CN113392182A - Knowledge matching method, device, equipment and medium fusing context semantic constraints - Google Patents

Abstract

The invention discloses a knowledge matching method, device, equipment and medium integrating context semantic constraints. The method includes: performing knowledge extraction on an acquired input sentence to obtain an entity vector and a relation vector in the input sentence; converting the input sentence into is a text vector; splicing the entity vector, the relationship vector and the text vector to obtain a target vector; performing similarity matching between the target vector and the vector records in the knowledge base to determine a set of similar vectors; The vector with the highest similarity in the vector set is used as the matching result. The invention improves the efficiency, reduces the time cost and labor cost for producing the labeling data, and can be widely used in the technical field of natural language processing.

Description

Knowledge matching method, device, device and medium incorporating context semantic constraints

technical field

The present invention relates to the technical field of natural language processing, in particular to a knowledge matching method, apparatus, device and medium integrating context semantic constraints.

Background technique

With the development of artificial intelligence, it has gone through the stage from computational intelligence to perceptual intelligence, and is moving towards the stage of cognitive intelligence. To achieve cognitive intelligence, machines must learn to understand, reason, and interpret knowledge like humans, which is a major problem that artificial intelligence urgently needs to solve. At present, by combining deep learning and knowledge graph reasoning, the semantic gap of human natural language can be well resolved. Knowledge graph is a structured way to describe the relationship between concepts, entities and their attributes in the objective world, usually expressed as triples, namely [entity 1, relationship, entity 2] or [entity, attribute, attribute value]. The knowledge graph can better organize, manage and understand the massive information on the Internet, and can transform the information on the Internet into a form that is closer to the human cognitive world. Therefore, it is widely used in intelligent retrieval, intelligent question answering, personalized recommendation and other fields.

In the intelligent retrieval and intelligent question answering system based on knowledge graph, for unstructured data, the basic tasks are: knowledge graph construction, information extraction (entity extraction, relation extraction, attribute extraction), entity linking, query construction Wait. The largest Chinese knowledge graph that has been open sourced has a scale of 140 million, and the knowledge graph construction technology is very mature, and the required knowledge graph can be constructed soon. Information Extraction (IE) is a text processing technology that extracts factual information such as entities, attributes, relationships, and events from natural language texts. With the needs of industrial applications and the promotion of academic competitions, Chinese information extraction has also achieved certain results. 's results. However, due to the complexity of text semantics, entity linking and query construction are more difficult. The existence of polysemous words, homographs, demonstrative pronouns, etc., makes the name of the same entity ambiguous and corresponds to different specific entities. For example, the entity "apple" can refer to "apple (fruit)" and "apple (tech company)" without context. How to distinguish and correctly connect to entities is the main research problem of entity linking and querying.

Existing entity linking methods mainly include two steps, namely entity referential recognition and entity disambiguation. Referential recognition first needs to build a referent-entity dictionary, such as extracting the title of the entity page, disambiguation page and redirecting page of the encyclopedia as entity referents, so as to establish a referent-entity dictionary. It takes a certain amount of time and material cost to construct this dictionary, and the construction cost increases sharply with the size of the data. Entity disambiguation mainly adopts methods such as sorting learning and graph models, which rely on a large amount of labeled data. In addition, using query languages such as SPARQL and Cypher to query the extracted entity-relationship and entity-relationship similarity matching is also a relatively common method for linking knowledge bases, but this will face the query results. There are many possibilities question. For example, entering "Who starred in "The King of Comedy"? Is her debut work in the question-and-answer system? Using entity relationship extraction, you can get [The King of Comedy, starring,? ],in'? ' means a placeholder. Using the above query language to query in the knowledge base, you may get: [King of Comedy, starring, Stephen Chow], [King of Comedy, starring, Karen Mok], [King of Comedy, starring, Cecilia Cheung] and other results, the similarity matching is the same. Since only triple information is usually stored in the knowledge base, the machine cannot use the following information of "debut work" to be constrained, so it is difficult to obtain the correct result of [The King of Comedy, starring, Cecilia Cheung].

