Discipline field lexicon construction method and system based on neural network, terminal and medium

Abstract

The neural network-based subject field dictionary construction method, system, terminal and medium of the present application automatically construct a subject field knowledge dictionary by performing semantic analysis on massive specific subject field data and mining subject professional terms; the present application can automatically construct a subject knowledge architecture from massive academic data, the method has high accuracy and recall rate, and significantly improves knowledge organization efficiency and reduces knowledge maintenance cost. Furthermore, the knowledge structure of the present application is academic, the subject knowledge system hierarchy is complete and unified, and is authoritative and practical, and the present application can realize automatic construction and regular incremental updating of the knowledge dictionary, adopts the most advanced artificial intelligence technology, and has strong core semantic correlation mapping capability. The present application can effectively replace 90% of manual work, save economic and time costs, and has strong universality and operability, has certain engineering application prospects, and effectively helps the knowledge service field to perform semantic indexing, organization and retrieval of academic resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and particularly relates to a subject field vocabulary construction method and system based on a neural network, a terminal and a medium. BACKGROUND

[0002] As the most important knowledge organization tool in the field of information retrieval, the subject vocabulary can be widely used in semantic indexing, organization and retrieval of literature resources. In the existing technology, there is little research on the automatic construction of subject knowledge vocabulary, and most of them are completed by manual arrangement, which requires a large amount of manpower, material resources and financial resources. For example, the science and technology knowledge organization system (STKOS) of the Chinese Academy of Sciences Literature Information Center also has problems such as weak timeliness and correlation mapping ability, high updating and maintenance cost, and non-uniform depth of knowledge system hierarchy, which is not conducive to knowledge organization management. This will greatly affect the application of subject vocabulary. Therefore, the intelligent construction of subject vocabulary is an urgent problem to be solved in the field of knowledge organization management. SUMMARY

[0003] In view of the above-mentioned shortcomings of the prior art, the purpose of the present application is to provide a subject field vocabulary construction method and system based on a neural network, a terminal and a medium, which solves the problem that there is little research on the automatic construction of subject knowledge vocabulary in the prior art, and most of them are completed by manual arrangement, which requires a large amount of manpower, material resources and financial resources, and has problems such as weak timeliness and correlation mapping ability, high updating and maintenance cost, and non-uniform depth of knowledge system hierarchy, which is not conducive to knowledge organization management.

[0004] To achieve the above-mentioned objectives and other related objectives, the present application provides a subject field vocabulary construction system based on a neural network, comprising: obtaining one or more basic seed libraries annotated by standardization; wherein each basic seed library comprises one or more seed words and one or more subject classification systems having a correlation relationship with the seed words; based on the seed words, obtaining one or more professional corpora of academic data, and generating a professional corpus; performing word segmentation on the data in the professional corpus, and calculating the similarity between each word segmentation corresponding to the professional corpus and the seed words to obtain one or more candidate academic words; performing semantic expansion on the seed words to obtain sentences respectively corresponding to the seed words; training the academic term automatic classification model using the sentences corresponding to the seed words and the subject classification systems having a correlation relationship with the seed words; performing semantic expansion on each candidate academic word to obtain sentences respectively corresponding to each candidate academic word; inputting the sentences respectively corresponding to each candidate academic word into the academic term automatic classification model, and filtering the final subject classification system corresponding to each candidate academic word through the subject classification system; and generating a subject field vocabulary according to each candidate academic word and the final subject classification system information corresponding thereto.

[0005] In one or more embodiments of the present application, the obtaining one or more professional corpora of the academic data based on various sub-words and generating a professional corpus comprises: determining a data source of the academic data; extracting core data in the academic data according to the data source; and screening one or more professional corpora of the academic data based on various sub-words and generating a professional corpus according to the core data.

[0006] In one or more embodiments of the present application, the word segmentation of the data in the professional corpus and the calculation of the similarity between each segmented word corresponding to the professional corpus and various sub-words to screen one or more candidate academic words comprises: word segmentation of the data in the professional corpus to obtain one or more segmented words; vector representation of each segmented word and various sub-words based on a trained word embedding space vector model to obtain a representation vector corresponding to each segmented word and various sub-words; calculation of the similarity between each segmented word and various sub-words of the professional corpus based on the representation vector corresponding to each segmented word and various sub-words; and screening one or more candidate academic words from each segmented word based on the similarity between each segmented word and various sub-words.

