CN107122421A - Information retrieval method and device - Google Patents

Abstract

This application discloses a kind of information retrieval method and device, belong to internet arena, with the accuracy for the result for improving the user's problem to be solved retrieved.Methods described includes：Receive the problem to be solved of input；Determine the technical field belonging to the problem to be solved；According to the knowledge base in the technical field pre-established, determine the destination document matched in the technical field with the problem to be solved, wherein, the knowledge base includes the corresponding relation between corresponding relation and the Object of Knowledge and the document object between problem objects, Object of Knowledge, document object, described problem object and the Object of Knowledge, and the Object of Knowledge is selected from a part for described problem object；Return to the destination document.The application is used to answer problem to be solved.

Description

Information retrieval method and device

technical field

The present application relates to the field of the Internet, in particular to an information retrieval method and device.

Background technique

With the rapid development of the Internet, users are more and more inclined to obtain answers to questions by asking questions on the Internet. After obtaining the user's question, the search engine will search based on one or more keywords appearing in the question, and return results matching the one or more keywords.

However, it is very difficult for a machine to understand a human question, and the result obtained through the above method may not be the result that the user wants to obtain, resulting in low retrieval accuracy.

Contents of the invention

The embodiment of the present application provides an information retrieval method and device, so as to improve the accuracy of the retrieved results of the user's problem to be solved. Described technical scheme is as follows:

In one aspect, an information retrieval method is provided, the method comprising:

receive input open problems;

Determine the technical field to which the problem to be solved belongs;

According to the pre-established knowledge base in the technical field, determine the target document in the technical field that matches the problem to be solved, wherein the knowledge base includes question objects, knowledge objects, document objects, the a correspondence between a question object and said knowledge object and a correspondence between said knowledge object and said document object, said knowledge object being selected from a part of said question object;

Returns the target document.

In another aspect, an information retrieval device is provided, and the information retrieval device includes:

an interface module for receiving input pending problems;

A processing module, configured to determine the technical field to which the problem to be solved belongs;

The processing module is further configured to determine a target document matching the problem to be solved in the technical field according to a pre-established knowledge base in the technical field, wherein the knowledge base includes problem objects, a knowledge object, a document object, a correspondence between the question object and the knowledge object, and a correspondence between the knowledge object and the document object, the knowledge object being selected from a part of the question object;

The interface module is further configured to return the target document.

The beneficial effects brought by the technical solutions provided by the embodiments of the present application include:

When searching based on the user's problem to be solved (that is, the user's question), not only one or more keywords in the problem are considered, but also the technical field of the problem is considered. By considering the technical field of the problem to be solved and using the pre-built The specific knowledge base can greatly improve the accuracy of the retrieved results of the user's problem to be solved.

Description of drawings

FIG. 1 is a schematic diagram of a four-layer knowledge map in a specific technical field provided by an embodiment of the present application;

FIG. 2 is an exemplary relational diagram of a problem node, a knowledge node and a file node provided by an embodiment of the present application;

Fig. 3 is a flow chart of an example information retrieval method provided by an embodiment of the present application;

Fig. 4 is a schematic diagram of an example information retrieval method provided by an embodiment of the present application;

FIG. 5 is a diagram of the relationship between nodes showing the random walk probability between nodes provided by the embodiment of the present application;

Fig. 6 is a structural block diagram of an example information retrieval device provided by an embodiment of the present application.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the implementation manner of the present invention will be further described in detail below in conjunction with the accompanying drawings. The "electronic devices" mentioned in this article can include smart phones, tablet computers, smart TVs, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III, moving picture experts compression standard audio layer 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic image experts compress standard audio layer 4) Players, laptops and desktop computers, etc. The "information retrieval device" mentioned in this article may be one or more servers and the like.

Relevant information retrieval methods only consider the keywords appearing in a question, and it is often difficult to understand the user's intention. To understand a problem, humans usually use their basic knowledge in technical domains. For example, for the problem "When the user tries to send some special forms in the outbox, the program is stuck in the waiting state". First of all, we will pay attention to the "special form" and "outbox", these are some parts of the product outlook, we can infer that this is some problem caused by outlook.