SUMMARY OF THE INVENTION

In view of this, embodiments of the present invention provide a low-cost and high-efficiency knowledge matching method, apparatus, device, and medium that integrate context semantic constraints.

One aspect of the present invention provides a knowledge matching method incorporating contextual semantic constraints, including:

Performing knowledge extraction on the obtained input sentence to obtain the entity vector and the relation vector in the input sentence;

converting the input sentence into a text vector;

Splicing the entity vector, the relation vector and the text vector to obtain a target vector;

Carry out similarity matching between the target vector and the vector record in the knowledge base, and determine a set of similar vectors;

The vector with the highest similarity in the similarity vector set is used as the matching result.

Preferably, the method further comprises the step of building a knowledge base, the step comprising:

Train word vectors according to the original training text;

Learning the vector representation of entities and relationships of each training text in the knowledge base, and obtaining entity training vectors and relationship training vectors;

Perform word segmentation processing on the original training text, and perform vector representation on the result of the word segmentation processing according to the trained word vector to obtain a first vector;

According to the splicing process of the entity training vector, the relation training vector and the first vector, the knowledge base construction is completed.

Preferably, before the step of training the word vector according to the original training text, the method further includes:

Preprocessing the original training text, wherein the preprocessing includes removing stop words and word segmentation;

The process of removing stop words includes: deleting useless words in the original training text, where the useless words include at least one of the following: modal particles, adverbs, prepositions, and conjunctions;

The word segmentation processing includes: converting the text into a word sequence separated by spaces by using the jieba word segmentation tool.

Preferably, the result of the word segmentation processing is represented by a vector according to the trained word vector to obtain a first vector, including:

After the trained word vectors are added one by one and averaged, the first vector is obtained.

Preferably, according to the splicing process of the entity training vector, the relationship training vector and the first vector, the knowledge base construction is completed, including:

A first triplet is constructed according to the entity and the relationship, and a second triplet is constructed according to the attributes of the entity, and the first triplet is [first entity, relationship, second entity], so The second triple is [entity, attribute, attribute value];

The first triplet and the second triplet are collectively represented as the first vector.

Preferably, the expression form of the target vector is: p=(e _p,1 ,r _p ,0,t _p );

Among them, _p is the target vector; e _{p ,1} is the entity vector; r _p is the relation vector; t _p is the text vector; The placeholder vector for two entities.

Preferably, the similarity matching between the target vector and the vector records in the knowledge base is performed to determine a set of similar vectors, including:

Calculate the similarity between the target vector and the vector records in the knowledge base by using the improved cosine similarity calculation formula, and determine the similarity vector set according to the calculated similarity;

Wherein, the cosine similarity calculation formula is:

e_s,1代表知识库的第一实体向量；r_s代表知识库的关系向量；e_s,2代表知识库的第二实体向量；t_s代表知识库的文本语义向量；β代表权重；

e_p,1代表待匹配的第一实体向量；r_p代表待匹配的关系向量；t_p代表待匹配的文本语义向量。in,

es _,1 represents the first entity vector of the knowledge base; _rs represents the relation vector of the knowledge base; es _,2 represents the second entity vector of the knowledge base; _ts represents the text semantic vector of the knowledge base; β represents the weight;

ep _,1 represents the first entity vector to be matched; r _p represents the relation vector to be matched; t _p represents the text semantic vector to be matched.

Another aspect of the embodiments of the present invention also provides a knowledge matching apparatus integrating context semantic constraints, including:

a knowledge extraction module, used for performing knowledge extraction on the acquired input sentence to obtain the entity vector and the relation vector in the input sentence;

a conversion module for converting the input sentence into a text vector;

a splicing module for splicing the entity vector, the relation vector and the text vector to obtain a target vector;

a similarity matching module, used for similarity matching between the target vector and the vector record in the knowledge base, to determine a set of similar vectors;

The determining module is configured to use the vector with the highest similarity in the similarity vector set as the matching result.

Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;

the memory is used to store programs;

The processor executes the program to implement the method as described above.

Another aspect of the embodiments of the present invention further provides a computer-readable storage medium, where the storage medium stores a program, and the program is executed by a processor to implement the aforementioned method.

The embodiment of the present invention also discloses a computer program product or computer program, where the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The computer instructions can be read from the computer-readable storage medium by a processor of the computer device, and the processor executes the computer instructions to cause the computer device to perform the foregoing method.

The embodiment of the present invention performs knowledge extraction on the acquired input sentence to obtain entity vector and relation vector in the input sentence; converts the input sentence into a text vector; performs knowledge extraction on the entity vector, relation vector and the text vector splicing to obtain a target vector; similarity matching is performed between the target vector and the vector records in the knowledge base to determine a similar vector set; the vector with the highest similarity in the similar vector set is used as a matching result; the present invention utilizes the similarity The matching method does not require additional data as support in entity links, eliminates the steps of processing and constructing additional data, improves efficiency, and reduces the time cost and labor cost of producing labeled data.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present application more clearly, the following briefly introduces the drawings that are used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

Fig. 1 is the overall step flow chart of the embodiment of the present invention;

FIG. 2 is an example diagram of a word vector provided by an embodiment of the present invention;

FIG. 3 is an example diagram of knowledge matching provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application.

Aiming at the problems existing in the prior art, an embodiment of the present invention provides a knowledge matching method integrating context semantic constraints, including:

Performing knowledge extraction on the obtained input sentence to obtain the entity vector and the relation vector in the input sentence;

converting the input sentence into a text vector;

Splicing the entity vector, the relation vector and the text vector to obtain a target vector;

Carry out similarity matching between the target vector and the vector record in the knowledge base, and determine a set of similar vectors;

The vector with the highest similarity in the similarity vector set is used as the matching result.

Preferably, the method further comprises the step of building a knowledge base, the step comprising:

Train word vectors according to the original training text;

Learning the vector representation of entities and relationships of each training text in the knowledge base, and obtaining entity training vectors and relationship training vectors;

Perform word segmentation processing on the original training text, and perform vector representation on the result of the word segmentation processing according to the trained word vector to obtain a first vector;

According to the splicing process of the entity training vector, the relation training vector and the first vector, the knowledge base construction is completed.

Preferably, before the step of training the word vector according to the original training text, the method further includes:

Preprocessing the original training text, wherein the preprocessing includes removing stop words and word segmentation;

The process of removing stop words includes: deleting useless words in the original training text, where the useless words include at least one of the following: modal particles, adverbs, prepositions, and conjunctions;

The word segmentation processing includes: converting the text into a word sequence separated by spaces by using the jieba word segmentation tool.

Preferably, the result of the word segmentation processing is represented by a vector according to the trained word vector to obtain a first vector, including:

After the trained word vectors are added one by one and averaged, the first vector is obtained.

Preferably, according to the splicing process of the entity training vector, the relationship training vector and the first vector, the knowledge base construction is completed, including:

A first triplet is constructed according to the entity and the relationship, and a second triplet is constructed according to the attributes of the entity, and the first triplet is [first entity, relationship, second entity], so The second triple is [entity, attribute, attribute value];

The first triplet and the second triplet are collectively represented as the first vector.

Preferably, the expression form of the target vector is: p=(e _p,1 ,r _p ,0,t _p );

Among them, _p is the target vector; e _{p ,1} is the entity vector; r _p is the relation vector; t _p is the text vector; The placeholder vector for two entities.