[0007] In one or more embodiments of the present application, the training of the term automatic classification model using sentences corresponding to various sub-words and various subject classification systems having an associated relationship with various sub-words comprises: obtaining sentences corresponding to various sub-words and various subject classification systems having an associated relationship with various sub-words from each basic seed library as training data; vector encoding of the training data to obtain a Bert semantic vector; inputting the Bert semantic vector into a TextCNN model to train a term automatic classification model.

[0008] In one or more embodiments of the present application, the inputting of the sentences corresponding to each candidate academic word into the term automatic classification model and screening of the final subject classification system corresponding to each candidate academic word through the subject classification system comprises: inputting the sentences corresponding to each candidate academic word into the term automatic classification model to obtain one or more subject classification systems having an associated relationship with each candidate academic word; and screening the final subject classification system corresponding to each candidate academic word based on the closeness of the associated relationship of each subject classification system corresponding to each candidate academic word.

[0009] In one or more embodiments of the present application, the subject field word library comprises: one or more candidate academic words and subject classification systems having an associated relationship with each candidate academic word.

[0010] To achieve the above object and other related objects, the present application provides a neural network-based subject field vocabulary construction system, comprising: a basic seed library acquisition module configured to acquire one or more basic seed libraries that are standardized and labeled; wherein each basic seed library comprises one or more seed words and one or more subject classification systems that are associated with the seed words; a professional corpus library module configured to acquire one or more professional corpora of academic data based on the seed words, and generate a professional corpus library; a candidate academic word acquisition module configured to perform word segmentation on data in the professional corpus library, and calculate the similarity between each segmented word in the professional corpus library and the seed words, to obtain one or more candidate academic words; a seed word semantic expansion module configured to perform semantic expansion on the seed words respectively, to obtain sentences respectively corresponding to the seed words; a classification model training module configured to train the automatic classification model of the technical terms using the sentences corresponding to the seed words and the subject classification systems that are associated with the seed words; a candidate academic word semantic expansion module configured to perform semantic expansion on the candidate academic words, to obtain sentences respectively corresponding to the candidate academic words; a candidate academic word classification module configured to input the sentences respectively corresponding to the candidate academic words into the automatic classification model of the technical terms, and filter the subject classification systems to obtain the final subject classification system corresponding to each candidate academic word; and a subject field vocabulary establishment module configured to generate a subject field vocabulary based on the candidate academic words and the final subject classification system information respectively corresponding to the candidate academic words.

[0011] In one or more embodiments of the present application, the training of the automatic classification model of the technical terms using the sentences corresponding to the seed words and the subject classification systems that are associated with the seed words comprises: using the sentences corresponding to the seed words and the subject classification systems that are associated with the seed words from the basic seed libraries as training data; performing vector encoding on the training data to obtain Bert semantic vectors; inputting the Bert semantic vectors into a TextCNN model to train an automatic classification model of technical terms.

[0012] To achieve the above object and other related objects, the present application provides a neural network-based subject field vocabulary construction terminal, comprising: one or more memories and one or more processors; the one or more memories are configured to store a computer program; the one or more processors are connected to the memories and are configured to run the computer program to execute the neural network-based subject field vocabulary construction method.

[0013] To achieve the above object and other related objects, the present application provides a computer-readable storage medium storing a computer program, wherein the computer program is run by one or more processors to execute the neural network-based subject field vocabulary construction method.

[0014] As described above, the neural network-based subject-domain lexicon construction method, system, terminal, and medium of this application automatically construct a subject-domain knowledge lexicon by performing semantic analysis and mining subject-specific terminology from massive amounts of data on specific subject domains. This invention can automatically construct a subject knowledge architecture from massive amounts of academic data. The method has high accuracy and recall, significantly improves knowledge organization efficiency, and reduces knowledge maintenance costs. Furthermore, the knowledge structure of this invention is academic, with a unified and complete hierarchical subject knowledge system that is authoritative and practical. It can achieve automated construction and regular incremental updates of the knowledge lexicon, employing state-of-the-art artificial intelligence technology with strong core semantic association mapping capabilities. This invention can also effectively replace 90% of manual work, saving economic and time costs. At the same time, the method has strong versatility and operability, and has certain engineering application prospects, effectively assisting the knowledge service field in the semantic indexing, organization, and retrieval of academic resources. Attached Figure Description

[0015] Figure 1 The diagram shows a flowchart of the method for constructing a subject domain thesaurus based on neural networks in an embodiment of this application.

