mention;
The original text where the entity reference is located is expanded to obtain a first text corresponding to the entity reference, and the text similarity between the first text and the second text corresponding to the potential link entity is obtained. Extended processing means to calculate,
Matching processing means for calculating a degree of coincidence between the potential link entity and the entity reference based on the text similarity;
An entity linking apparatus comprising: linking means for linking the entity reference with a link to a potential link entity having a matching degree satisfying a predetermined condition.

(Appendix 18)
The expansion processing means includes
Selecting text within a predetermined length around the entity reference and dividing the text into words;
The degree of association between each word and the entity reference is calculated, and the n words having the highest degree of association are used as keywords,
Searching a phrase library by the n words and the entity reference to obtain a text segment including the entity reference and the n keywords;
18. The entity linking device according to appendix 17, wherein the first text is obtained by combining the text segment and the original text where the entity reference is located.

(Appendix 19)
18. The entity linking device according to appendix 17, wherein the acquisition unit expands the entity reference to acquire a temporary entity reference, and acquires the potential link entity based on the temporary entity reference.

(Appendix 20)
20. The entity linking apparatus according to appendix 19, wherein the acquisition unit acquires the temporary entity reference by expanding the entity reference based on the type of the entity reference.

Claims (10)

Obtaining a potential link entity for an entity mention;
The original text where the entity reference is located is expanded to obtain a first text corresponding to the entity reference, and the text similarity between the first text and the second text corresponding to the potential link entity is obtained. A calculating step;
Calculating a degree of coincidence between the potential link entity and the entity reference based on the text similarity;
Attaching the entity reference to a link to a potential link entity having a matching degree that satisfies a predetermined condition. Calculating the degree of coincidence between the potential link entity and the entity reference based on the text similarity;
Similarity between the character string of the entity reference and the character string of the entity name of the potential link entity, the popularity of the link between the entity reference and the potential link entity, and the potential link entity and the entity reference Calculating the degree of coincidence between the potential link entity and the entity reference based on at least one of the cross-reference probabilities and the text similarity. Linking method. Extending the original text in which the entity reference is located to obtain a first text corresponding to the entity reference;
Selecting text within a predetermined length around the entity reference and dividing the text into words;
Calculating the relevance between each word and the entity reference, and using the n words with the highest relevance as keywords,
Searching a phrase library by the n keywords and the entity reference to obtain a text segment including the entity reference and the n keywords;
The entity linking method according to claim 1, further comprising: obtaining the first text by combining the text segment and the original text where the entity reference is located.   The entity linking method according to claim 3, wherein when a plurality of the text segments are acquired, the text segments are selected so as to acquire the text segments whose generation time satisfies a predetermined condition.   If there are two or more text segments obtained by sorting, a vector is constructed for each text segment and the original text, and a cosine distance between each text segment vector and the original text vector is calculated. 5. The entity linking method according to claim 4, wherein the text segment is selected so as to obtain the text segment whose distance satisfies the threshold condition. Obtaining a potential link entity for the entity reference;
The entity linking method according to claim 1, further comprising: expanding the entity reference to obtain a temporary entity reference, and obtaining the potential link entity based on the temporary entity reference. Extending the entity reference to obtain a provisional entity reference comprises:
The entity linking method according to claim 6, further comprising: extracting a text corresponding to the entity reference from the original text as the temporary entity reference by pattern matching when the entity reference is an abbreviation. Extending the entity reference to obtain a provisional entity reference comprises:
The entity linking method according to claim 6, further comprising the step of expanding the entity reference to obtain the temporary entity reference based on the type of the entity reference. Extending the entity reference based on the entity reference type to obtain the temporary entity reference comprises:
If the type of the entity reference is human, extracting from the original text another representation as an extension of the entity reference as the provisional entity reference by coreference resolution;
If the type of the entity reference is an administrative agency, a step of searching for a place name from the original text and combining the entity reference with the searched place name to form the temporary entity reference;
The entity link according to claim 8, further comprising: extracting a text corresponding to the entity reference from the original text as the temporary entity reference by pattern matching when the type of the entity reference is an institution. Attaching method. An acquisition means for acquiring a potential link entity of an entity mention;
The original text where the entity reference is located is expanded to obtain a first text corresponding to the entity reference, and the text similarity between the first text and the second text corresponding to the potential link entity is obtained. Extended processing means to calculate,
Matching processing means for calculating a degree of coincidence between the potential link entity and the entity reference based on the text similarity;
An entity linking apparatus comprising: linking means for linking the entity reference with a link to a potential link entity having a matching degree satisfying a predetermined condition.

Images