Preferably, the similarity matching between the target vector and the vector records in the knowledge base is performed to determine a set of similar vectors, including:

Calculate the similarity between the target vector and the vector records in the knowledge base by using the improved cosine similarity calculation formula, and determine the similarity vector set according to the calculated similarity;

Wherein, the cosine similarity calculation formula is:

e_s,1代表知识库的第一实体向量；r_s代表知识库的关系向量；e_s,2代表知识库的第二实体向量；t_s代表知识库的文本语义向量；β代表权重；

e_p,1代表待匹配的第一实体向量；r_p代表待匹配的关系向量；t_p代表待匹配的文本语义向量。in,

es _,1 represents the first entity vector of the knowledge base; _rs represents the relation vector of the knowledge base; es _,2 represents the second entity vector of the knowledge base; _ts represents the text semantic vector of the knowledge base; β represents the weight;

ep _,1 represents the first entity vector to be matched; r _p represents the relation vector to be matched; t _p represents the text semantic vector to be matched.

The specific implementations of the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings:

As shown in FIG. 1 , the present invention includes two stages of building a knowledge base and knowledge matching. It should be noted that the present invention is developed on the premise of an existing knowledge graph and text data for constructing the knowledge graph.

(1) Build a knowledge base:

Step 1), training word vectors: using the text data (hereinafter referred to as the source text) used to construct the knowledge graph, use the text representation method to train the word vectors. Further, the source text needs to be preprocessed such as removing stop words and word segmentation. The purpose of removing stop words is to save space and improve processing efficiency. According to the open source stop word table, non-meaning words or words without actual semantic information in the text are deleted. These words are usually modal particles, adverbs, prepositions, conjunctions, etc. , and only the keywords are retained in the end. The word segmentation is to use a word segmentation tool such as jieba to convert the text into a sequence of words separated by spaces.

Specifically, the embodiment of the present invention performs word segmentation and cleaning operations on the text corpus used to build the knowledge base, and the text corpus comes from the "2019 Language and Intelligence Technology Competition" jointly organized by the Chinese Computer Society, the Chinese Information Society of China and Baidu. "Datasets for Information Extraction Tasks. It is preferred to perform special processing on the corpus, which is divided into two parts, one is the triplet dataset, and the other is the source text dataset corresponding to the triplet. After removing stop words from the source text data set, use jieba and other word segmentation tools for word segmentation, and then input the open source Glove model to train a 300-dimensional word vector. The representation of the resulting word vector is shown in Figure 2.

Step 2), training knowledge graph embedding: similar to word vector, use knowledge graph embedding technology to learn the vector representation of entities and relationships in the knowledge base, and finally obtain entity vectors and relationship vectors. Further, this knowledge graph embedding does not distinguish between relations and attributes, as well as entity 2 and attribute value, that is, [entity 1, relation, entity 2] and [entity, attribute, attribute value] are uniformly expressed as [E1, R, E2].

Specifically, the embodiment of the present invention uses an open-source knowledge graph to embed the Complex model to convert the triple data set obtained in step 1) into a vector representation. Suppose a triple is [The King of Comedy, starring, Cecilia Cheung], vectorized The latter is es _,1 , r,es _,2 where es _,1 , r,es _,2 are all 100-dimensional, and the triplet is similar to the word vector.

Step 3), find the source text vector: after the word segmentation of the source text and the removal of stop words, add the trained word vectors one by one to obtain an average vector t to represent the entire source text.

Specifically, the embodiment of the present invention assumes that there is a sentence in the source text data set as "Cecilia Cheung's debut work is "The King of Comedy" directed by Stephen Chow", and the word sequence "Cecilia Cheung / debut / work / Stephen Chow / director / is obtained from step 1). "King of Comedy", the word vector of each word in the word sequence is added and averaged to obtain a 300-dimensional vector t _s to represent the source text. So far, all triples and corresponding source texts have been vectorized.