[0016] Figure 2 The diagram shown is a schematic representation of the subject domain lexicon construction system based on neural networks in this application embodiment.

[0017] Figure 3 The diagram shown is a schematic representation of the terminal for constructing a subject-specific thesaurus based on a neural network, as described in this application embodiment. Detailed Implementation

[0018] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and application systems without departing from the spirit of this application. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0019] The embodiments of this application will now be described in detail with reference to the accompanying drawings, so that those skilled in the art can easily implement the application. This application may be embodied in many different forms and is not limited to the embodiments described herein.

[0020] To clearly illustrate this application, components unrelated to the description are omitted, and the same or similar constituent elements throughout the specification are given the same reference numerals.

[0021] Throughout the specification, when it is said that a certain component is "connected" to another component, this includes not only the case of "direct connection", but also the case of "indirect connection" in which other elements are interposed therebetween. In addition, when it is said that a certain component "includes" a certain constituent element, unless specifically stated to the contrary, other constituent elements are not excluded, but it means that other constituent elements can also be included.

[0022] When it is said that a certain component is "on" another component, this can be directly on the other component, but can also be accompanied by other components therebetween. When it is said in contrast that a certain component is "directly" on another component, there are no other components therebetween.

[0023] Although the terms first, second, etc. can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, the first interface and the second interface, etc. descriptions. Also, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises", "comprising", "includes" and / or "including", mean that there are the mentioned features, steps, operations, elements, components, items, kinds and / or groups, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, items, kinds and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean either one or any combination. Thus, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C". This exception occurs only when the combination of elements, functions, steps or operations is inherently mutually exclusive by its nature.

[0024] The professional terms used herein are only used to refer to specific embodiments, and are not intended to limit the present application. The singular form used herein, unless the sentence explicitly indicates the opposite meaning, also includes the plural form. The meaning of "include" used in the specification is to specify certain characteristics, regions, integers, steps, operations, elements and / or components, and does not exclude the presence or addition of other characteristics, regions, integers, steps, operations, elements and / or components.

[0025] Terms denoting relative positions such as "lower," "upper," and the like can be used for ease of describing the relationship of one component to another component as illustrated in the drawings. Such terms encompass not only the meanings indicated in the drawings, but also other meanings or operations of the device in use. For example, if the device in the drawing is turned upside down, a component that was described as being "lower" than another component is now described as being "upper" than the other component. Thus, the exemplary term "lower" encompasses both upper and lower positions. The device can be rotated 90° or another angle, and the terms denoting relative positions are interpreted accordingly.

[0026] Although not differently defined, technical and scientific terms used herein include those terms that are commonly used in the art to which the present application pertains, and all terms have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. Terms defined in a general dictionary are additionally interpreted to have meanings consistent with the contents of related technical documents and the present disclosure, and are not over-interpreted as ideal or very formal meanings unless defined.

[0027] In recent years, the rapid development of big data and natural language processing technology makes it possible to accurately mine large-scale academic resources. Through the application and improvement of these cutting-edge technologies, a discipline field vocabulary can be automatically constructed, which can reduce costs and increase efficiency, and effectively solve the problem of knowledge application management.

[0028] Therefore, in view of the deficiencies of the prior art, by performing semantic analysis on massive specific discipline field data and mining discipline professional terms, a discipline field knowledge vocabulary is automatically constructed. The present application can automatically construct a discipline knowledge architecture from massive academic data. The method has high accuracy and recall rate, and significantly improves the efficiency of knowledge organization and reduces the cost of knowledge maintenance. The knowledge structure of the present application is academic, the discipline knowledge system is hierarchical and complete, and is authoritative and practical. The knowledge vocabulary can be automatically constructed and periodically updated, and the most advanced artificial intelligence technology is used, with strong core semantic correlation mapping capability. The present application can effectively replace 90% of manual work, saving economic and time costs. The method is universal and operable, has certain engineering application prospects, and effectively helps the knowledge service field to perform semantic indexing, organization and retrieval of academic resources.