From the above analysis, it can be concluded that background knowledge in the technical field plays an important role in problem understanding. In this application, a knowledge base of a specific technical field is constructed to facilitate the machine to understand user questions.

The information retrieval method in this application is based on a pre-built knowledge base. The knowledge base includes question objects, knowledge objects, document objects, correspondences between the question objects and the knowledge objects, and correspondences between the knowledge objects and the document objects. Wherein, the question object can be each unresolved problem input by the user, the knowledge object can be selected from a part of the unresolved problem, and the document object can be a document that solves each unresolved problem.

In order to facilitate the understanding of the knowledge base mentioned in this application, the following description presents various parts and their relationships in the knowledge base in the form of a knowledge graph.

A technical question usually consists of three parts: product, component and event words. Generally speaking, the knowledge graph in this application can include four parts: concept layer, product layer, component layer and event layer. in:

Concept layer: In the concept layer, a node represents a concept, a concept represents a group of products with similar functions, and a concept is usually also a sub-concept of another concept.

Product layer: The product layer contains all products and product attributes. The product layer is the core of the entire knowledge graph, and the nodes of the product layer represent a specific product or product attributes. Several attributes of a product, such as version, language, and runtime, can be predefined.

Component layer: Generally speaking, a technical problem is about a certain component of the product, and the component layer includes all product components.

Event layer: After the product or component has been defined, it is necessary to understand the specific phenomenon of a problem. The component layer contains some nouns, verbs, adjectives, etc. that describe the problem phenomenon.

An example of a knowledge graph can be shown in Figure 1. In Figure 1, four layers are divided by dotted lines from top to bottom: concept layer, product layer, component layer and event layer.

This paper uses technical corpus to construct a knowledge graph, and the specific construction method will be described below.

concept layer and product layer

This paper extracts concepts and products from product information. A total of eg 6052 products are obtained, belonging to eg 214 different categories. At the same time, this paper uses predefined rules to extract product attributes. For example, "Office ProWin32IT" means that the product name is Office, the version is Pro, the language is Italian, and it is installed on a 32-bit Windows operating system.

component layer

This paper utilizes technical corpus and user's question logs to extract components. First, identify the components mentioned in the corpus using some sequence annotation methods. These extracted phrases are represented as nodes of the component layer, which are measured by the PMI values of products and components in this paper. PMI is a common method used to measure the similarity between two phrases. If the PMI value of a component c and a product p exceeds a threshold, then we consider c to be a component of p. The definition of PMI is as follows:

in

#(c) indicates the number of occurrences of c, #(p) indicates the number of occurrences of p, and #(p, c) indicates the number of co-occurrences of p and c.

event layer

There are two different kinds of edges in the event layer, namely "event word (EventWordOf)" and "about (RelatedTo)", we discuss these two kinds of edges separately. First, the event word (EventWordOf) connects a product and an action word. We use a method similar to the component layer and use PMI to extract such a relationship. Generally speaking, users use verbs, adjectives, adverbs, nouns, etc. to describe the phenomenon of a problem. Given a large-scale technical corpus, first use some mature position tag (POS-TAG) methods to mark the part of speech of the technical corpus. At the same time, it is assumed that if two technical problems can be solved by the same problem, then they should be very similar in semantics. For example, document d can solve three technical problems, which are as follows:

q2: Outlook 2007 does not move (Outlook 2007gets frozen).

q9: The Outlook sending status remains for hours.

q15: Emails get stuck in outbox.

Therefore, we can conclude that the three words “frozen”, “remain” and “stuck” are semantically similar, so between the time nodes corresponding to these three words, there will be a connection “about ( RelatedTo)" relationship.

In order to return the target document associated with the user's problem to be solved, the knowledge object is associated with the document object in this application. Wherein, the document object can be obtained according to technical problem logs collected on the Internet.

An example connection relationship of question objects, knowledge objects and document objects can be shown in FIG. 2 . Each node in Figure 2 can represent an object, such as a question object, a knowledge object or a document object. The unresolved problem q in Figure 2 is: some specific Custom forms get stuck in Outbox when users send it. Document d1 is a description of Microsoft Office Suite SP2 (Description of 2007Microsoft Office Suite SP2).