Step 4) Knowledge splicing: splicing the triples in the knowledge base, namely the three vectors es _,1 ,r,es _,2 of [E1, R, E2], and splicing the representation of the source text of the triples The vectors _ts are spliced together to obtain the vector s=(es _,1 , _rs ,es _,2 , _ts ), that is, the vector _s is composed of four vectors es _,1 ,r,es _,2 ,ts along the line The vector direction splicing finally contains the triple information and the semantic information of the source text. The representation vector _ts of the source text is obtained in step 3). So far, the entire knowledge base has been constructed.

Specifically, in the embodiment of the present invention, the triplet vector obtained in step 2) and the source text vector obtained in step 3) are spliced along the row vector direction to form a record in the knowledge base. The dimension of each record is 600 dimensions. In the present invention, the entire knowledge base has a total of 340,000 records, expressed as s=(es _,1 , _rs ,es _,2 , _ts ).

(2) Knowledge matching stage:

Step 1), perform knowledge extraction on the input sentence, that is, extract the entity and relationship in the sentence, and convert the entity into a vector ep _{, 1} according to the knowledge map embedding obtained in step 2) of the above-mentioned construction knowledge base stage, and the relationship is transformed. is represented by a vector _rp . It should be noted that the focus of the present invention is not on the knowledge extraction step, so the knowledge extraction related technology will not be described in detail.

Specifically, the embodiments of the present invention perform knowledge extraction on the questions input to the question answering system, and the knowledge extraction model may refer to the entry model published in the 2019 Language and Intelligence Technology Competition. Suppose the input question is "Who starred in "The King of Comedy", and the King of Comedy is her debut work", then through the knowledge extraction model, we can get [The King of Comedy, starring,? ],in? is a placeholder, then according to the knowledge graph embedding trained in step 2) of the knowledge base construction phase, the entity "King of Comedy" and the relationship "starring" are vectorized as ep _,1 , _rp , ? It is represented by a zero vector of the same dimension.

In step 2), the input sentence is converted into a text vector t _p , and the adopted method is step 3) in the above-mentioned stage of building a knowledge base.

Specifically, in the embodiment of the present invention, the input sentence is converted into a text vector _tp : the same as the knowledge base construction step 3). starring/king of comedy/king of comedy/she/debut/work", the sentence is vectorized as _tp .

Step 3), splicing the entity vector ep _,1 obtained in step 1), the relation vector r _p and the text vector t _p obtained in step 2) into p=(e _p,1 ,r _p ,0,t _p ), , where 0 is a zero vector of the same dimension as e _p,1 , r _p , which is used as the placeholder vector of entity 2.

Specifically, the splicing vector _p =(ep _,1 , rp , 0, t _p ) in the embodiment of the present invention is spliced along the row vector direction by four vectors _{ep , 1} , rp , 0, t _p .

Step 4), using the improved cosine similarity to perform similarity matching between the vector obtained in step 3) and the record in the knowledge base. The improved cosine similarity is expressed as:

是由向量s在源文本向量t_s乘以一个权重系数β所得。in,

is obtained by multiplying the vector s by a weight coefficient β in the source text vector t _s .

是由向量p在源文本向量t_p乘以一个权重系数β所得。

is obtained by multiplying the vector p by a weight coefficient β in the source text vector t _p .

Specifically, the embodiment of the present invention performs similarity matching between p and each record in the knowledge base s, and the matching result is shown in FIG. 3 . The similarity algorithm is an improved cosine similarity, and the calculation formula is:

是由向量s在源文本向量t_s乘以一个权重系数β所得。in,

is obtained by multiplying the vector s by a weight coefficient β in the source text vector t _s .

是由向量p在源文本向量t_p乘以一个权重系数β所得。

is obtained by multiplying the vector p by a weight coefficient β in the source text vector t _p .

The weight coefficient β takes 1.8 as the optimum in the present invention.

Step 5), sort the results obtained in step 4), and take the one with the highest similarity as the final knowledge matching result.