[0029] The embodiments of the present application will be described in detail below with reference to the accompanying drawings. The present application can be embodied in various different forms, and is not limited to the embodiments described herein.

[0030] As shown in Figure 1 FIG. 1 shows a flowchart of a method for constructing a discipline field vocabulary based on a neural network according to an embodiment of the present application.

[0031] The method comprises:

[0032] S11: Obtain one or more basic seed libraries with standardized annotations.

[0033] In detail, each basic seed library includes one or more seed words and one or more subject classification systems associated with the seed words. Specifically, the basic seed library includes subject terminology words, subject systems, and the association between the subject terminology words and the subject systems. Some of the subject terminology words are annotated as seed words and have an annotated subject classification system associated with the seed words. The association between the subject terminology words and the subject classification systems can also be obtained.

[0034] Optionally, the standardized annotations of the basic seed library are implemented by manual standardized processes. Preferably, a small number of basic seed libraries are established, which include seed words and subject classification systems annotated by subject field experts through manual annotation using strict standardized processes, including:

[0035] 1) The annotators independently complete the annotation of the seed words and the subject classification systems associated with the seed words in the seed word library they are responsible for to obtain annotation results.

[0036] 2) According to the annotation results, the consistency degree between each two annotators is calculated using the kappa coefficient.

[0037] 3) The annotators check their respective annotations and negotiate the reasons for the differences in the annotations. If a consensus can be reached, a consistent annotation result is formed. If not, the ambiguous part of the annotation is marked with color.

[0038] 4) The instructor reviews the annotation results of the annotators and confirms the annotation results to form the final basic seed library as a template for model learning and training.

[0039] Step S12: Based on the various seed words, one or more professional corpora of academic data are obtained, and a professional corpus library is generated.

[0040] Optionally, step S12 includes determining the data source of the academic data, extracting core data from the academic data according to the data source, and based on the various seed words, filtering one or more high-quality professional corpora of academic data from the core data, and generating a professional corpus library.

[0041] It should be noted that the data source can be any academic database, such as Wanfang database, Zhiwu database, and encyclopedia database, etc. Each database has different formats and storage types, so different extraction methods are required for core data extraction. The core data includes but is not limited to literature abstracts, keywords, and other data.

[0042] Optionally, the professional corpus is subjected to multi-source heterogeneous data integration, cleaning and preprocessing operations to generate a professional corpus.

[0043] Step S13: Tokenizing the data in the professional corpus and calculating the similarity between each token in the professional corpus and various subwords to obtain one or more candidate academic words.

[0044] Optionally, step S13 includes:

[0045] Tokenizing the data in the professional corpus using a tokenization tool to obtain one or more tokens; wherein the tokenization tool can be any tokenization tool, for example, using the Jieba tool.

[0046] Based on the trained word embedding space vector model, the tokens and various subwords are vector represented to obtain the representation vectors corresponding to the tokens and various subwords;

[0047] Based on the representation vectors corresponding to the tokens and various subwords, the similarity between the tokens and various subwords of the professional corpus is calculated;

[0048] Based on the similarity between each token and various subwords, one or more candidate academic words are selected from the tokens, and a candidate academic word set can also be generated.

[0049] Optionally, based on the trained word embedding space vector model, the tokens and various subwords are vector represented to obtain the representation vectors corresponding to the tokens and various subwords, which includes using a word embedding space vector model trained using the FastText algorithm to vector represent the tokens and various subwords; in this process, the algorithm and parameters can be adjusted according to the data to obtain better representation vectors.

[0050] Optionally, the way of selecting one or more candidate academic words from the tokens based on the similarity between each token and various subwords includes:

[0051] (1) According to the set similarity threshold, the candidate academic words are selected based on the similarity threshold;

[0052] (2) Based on the similarity corresponding to each token, the similarity is arranged, and the candidate academic words are selected based on the arrangement order.