In Figure 2, there are three types of connection edges: an edge connecting a question node to a knowledge node, an edge connecting two knowledge nodes, and an edge connecting a knowledge node to a document node. Wherein, for the same problem node, the edges connecting the problem node to each knowledge node have the same weight. The weight of the edge connecting two knowledge nodes can be expressed by conditional probability, that is, the weight from node x to node y is expressed as the probability of occurrence of y when x appears, as follows:

Among them, #(x,y) represents the number of co-occurrences of x and y.

The weight of the edge connecting the knowledge node to the document node can be expressed by the following formula:

Among them, the numerator represents the number of all problems that can be solved by d and include the number of problems belonging to x, the denominator is the number of all problems that can be solved by d, and QL(d) represents all the problems that can be solved by d.

After the knowledge base is pre-built, information retrieval can be performed according to the questions entered by the user.

Referring to Fig. 3, an embodiment of the present invention provides an information retrieval method, the method comprising:

Step 31, receiving an input problem to be solved.

Wherein, the input problem to be solved may be a problem to be solved input by the user through the electronic device.

Step 32, determining the technical field to which the problem to be solved belongs.

In the embodiment of the present application, the technical field to which the problem to be solved belongs can be determined through one or more keywords in the problem to be solved.

Step 33, according to the pre-established knowledge base in the technical field, determine the target document in the technical field that matches the problem to be solved, wherein the knowledge base includes question objects, knowledge objects, document objects , the corresponding relationship between the question object and the knowledge object and the corresponding relationship between the knowledge object and the document object, the knowledge object is selected from a part of the question object;

Step 34, returning the target document.

In this application, the target document matching the problem to be solved may be, for example, a target document that solves the problem to be solved, a target document that contains the problem to be solved, or contains one or more keys in the problem to be solved The target document for words.

In this application, returning the target document in step 34 may include: returning the name of the target document and/or returning the content in the target document.

In the embodiment of the present application, when searching based on the user's problem to be solved (that is, the user's question), not only one or more keywords in the problem are considered, but also the technical field of the problem is considered. By considering the technical field of the problem to be solved and Utilizing the pre-built specific knowledge base can greatly improve the accuracy of the retrieved results of the user's problem to be solved.

In the embodiment of the present application, the determination of the target document in the technical field that matches the problem to be solved in step 33 may include:

determining problems similar to the problem to be solved in the technical field according to the problem object, the knowledge object, and the correspondence between the problem object and the knowledge object in the knowledge base;

determining a similarity score between each of said similar problems and said problem to be solved;

Based on the similarity score and the target document corresponding to each of the similar questions, the target document matching the problem to be solved is determined.

What needs to be understood here is that in this embodiment of the present application, based on the similarity score and the target document corresponding to each of the similar problems, the target document corresponding to the similar problem with the highest similarity score can be directly selected as the target document that matches the problem to be solved. target document. This way, results are returned to the user as quickly as possible. This method can be applied to scenarios where users have extremely high speed requirements.

Of course, in this application, the target document corresponding to each of the similar questions can be used as a candidate document, and based on the similarity score and the target document corresponding to each of the similar questions, determine the Problem-solving matches to target documentation may include:

determining a similarity between the problem to be solved and each of the candidate documents based on the similarity scores;

Selecting one or more candidate documents as target documents matching the problem to be solved according to the descending order of similarity between the problem to be solved and the candidate document;

Wherein, the degree of similarity between the problem to be solved and each of the candidate documents is determined in the following manner:

q represents the problem to be solved, d represents a candidate document, score(q,d) represents the similarity between the problem to be solved q and the candidate document d, #(d, C) represents the total number of times d appears in C, # (d, C ₀ ) indicates the number of times d appears in C ₀ , (q' _i ,d)∈C ₀ indicates that d can solve the problem q' _i in C ₀ , score(q' _i ,q) indicates q ' The similarity score between _i and q; and C ₀ represents a subset of the problem log C, q' represents a problem similar to the problem q to be solved, and

C ₀ = {(q′ ₀ , d′ ₀ ), {(q′ ₁ , d′ ₁ ), ..., {q′ _m , d′ _m )}, q' _i means that the i-th one is similar to q , m represents the total number of questions similar to q, and d' represents the target document corresponding to q'.