Compared with the prior art, the beneficial effects of the present invention are:

(1) Using the similarity matching method, there is no need for additional data as support in the entity link, which eliminates the steps of processing and constructing additional data, and reduces the time cost and labor cost of producing labeled data. Further, compared with the method using query language, the knowledge matching result obtained by the present invention has no ambiguity and multiple selection.

(2) Different from the existing invention, the constraints of context information are introduced, so that the output of knowledge matching tends to be unique, and the weights of triple similarity and text similarity can be dynamically adjusted, and the matching results perform better.

(3) The present invention can be applied to different fields, the difference lies in the specific steps of research planning, and it has great versatility.

The embodiment of the present invention also provides a knowledge matching device integrating context semantic constraints, including:

a knowledge extraction module, used for performing knowledge extraction on the acquired input sentence to obtain the entity vector and the relation vector in the input sentence;

a conversion module for converting the input sentence into a text vector;

a splicing module for splicing the entity vector, the relation vector and the text vector to obtain a target vector;

a similarity matching module, used for similarity matching between the target vector and the vector record in the knowledge base, to determine a set of similar vectors;

The determining module is configured to use the vector with the highest similarity in the similarity vector set as the matching result.

Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;

the memory is used to store programs;

The processor executes the program to implement the method as described above.

Another aspect of the embodiments of the present invention further provides a computer-readable storage medium, where the storage medium stores a program, and the program is executed by a processor to implement the aforementioned method.

The embodiment of the present invention also discloses a computer program product or computer program, where the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The computer instructions can be read from the computer-readable storage medium by a processor of the computer device, and the processor executes the computer instructions to cause the computer device to perform the foregoing method.

The embodiment of the present invention also discloses a computer program product or computer program, where the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The computer instructions can be read from the computer-readable storage medium by a processor of the computer device, and the processor executes the computer instructions to cause the computer device to perform the foregoing method.

In some alternative implementations, the functions/operations noted in the block diagrams may occur out of the order noted in the operational diagrams. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality/operations involved. Furthermore, the embodiments presented and described in the flowcharts of the present invention are provided by way of example in order to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of the various operations are altered and in which sub-operations described as part of larger operations are performed independently.

Furthermore, while the invention is described in the context of functional modules, it is to be understood that, unless stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or or software modules, or one or more functions and/or features may be implemented in separate physical devices or software modules. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary to understand the present invention. Rather, given the attributes, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the modules will be within the routine skill of the engineer. Accordingly, those skilled in the art, using ordinary skill, can implement the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are illustrative only and are not intended to limit the scope of the invention, which is to be determined by the appended claims along with their full scope of equivalents.

The functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

The logic and/or steps represented in flowcharts or otherwise described herein, for example, may be considered an ordered listing of executable instructions for implementing the logical functions, may be embodied in any computer-readable medium, For use with, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from and execute instructions from an instruction execution system, apparatus, or apparatus) or equipment. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with an instruction execution system, apparatus, or apparatus.

More specific examples (non-exhaustive list) of computer readable media include the following: electrical connections with one or more wiring (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), Read Only Memory (ROM), Erasable Editable Read Only Memory (EPROM or Flash Memory), Fiber Optic Devices, and Portable Compact Disc Read Only Memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program may be printed, as the paper or other medium may be optically scanned, for example, followed by editing, interpretation, or other suitable medium as necessary process to obtain the program electronically and then store it in computer memory.

It should be understood that various parts of the present invention may be implemented in hardware, software, firmware or a combination thereof. In the above-described embodiments, various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or a combination of the following techniques known in the art: Discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, Programmable Gate Arrays (PGA), Field Programmable Gate Arrays (FPGA), etc.

In the description of this specification, description with reference to the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples", etc., mean specific features described in connection with the embodiment or example , structure, material or feature is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, The scope of the invention is defined by the claims and their equivalents.

The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements on the premise of not violating the spirit of the present invention, These equivalent modifications or substitutions are all included within the scope defined by the claims of the present application.