[0053] For example, the top few tokens in the similarity ranking are selected as candidate academic words. For example, a Top20 candidate academic word set is obtained.

[0054] Step S14: Each subword is subjected to semantic expansion to obtain sentences respectively corresponding to each subword.

[0055] Optionally, each sub-word is subjected to complete context semantic expansion to form a sentence containing the seed word.

[0056] Step S15: training the automatic classification model of the term word by using the sentences corresponding to the various sub-words and the various subject classification systems having the association relationship with the various sub-words.

[0057] Optionally, the automatic classification model of the term word is trained by using the sentences corresponding to the various sub-words and the various subject classification systems having the association relationship with the various sub-words as training data.

[0058] Optionally, the training of the automatic classification model of the term word by using the sentences corresponding to the various sub-words and the various subject classification systems having the association relationship with the various sub-words comprises:

[0059] The sentences corresponding to the various sub-words and the various subject classification systems having the association relationship with the various sub-words from the basic seed library are obtained as training data.

[0060] The training data is vector encoded by the Bert semantic embedding layer to respectively generate word embedding vectors, sentence embedding vectors, and position embedding vectors, and the Bert semantic vectors are generated by a pre-training model.

[0061] The Bert semantic vectors are input into the TextCNN model, convolution and pooling are sequentially performed, and the flatten operation is performed, and finally the softmax classification results of the semantic expansion sentences of the various sub-words are obtained as the final model output data, and the automatic classification model of the term word is trained and obtained.

[0062] Step S16: performing semantic expansion on each candidate academic word to obtain a sentence corresponding to each candidate academic word.

[0063] Optionally, each candidate academic word is subjected to complete context semantic expansion to form a sentence containing the seed word.

[0064] Step S17: inputting the sentences corresponding to each candidate academic word into the automatic classification model of the term word, and screening the subject classification system to obtain the final subject classification system corresponding to each candidate academic word.

[0065] Optionally, step S17 comprises: inputting the sentences corresponding to each candidate academic word into the automatic classification model of the term word, automatically classifying and inferring the subject system corresponding to each candidate academic word to obtain one or more subject classification systems having an association relationship with each candidate academic word; screening the final subject classification system corresponding to each candidate academic word based on the closeness of the association relationship of each subject classification system corresponding to each candidate academic word, completing the multi-label subject classification of the academic new word, and forming the hierarchical association relationship between the academic word and the subject system.

[0066] Optionally, the final subject classification system corresponding to each candidate academic term is obtained based on the closeness of the association relationship of each subject classification system corresponding to each candidate academic term, comprising:

[0067] The model output categories corresponding to all sentence corpora of each candidate academic term are counted, one or more categories with the highest closeness of the association relationship of each subject classification system corresponding to each candidate academic term are selected as the candidate categories of the candidate academic term, and the categories greater than or equal to the average value are selected as the final categories of the candidate academic term.

[0068] Step S18: generating a subject field vocabulary according to each candidate academic term and the final subject classification system information corresponding thereto.

[0069] Optionally, the subject field vocabulary comprises one or more candidate academic terms and subject classification systems having an association relationship with each candidate academic term.

[0070] Optionally, the generated knowledge vocabulary is verified by a field expert to control the quality of the subject vocabulary.

[0071] Optionally, the academic terms and the association relationship thereof with the subject system that pass the expert verification are used as the final generated subject field vocabulary, which is finally composed of a multi-level subject classification system, subject terminology, and subject word association mapping relationship for a knowledge system.

[0072] Similar to the above-mentioned embodiment principle, the present application provides a subject field vocabulary construction system based on a neural network.

[0073] The following provides specific embodiments in combination with the drawings:

[0074] As Figure 2 The structure of a subject field vocabulary construction system based on a neural network in an embodiment of the present application is shown.

[0075] The system comprises:

[0076] The basic seed library acquisition module 21 is configured to acquire one or more basic seed libraries that are standardized and labeled, wherein each basic seed library comprises one or more seed words and one or more subject classification systems having an association relationship with each seed word.