Referring to FIG. 4 , FIG. 4 shows similar problems belonging to the same technical field as the problem to be solved, and a target document corresponding to each similar problem. In the process of determining the target document matching the problem to be solved in the technical field, the present application refers to FIG. 4 , if problem q000 is a similar problem with the highest similarity score between the problem to be solved, then the The document d1 corresponding to the question q000 is used as the target document of the question to be solved. Of course, both d5 and d1 (just an example) can be used as the target documents of the problem to be solved, and d1 is ranked before d5 when the result is returned.

As an optional manner, the ranking order of the returned results may also be determined based on the similarity between the problem to be solved and each of the candidate documents. Correspondingly, after the target document is determined in step 33, the information retrieval method provided by the embodiment of the present application may further include: based on a random walk (random walk) algorithm, calculating each of the problem to be solved and the knowledge base Similarity of document objects: reordering the plurality of target documents based on the similarity between the problem to be solved and each document object in the knowledge base.

After multiple target documents are reordered, the target documents can be returned according to the reordered results.

Wherein, the calculation of the similarity between the problem to be solved and each document object in the knowledge base based on the random walk algorithm described in the embodiment of the present application includes: selecting the relationship between the problem to be solved and the document object Indexes are set for one or more nodes of the node, wherein the index of the node represents the similarity between the node and each document object in the knowledge base; based on the index set for the one or more nodes, the calculated The similarity between the problem to be solved and each document object in the knowledge base.

One way of selecting nodes for setting indexes is: selecting frequent nodes on the path to set indexes, wherein a frequent node is a node whose in-degree and out-degree product is greater than a threshold.

The random walk algorithm is a method to measure the similarity of nodes. Generally speaking, if you start from a node and walk to another node randomly according to the probability of each edge, the probability of reaching another node is the initial The similarity of a node to another node. The similarity calculated by the random walk algorithm can be calculated as follows:

Among them, s(x, y) is the similarity between node x and node y based on random walk, N(x) represents all nodes connected to x, T(x, x') represents walking from node x Probability of going to node x'.

In this application, normalized weights are used as transition probabilities. When calculating the similarity based on the random walk algorithm, only the edge connecting the knowledge space node to the document node is kept, and the same method is used for the edge connecting the question node to the knowledge node.

The similarity between user question node q and document node d based on random walk can be calculated in different ways. One way is based on sampling method. We start from the user problem node q, and randomly move to an adjacent node with the edge weight as the transition probability. Assuming that the number of sampling is N, among which there are r times staying on the document node d, the similarity between q and d is r/N. Experiments show that it takes about 4 million samples before the node similarity tends to converge, which shows that Using sampling-based methods in online queries is time-consuming because during the query phase, real-time responses from the system are required. Another way is to create a system of linear equations based on the definition of random walk similarity, and solve the system of linear equations to get the answer. Referring to FIG. 5 , the connection line between every two nodes in FIG. 5 represents the probability of walking from one node to another adjacent node. The linear equations that can be listed based on the values shown in Figure 5 are as follows:

However, the complexity of solving a system of linear equations is very high, and the complexity of solving a system of linear equations by Gaussian elimination is O(n ³ ), where n is the number of unknowns in the system of equations. In the knowledge graph constructed in this paper, the number of nodes is very large, and the complexity of solving a linear equation system is very high. In order to improve the calculation speed, indexes can be built on some nodes in advance. For an indexed node, the form of the index is a series of floating-point numbers, indicating the similarity between the current node and all documents. For example, to index node x, the index form of x is:

Idex(x)={s(x, d ₀ ), s(x, d ₁ ), . . . , s(x, d _m )}

Among them, m is the number of documents, assuming that the index is built on the node, the similarity with each document can be obtained.