Claims (10)

1. A knowledge matching method incorporating contextual semantic constraints is characterized in that, comprising: Performing knowledge extraction on the obtained input sentence to obtain the entity vector and the relation vector in the input sentence; converting the input sentence into a text vector; Splicing the entity vector, the relation vector and the text vector to obtain a target vector; Carry out similarity matching between the target vector and the vector record in the knowledge base, and determine a set of similar vectors; The vector with the highest similarity in the similarity vector set is used as the matching result. 2. The knowledge matching method of merging context semantic constraints according to claim 1, wherein the method further comprises the step of building a knowledge base, the step comprising: Train word vectors according to the original training text; Learning the vector representation of entities and relationships of each training text in the knowledge base, and obtaining entity training vectors and relationship training vectors; Perform word segmentation processing on the original training text, and perform vector representation on the result of the word segmentation processing according to the trained word vector to obtain a first vector; According to the splicing process of the entity training vector, the relation training vector and the first vector, the knowledge base construction is completed. 3. the knowledge matching method of fusion context semantic constraint according to claim 2, is characterized in that, before the described step of training word vector according to original training text, also comprises: Preprocessing the original training text, wherein the preprocessing includes removing stop words and word segmentation; The process of removing stop words includes: deleting useless words in the original training text, where the useless words include at least one of the following: modal particles, adverbs, prepositions, and conjunctions; The word segmentation processing includes: converting the text into a word sequence separated by spaces by using the jieba word segmentation tool. 4. The knowledge matching method of merging context semantic constraints according to claim 2, wherein the result of the word segmentation process is represented by a vector according to the trained word vector to obtain a first vector, comprising: After the trained word vectors are added one by one and averaged, the first vector is obtained. 5. The knowledge matching method of merging context semantic constraints according to claim 2, wherein, according to the entity training vector, the relationship training vector and the first vector splicing processing, the knowledge base construction is completed, include: A first triplet is constructed according to the entity and the relationship, and a second triplet is constructed according to the attributes of the entity, and the first triplet is [first entity, relationship, second entity], so The second triple is [entity, attribute, attribute value]; The first triplet and the second triplet are collectively represented as the first vector. 6. The knowledge matching method of integrating context semantic constraints according to claim 1, wherein the expression form of the target vector is: p=(e _p,1 ,r _p ,0,t _p ); Among them, _p is the target vector; e _{p ,1} is the entity vector; r _p is the relation vector; t _p is the text vector; The placeholder vector for two entities. 7. The knowledge matching method of fused context semantic constraints according to claim 1, wherein the described target vector and the vector record in the knowledge base are carried out similarity matching, and the set of similar vectors is determined, comprising: Calculate the similarity between the target vector and the vector records in the knowledge base by using the improved cosine similarity calculation formula, and determine the similarity vector set according to the calculated similarity; Wherein, the cosine similarity calculation formula is:

in,

es _,1 represents the first entity vector of the knowledge base; _rs represents the relation vector of the knowledge base; es _,2 represents the second entity vector of the knowledge base; _ts represents the text semantic vector of the knowledge base; β represents the weight;

ep _,1 represents the first entity vector to be matched; r _p represents the relation vector to be matched; t _p represents the text semantic vector to be matched. 8. A knowledge matching device fused with context semantic constraints, characterized in that, comprising: a knowledge extraction module, used for performing knowledge extraction on the acquired input sentence to obtain the entity vector and the relation vector in the input sentence; a conversion module for converting the input sentence into a text vector; a splicing module for splicing the entity vector, the relation vector and the text vector to obtain a target vector; a similarity matching module, used for similarity matching between the target vector and the vector record in the knowledge base, to determine a set of similar vectors; The determining module is configured to use the vector with the highest similarity in the similarity vector set as the matching result. 9. An electronic device, comprising a processor and a memory; the memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1-7. 10. A computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the method according to any one of claims 1-7.

Images