[0077] The professional corpus module 22 is configured to acquire one or more professional corpora of academic data based on each seed word and generate a professional corpus. It should be noted that each seed word can be directly input, or the professional corpus module 22 is connected to the basic seed library acquisition module 21 to acquire the seed library from the seed library acquisition module 21, or the basic seed library output by the seed library acquisition module 21 stored in the storage.Figure 2 For example, the professional corpus module 22 is connected to the basic seed library acquisition module 21.

[0078] The candidate academic term acquisition module 23 is connected to the basic seed library acquisition module 21 and the professional corpus module 22, and is configured to perform word segmentation on data in the professional corpus, and calculate similarity between each word segmentation corresponding to the professional corpus and each sub-term, so as to screen one or more candidate academic terms;

[0079] The seed term semantic expansion module 24 is configured to perform semantic expansion on each sub-term respectively, to obtain a sentence corresponding to each sub-term respectively; it should be noted that the sub-term can be directly input, or the seed term semantic expansion module 24 is connected to the basic seed library acquisition module 21 to obtain the sub-term from the basic seed library acquisition module 21, or the sub-term is obtained from the basic seed library output by the seed library acquisition module 21 stored in the storage. Figure 2 For example, the seed term semantic expansion module 24 is connected to the basic seed library acquisition module 21,

[0080] The classification model training module 25 is connected to the seed term semantic expansion module 24 and the basic seed library acquisition module 21, and is configured to train the automatic classification model of the technical term by using the sentence corresponding to each sub-term and each subject classification system having a correlation with the sub-term;

[0081] The candidate academic term semantic expansion module 26 is connected to the seed term semantic expansion module 24, and is configured to perform semantic expansion on each candidate academic term to obtain a sentence corresponding to each candidate academic term respectively;

[0082] The candidate academic term classification module 27 is connected to the candidate academic term semantic expansion module 26, and is configured to input the sentence corresponding to each candidate academic term into the automatic classification model of the technical term, and screen a final subject classification system corresponding to each candidate academic term through the subject classification system;

[0083] The subject field library establishing module 28 is connected to the candidate academic term classification module 27, and is configured to generate a subject field library according to each candidate academic term and the final subject classification system information corresponding to the candidate academic term respectively.

[0084] It should be noted that it should be understood that Figure 2 The division of each module in the system embodiment is only a logical function division, and all or part of the modules can be integrated into one physical entity, or can be physically separated. Moreover, the modules can all be implemented in the form of software called by a processing element; or all can be implemented in the form of hardware; or part of the modules can be implemented in the form of software called by a processing element, and part of the modules can be implemented in the form of hardware.

[0085] For example, each module can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), or one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs), etc. For another example, when a certain module above is implemented in the form of a processing element scheduling code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor that can invoke program code. For another example, these modules can be integrated together to implement in the form of a system-on-a-chip (SOC).

[0086] Therefore, since the implementation principle of the neural network-based subject field corpus construction system has been described in the foregoing embodiments, it is not repeated here.

[0087] Optionally, the professional corpus module 22 is configured to determine a data source of the academic data, extract core data in the academic data according to the data source, and filter one or more professional corpora of the academic data based on various sub-words and the core data, and generate a professional corpus.

[0088] Optionally, the candidate academic word acquisition module 23 is configured to perform word segmentation on data in the professional corpus to obtain one or more segmented words, perform vector representation on each segmented word and various sub-words based on a trained word embedding space vector model to obtain a representation vector corresponding to each segmented word and various sub-words, calculate a similarity between each segmented word and various sub-words of the professional corpus based on the representation vector corresponding to each segmented word and various sub-words, and filter one or more candidate academic words from the segmented words based on the similarity between each segmented word and various sub-words.

[0089] Optionally, the classification model training module 25 is configured to obtain sentences corresponding to various sub-words and various subject classification systems having an association relationship with various sub-words from each basic seed library as training data, perform vector encoding on the training data to obtain a Bert semantic vector, and input the Bert semantic vector into a TextCNN model to train an automatic classification model of academic words.

[0090] Optionally, the candidate academic word classification module 27 is configured to input the sentence corresponding to each candidate academic word into the automatic classification model of the term dictionary, to obtain one or more subject classification systems having a correlation relationship with each candidate academic word; and based on the closeness of the correlation relationship of each subject classification system corresponding to each candidate academic word, to screen a final subject classification system corresponding to each candidate academic word.