For example, if indexes are pre-established on nodes v ₅ , v ₈ , and v ₁₀ , s(v ₅ ,d ₁ )=0.701, s(v ₈ ,d ₁ )=0.668, s(v ₁₀ , d ₁ )=0.642, then the simplified result of the above linear equations can be as follows:

From the example above, it can be seen that the number of unknowns of the equation will be greatly reduced (from 11 to 3) if the indexes are built on some nodes in advance.

In the embodiment of this application, a greedy algorithm is proposed to select materialized (indexed) nodes. This greedy algorithm selects some nodes that frequently appear on many paths each time, because frequent nodes are easier to cover more paths, this paper uses in-degree × out-degree as the measurement index of frequent nodes, that is to say, in-degree The greater the degree of ×, the higher the frequency. In the greedy algorithm, the node with the highest frequency is selected each time, this node is added to the index node, and then the frequency of other nodes is recalculated, and finally all materialized nodes are obtained.

The index-based similarity calculation method provided by the embodiment of the present application can greatly reduce the calculation amount and improve the calculation efficiency. At the same time, by selecting the nodes for indexing based on frequency, some nodes can be selected for indexing instead of indexing for all nodes, which further reduces the amount of calculation.

FIG. 6 is a structural block diagram of an information retrieval device provided in the embodiment of the present application. Referring to FIG. 6 , the information retrieval device 600 provided in the embodiment of the present application includes: an interface module 601 and a processing module 602 . in:

An interface module 601, configured to receive an input problem to be solved;

A processing module 602, configured to determine the technical field to which the problem to be solved belongs;

The processing module 602 is further configured to determine a target document in the technical field that matches the problem to be solved according to a pre-established knowledge base in the technical field, wherein the knowledge base includes problem objects , a knowledge object, a document object, a correspondence between the question object and the knowledge object, and a correspondence between the knowledge object and the document object, the knowledge object being selected from a part of the question object;

The interface module 601 is further configured to return the target document.

The information retrieval device provided in the embodiment of the present application not only takes into account one or more keywords in the question, but also considers the technical field of the question when searching based on the user's problem to be solved (that is, the user's question). The technical field of problem solving and the use of pre-built specific knowledge base can greatly improve the accuracy of the results retrieved for the user's problem to be solved.

Optionally, the target document matching the problem to be solved is a target document for solving the problem to be solved.

The interface module is specifically configured to: return the name of the target document and/or return the content in the target document.

Optionally, the processing module 602 is specifically configured to:

determining problems similar to the problem to be solved in the technical field according to the problem object, the knowledge object, and the correspondence between the problem object and the knowledge object in the knowledge base;

determining a similarity score between each of said similar problems and said problem to be solved;

Based on the similarity score and the target document corresponding to each of the similar questions, the target document matching the problem to be solved is determined.

Optionally, the target document corresponding to each of the similar questions is used as a candidate document, and the processing module 602 is specifically configured to:

determining a similarity between the problem to be solved and each of the candidate documents based on the similarity scores;

Selecting one or more candidate documents as target documents matching the problem to be solved according to the descending order of similarity between the problem to be solved and the candidate document;

Wherein, the degree of similarity between the problem to be solved and each of the candidate documents is determined in the following manner:

q represents the problem to be solved, d represents a candidate document, score(q,d) represents the similarity between the problem to be solved q and the candidate document d, #(d, C) represents the total number of times d appears in C, # (d, C ₀ ) indicates the number of times d appears in C ₀ , (q' _i ,d)∈C ₀ indicates that d can solve the problem q' _i in C ₀ , score(q' _i ,q) indicates q ' The similarity score between _i and q; and C ₀ represents a subset of the problem log C, q' represents a problem similar to the problem q to be solved, and

C ₀ = {(q′ ₀ , d′ ₀ ), {(q′ ₁ , d′ ₁ ), ..., {q′ _m , d′ _m )}, q' _i means that the i-th one is similar to q , m represents the total number of questions similar to q, and d' represents the target document corresponding to q'.

Optionally, after the target document is determined, the processing module 602 is further configured to:

Based on a random walk algorithm, calculating the similarity between the problem to be solved and each document object in the knowledge base;

The plurality of target documents are reordered based on the similarity between the problem to be solved and each document object in the knowledge base.