[0091] Optionally, the subject field dictionary includes one or more candidate academic words and a subject classification system having a correlation relationship with each candidate academic word.

[0092] As Figure 3 The structure of the neural network-based subject field dictionary construction terminal 30 in the embodiment of the present application is shown.

[0093] The neural network-based subject field dictionary construction terminal 30 includes a memory 31 and a processor 32. The memory 31 is configured to store a computer program; and the processor 32 is configured to run the computer program to implement the following functions. Figure 1 The neural network-based subject field dictionary construction method.

[0094] Optionally, the number of memories 31 can be one or more, the number of processors 32 can be one or more, and the number of application programs can be one or more. Figure 3 In the embodiment, one is taken as an example.

[0095] Optionally, the processor 32 in the neural network-based subject field dictionary construction terminal 30 will load one or more instructions corresponding to the process of the application program into the memory 31, and run the application program stored in the first memory 31 by the processor 32, so as to implement the various functions in the neural network-based subject field dictionary construction method. Figure 1 The neural network-based subject field dictionary construction method. Figure 1 The neural network-based subject field dictionary construction method.

[0096] Optionally, the memory 31 can include, but is not limited to, a high-speed random access memory, a nonvolatile memory, for example, one or more disk storage devices, a flash memory device, or other nonvolatile solid-state storage device; the processor 32 can include, but is not limited to, a central processing unit (CPU), a network processor (NP), etc., and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic component, a discrete hardware component.

[0097] Optionally, the processor 32 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc., and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic component, a discrete hardware component.

[0098] The application also provides a computer readable storage medium storing a computer program, the computer program being configured to implement the neural network-based subject field vocabulary construction method shown in the above Figure 1 The computer readable storage medium can include, but is not limited to, a floppy disk, an optical disk, a CD-ROM (compact disk-read only memory), a magneto-optical disk, a ROM (read only memory), a RAM (random access memory), an EPROM (erasable programmable read only memory), an EEPROM (electrically erasable programmable read only memory), a magnetic or optical card, a flash memory, or other types of media / machine readable media suitable for storing machine executable instructions. The computer readable storage medium can be a product that is not connected to a computer device, or can be a component connected to a computer device for use.

[0099] To sum up, the neural network-based subject field lexicon construction method, system, terminal and medium of the application automatically construct the subject field knowledge lexicon by performing semantic analysis on massive specific subject field data and mining subject professional terms; the application can automatically construct a subject knowledge architecture from massive academic data, the method has high accuracy and recall rate, and significantly improves knowledge organization efficiency and reduces knowledge maintenance cost. Moreover, the knowledge structure of the application has academic nature, the subject knowledge system hierarchy is complete and unified, is authoritative and practical, can realize automatic construction and regular incremental update of the knowledge lexicon, adopts the most advanced artificial intelligence technology, and has strong core semantic correlation mapping capability. The application can effectively replace 90% of manual work, save economic and time costs, and has strong universality and operability, has certain engineering application prospect, and effectively helps semantic indexing, organization and retrieval of academic resources in the knowledge service field.

[0100] The above embodiments only exemplarily illustrate the principles and effects of the application, and are not used to limit the application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the application. Therefore, all equivalent modifications or changes completed by those skilled in the art without departing from the spirit and technical thought disclosed by the application should be covered by the claims of the application.

Claims

1. A neural network-based academic field lexicon construction method, characterized by, The method comprises the following steps: obtaining one or more basic seed libraries with standardized annotations, wherein each basic seed library comprises one or more seed words and one or more subject classification systems associated with the seed words; obtaining one or more professional corpora of academic data based on the seed words and generating a professional corpus; performing word segmentation on the data in the professional corpus, calculating the similarity between each segmented word in the professional corpus and the seed words, and screening one or more candidate academic words; performing semantic expansion on the seed words respectively to obtain sentences respectively corresponding to the seed words; training the automatic classification model of the term word by using the sentences corresponding to the seed words and the subject classification systems associated with the seed words; performing semantic expansion on each candidate academic word to obtain sentences respectively corresponding to each candidate academic word; inputting the sentences respectively corresponding to each candidate academic word into the automatic classification model of the term word, and screening the final subject classification system corresponding to each candidate academic word through the subject classification system; generating a subject field word library according to each candidate academic word and the final subject classification system information respectively corresponding to the candidate academic word.

2. The neural network-based academic field lexicon construction method of claim 1, wherein, The method comprises the following steps: determining the data source of the academic data; extracting core data from the academic data according to the data source; screening one or more professional corpora of academic data based on the seed words and the core data, and generating a professional corpus.

3. The neural network-based academic field lexicon construction method of claim 1, wherein, The method comprises the following steps: performing word segmentation on the data in the professional corpus to obtain one or more segmented words; performing vector representation on each segmented word and the seed words based on a trained word embedding space vector model to obtain a representation vector corresponding to each segmented word and the seed words; calculating the similarity between each segmented word and the seed words based on the representation vector corresponding to each segmented word and the seed words; screening one or more candidate academic words from the segmented words based on the similarity between each segmented word and the seed words.

4. The neural network-based academic field lexicon construction method of claim 1, wherein, The method comprises the following steps: obtaining the sentences corresponding to the seed words and the subject classification systems associated with the seed words from the basic seed libraries as training data; performing vector encoding on the training data to obtain a Bert semantic vector; inputting the Bert semantic vector into a TextCNN model to train the automatic classification model of the term word.

5. The neural network-based academic field lexicon construction method of claim 1, wherein, The method comprises the following steps: inputting the sentences respectively corresponding to each candidate academic word into the automatic classification model of the term word to obtain one or more subject classification systems associated with each candidate academic word; The final subject classification system corresponding to each candidate academic term is obtained based on the closeness of the association relationship of each subject classification system corresponding to each candidate academic term.

6. The neural network-based academic field lexicon construction method of claim 1, wherein, The subject field library includes one or more candidate academic terms and subject classification systems having an association relationship with each candidate academic term. 7.A neural network-based academic field lexicon construction system, characterized by, The method comprises the following steps: The basic seed library acquisition module is configured to acquire one or more basic seed libraries that are standardized and labeled; each basic seed library includes one or more seed terms and one or more subject classification systems having an association relationship with each seed term; The professional corpus module is configured to acquire one or more professional corpora of academic data based on each seed term and generate a professional corpus; The candidate academic term acquisition module is configured to perform word segmentation on the data in the professional corpus and calculate the similarity between each segmented word corresponding to the professional corpus and each seed term to obtain one or more candidate academic terms; The seed term semantic expansion module is configured to perform semantic expansion on each seed term to obtain a sentence corresponding to each seed term, respectively; The classification model training module is configured to train the automatic classification model of the term based on the sentences corresponding to each seed term and the subject classification systems having an association relationship with each seed term; The candidate academic term semantic expansion module is configured to perform semantic expansion on each candidate academic term to obtain a sentence corresponding to each candidate academic term, respectively; The candidate academic term classification module is configured to input the sentences corresponding to each candidate academic term, respectively, into the automatic classification model of the term and perform subject classification system screening to obtain a final subject classification system corresponding to each candidate academic term; The subject field library establishment module is configured to generate a subject field library based on each candidate academic term and the final subject classification system information corresponding to each candidate academic term, respectively.

8. The neural network-based academic domain lexicon construction system of claim 7, wherein, The training of the automatic classification model of the term based on the sentences corresponding to each seed term and the subject classification systems having an association relationship with each seed term comprises the following steps: The sentences corresponding to each seed term and the subject classification systems having an association relationship with each seed term from each basic seed library are obtained as training data; The training data is vector encoded to obtain a Bert semantic vector; The Bert semantic vector is input into a TextCNN model to train an automatic classification model of the term. 9.A neural network-based academic field lexicon construction terminal, comprising: The method comprises the following steps: One or more memories and one or more processors; The one or more memories are configured to store a computer program; The one or more processors are connected to the memories and are configured to run the computer program to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored, and the computer program is run by one or more processors to perform the method of any one of claims 1 to 7.

Images