Optionally, when calculating the similarity between the problem to be solved and each document object in the knowledge base based on the random walk algorithm, the processing module 602 is specifically configured to:

Selecting one or more nodes between the problem to be solved and the document object to set an index, wherein the index of the node indicates the similarity between the node and each document object in the knowledge base;

Based on the index set for the one or more nodes, the similarity between the problem to be solved and each document object in the knowledge base is calculated.

Optionally, when selecting a node for index setting, the processing module 602 is specifically configured to:

Select the frequent nodes on the path to set the index, where the frequent node is the node whose in-degree and out-degree product is greater than the threshold.

It should be noted that: the information retrieval device provided by the above-mentioned embodiment is only illustrated by the division of the above-mentioned functional modules. The internal structure is divided into different functional modules to complete all or part of the functions described above. In addition, the information retrieval device and the information retrieval method embodiments provided in the above embodiments belong to the same idea, and the specific implementation process thereof is detailed in the method embodiments, and will not be repeated here.

What needs to be understood here is that the interface module 601 and the processing module 602 can be different modules in the same physical device, and depending on the application, the interface module 601 can be one or more physical devices distributed in different locations, processing Module 602 may also be one or more physical devices distributed at different locations.

An embodiment of the present invention also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the memory in the above-mentioned embodiments; or exist independently and not be assembled into the terminal computer readable storage media. The computer-readable storage medium stores one or more programs, and the one or more programs are used by one or more processors to execute the above information retrieval method.

Unless otherwise defined, the technical terms or scientific terms used herein shall have the usual meanings understood by those skilled in the art to which this application belongs. "First", "second" and similar words used in the patent application specification and claims of this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Likewise, words like "a" or "one" do not denote a limitation in quantity, but indicate that there is at least one. Words such as "connected" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware, and can also be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. The above-mentioned The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, and the like.

The above are only exemplary embodiments of the application, and are not intended to limit the application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the application shall be included in the protection scope of the application. within.

Claims (14)

1. a kind of information retrieval method, it is characterised in that methods described includes：

Receive the problem to be solved of input；

Determine the technical field belonging to the problem to be solved；

According to the knowledge base in the technical field pre-established, determine in the technical field with the problem phase to be solved The destination document of matching, wherein, the knowledge base include problem objects, Object of Knowledge, document object, described problem object and The corresponding relation between corresponding relation and the Object of Knowledge and the document object between the Object of Knowledge, it is described to know Know the part that object is selected from described problem object；

Return to the destination document.

2. according to the method described in claim 1, it is characterised in that the destination document matched with the problem to be solved To solve the destination document of the problem to be solved；

The return destination document includes：Return to the title of the destination document and/or return in the destination document Content.

3. according to the method described in claim 1, it is characterised in that described to determine in the technical field to be solved to ask with described Inscribing the technical documentation matched includes：

The problem objects according to the knowledge base, the Object of Knowledge and described problem object and the Object of Knowledge it Between corresponding relation, the problem of determining similar with the problem to be solved in the technical field；

It is determined that it is each described similar the problem of and the problem to be solved between similarity score；

Based on the similarity score, and it is each described similar the problem of corresponding destination document, it is determined that with it is described to be solved The destination document that problem matches.

4. method according to claim 3, it is characterised in that corresponding destination document conduct the problem of each described similar Candidate documents, it is described based on the similarity score and it is each described similar the problem of corresponding destination document, it is determined that and institute Stating the destination document that problem to be solved matches includes：

Based on the similarity score, the problem to be solved and the similarity of each in the candidate documents are determined；

According to one or more candidates of the sequential selection of similarity from high to low between the problem to be solved and the candidate documents Document is used as the destination document matched with the problem to be solved；

Wherein, the problem to be solved and the similarity of each in the candidate documents are determined as follows：

<mrow> <mi>s</mi> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>e</mi> <mrow> <mo>(</mo> <mrow> <mi>q</mi> <mo>,</mo> <mi>d</mi> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <mi>log</mi> <mo>#</mo> <mrow> <mo>(</mo> <mrow> <mi>d</mi> <mo>,</mo> <mi>C</mi> </mrow> <mo>)</mo> </mrow> <mo>&amp;times;</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mrow> <mo>(</mo> <mrow> <msub> <msup> <mi>q</mi> <mo>&amp;prime;</mo> </msup> <mi>i</mi> </msub> <mo>,</mo> <mi>d</mi> </mrow> <mo>)</mo> </mrow> <mo>&amp;Element;</mo> <msub> <mi>C</mi> <mn>0</mn> </msub> </mrow> </munder> <mfrac> <mrow> <mo>#</mo> <mrow> <mo>(</mo> <mrow> <mi>d</mi> <mo>,</mo> <msub> <mi>C</mi> <mn>0</mn> </msub> </mrow> <mo>)</mo> </mrow> </mrow> <mrow> <mo>#</mo> <mrow> <mo>(</mo> <mrow> <mi>d</mi> <mo>,</mo> <mi>C</mi> </mrow> <mo>)</mo> </mrow> <mo>&amp;times;</mo> <mi>i</mi> </mrow> </mfrac> <mo>&amp;times;</mo> <mi>s</mi> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>e</mi> <mrow> <mo>(</mo> <mrow> <msub> <msup> <mi>q</mi> <mo>&amp;prime;</mo> </msup> <mi>i</mi> </msub> <mo>,</mo> <mi>q</mi> </mrow> <mo>)</mo> </mrow> <mo>;</mo> </mrow>

Q represents problem to be solved, and d represents a candidate documents, score (q, d) represent problem q and candidate documents d to be solved it Between similarity, # (d, C) represents the total degrees that occur in C of d, # (d, C₀) represent d in C₀The number of times of middle appearance, (q '_i,d)∈ C₀Represent that d can be solved in C₀Middle the problem of q '_i, score (q '_i, q) represent q '_iWith q similarity score；And C₀Problem of representation day Will C subset, the problem of q ' represents similar with problem q to be solved, and

C₀={ (q '₀, d '₀), { (q '₁, d '₁) ..., { q '_m, d '_m), q '_iRepresent i-th it is similar with q the problem of, m is represented and q The sum of similar the problem of, d ' represents destination document corresponding with q '.

5. according to any described methods of claim 1-4, it is characterised in that it is determined that after destination document, methods described is also Including：

Based on Random Walk Algorithm, the problem to be solved is calculated similar to each document object in the knowledge base Degree；

Based on the similarity of each document object in the problem to be solved and the knowledge base, to the multiple target text Shelves are reordered.

6. method according to claim 5, it is characterised in that described to be based on Random Walk Algorithm, is calculated described to be solved Problem and the similarity of each document object in the knowledge base include：

Select one or more nodes between the problem to be solved and the document object that index is set, wherein, the node Index represent the node to each document object in the knowledge base similarity；

Be based upon the index that one or more described nodes are set, calculate the problem to be solved with it is every in the knowledge base The similarity of one document object.

7. method according to claim 6, it is characterised in that selection sets the node of index to include：

Select the frequent node on path that index is set, wherein, frequent node is section of the product more than threshold value of in-degree and out-degree Point.

8. a kind of information indexing device, it is characterised in that described information retrieval device includes：

Interface module, the problem to be solved for receiving input；

Processing module, for determining the technical field belonging to the problem to be solved；

The processing module, is additionally operable to, according to the knowledge base in the technical field pre-established, determine the technical field In the destination document that matches with the problem to be solved, wherein, the knowledge base includes problem objects, Object of Knowledge, text Corresponding relation and the Object of Knowledge and the document object between shelves object, described problem object and the Object of Knowledge Between corresponding relation, the Object of Knowledge be selected from described problem object a part；

The interface module, is additionally operable to return to the destination document.

9. information indexing device according to claim 8, it is characterised in that what the described and problem to be solved matched Destination document is the destination document for solving the problem to be solved；

The interface module specifically for：Return to the content in the title and/or the return destination document of the destination document.

10. information indexing device according to claim 8, it is characterised in that the processing module specifically for：

The problem objects according to the knowledge base, the Object of Knowledge and described problem object and the Object of Knowledge it Between corresponding relation, the problem of determining similar with the problem to be solved in the technical field；

It is determined that it is each described similar the problem of and the problem to be solved between similarity score；

Based on the similarity score, and it is each described similar the problem of corresponding destination document, it is determined that with it is described to be solved The destination document that problem matches.

11. information indexing device according to claim 10, it is characterised in that corresponding mesh the problem of each described similar Mark document as candidate documents, the processing module specifically for：

Based on the similarity score, the problem to be solved and the similarity of each in the candidate documents are determined；

According to one or more candidates of the sequential selection of similarity from high to low between the problem to be solved and the candidate documents Document is used as the destination document matched with the problem to be solved；

Wherein, the problem to be solved and the similarity of each in the candidate documents are determined as follows：

<mrow> <mi>s</mi> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>e</mi> <mrow> <mo>(</mo> <mrow> <mi>q</mi> <mo>,</mo> <mi>d</mi> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <mi>log</mi> <mo>#</mo> <mrow> <mo>(</mo> <mrow> <mi>d</mi> <mo>,</mo> <mi>C</mi> </mrow> <mo>)</mo> </mrow> <mo>&amp;times;</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mrow> <mo>(</mo> <mrow> <msub> <msup> <mi>q</mi> <mo>&amp;prime;</mo> </msup> <mi>i</mi> </msub> <mo>,</mo> <mi>d</mi> </mrow> <mo>)</mo> </mrow> <mo>&amp;Element;</mo> <msub> <mi>C</mi> <mn>0</mn> </msub> </mrow> </munder> <mfrac> <mrow> <mo>#</mo> <mrow> <mo>(</mo> <mrow> <mi>d</mi> <mo>,</mo> <msub> <mi>C</mi> <mn>0</mn> </msub> </mrow> <mo>)</mo> </mrow> </mrow> <mrow> <mo>#</mo> <mrow> <mo>(</mo> <mrow> <mi>d</mi> <mo>,</mo> <mi>C</mi> </mrow> <mo>)</mo> </mrow> <mo>&amp;times;</mo> <mi>i</mi> </mrow> </mfrac> <mo>&amp;times;</mo> <mi>s</mi> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>e</mi> <mrow> <mo>(</mo> <mrow> <msub> <msup> <mi>q</mi> <mo>&amp;prime;</mo> </msup> <mi>i</mi> </msub> <mo>,</mo> <mi>q</mi> </mrow> <mo>)</mo> </mrow> <mo>;</mo> </mrow>

Q represents problem to be solved, and d represents a candidate documents, score (q, d) represent problem q and candidate documents d to be solved it Between similarity, # (d, C) represents the total degrees that occur in C of d, # (d, C₀) represent d in C₀The number of times of middle appearance, (q '_i,d)∈ C₀Represent that d can be solved in C₀Middle the problem of q '_i, score (q '_i, q) represent q '_iWith q similarity score；And C₀Problem of representation day Will C subset, the problem of q ' represents similar with problem q to be solved, and

C₀={ (q '₀, d '₀), { (q '₁, d '₁) ..., { q '_m, d '_m), q '_iRepresent i-th it is similar with q the problem of, m is represented and q The sum of similar the problem of, d ' represents destination document corresponding with q '.

12. according to any described information indexing devices of claim 8-11, it is characterised in that it is determined that after destination document, The processing module is additionally operable to：

Based on Random Walk Algorithm, the problem to be solved is calculated similar to each document object in the knowledge base Degree；

Based on the similarity of each document object in the problem to be solved and the knowledge base, to the multiple target text Shelves are reordered.

13. information indexing device according to claim 12, it is characterised in that based on Random Walk Algorithm, calculate institute When stating the similarity of each document object in problem to be solved and the knowledge base, the processing module specifically for：

Select one or more nodes between the problem to be solved and the document object that index is set, wherein, the node Index represent the node to each document object in the knowledge base similarity；

Be based upon the index that one or more described nodes are set, calculate the problem to be solved with it is every in the knowledge base The similarity of one document object.

14. information indexing device according to claim 13, it is characterised in that when selection sets the node of index, institute State processing module specifically for：

Select the frequent node on path that index is set, wherein, frequent node is section of the product more than threshold value of in-degree and out-degree Point.