Scientific research document information extraction method and device

Abstract

The invention provides a scientific research document information extraction method and device, and the method comprises the steps: constructing a term semantic network from a scientific research document; performing confidence degree sorting on the term semantic network to obtain a candidate term extraction result in the scientific research document; according to a candidate term extraction result, extracting concept attributes from the term semantic network; and based on the candidate term extraction result and the concept attribute, clustering is carried out, and a domain knowledge graph is constructed. The method comprises the following steps: constructing a term semantic network, carrying out confidence ranking, obtaining candidate term extraction results of scientific research documents, obtaining concept attributes of the term semantic network, and clustering the candidate term extraction results and the concept attributes to obtain a domain knowledge graph. Automatic acquisition of key information such as domain terms, concepts and attributes in different types of scientific research documents is realized, a description system of corresponding domain knowledge is constructed, a logic structure implied in the scientific research documents is accurately acquired, and powerful support is provided for management, analysis and utilization of scientific research information.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for extracting scientific research document information. Background Art

[0002] In the process of enterprise R&D and operation, there are a large number of scientific research documents. These scientific research documents contain a large amount of information such as professional terms and concepts that need to be extracted and checked. Then scientific research documents usually exist in semi-structured or unstructured forms. It is difficult for existing information extraction methods to accurately and efficiently extract the required information. For example, the existing rule-based information extraction method matches and extracts the target information in scientific research documents by manually defining extraction rules, such as regular expressions and dictionaries. Although it can achieve high extraction accuracy, it has poor scalability and generalization capabilities, and it is difficult to apply to information extraction needs of multiple fields and different types of documents. Another method is to extract information through machine learning. Although it has good generalization capabilities, its performance usually depends on the quality and quantity of training data, and requires a lot of manpower and time costs.

[0003] It should be noted that the above introduction to the technical background is only for the convenience of providing a clear and complete description of the technical solutions of the present application and for the convenience of understanding by those skilled in the art. It cannot be considered that the above technical solutions are well known to those skilled in the art simply because they are described in the background technology section of the present application. Summary of the invention

[0004] The embodiments of the present application provide a method and device for extracting scientific research document information.

[0005] The first embodiment of the present application proposes a method for extracting scientific research document information, comprising:

[0006] Build a semantic network of terms from scientific research documents;

[0007] Based on ranking the confidence of the term semantic network, obtaining the extraction results of candidate terms in the scientific research document;

[0008] Extracting concept attributes from the term semantic network according to the candidate term extraction result;

[0009] Based on clustering the candidate term extraction results and the concept attributes, a domain knowledge graph is constructed.

[0010] The second embodiment of the present application proposes a scientific research document information extraction device, comprising:

[0011] A first generation module, the first generation module is used to construct a term semantic network from scientific research documents;

[0012] A first acquisition module, the first acquisition module is used to obtain the extraction results of candidate terms in the scientific research document based on the confidence ranking of the term semantic network;

[0013] A second acquisition module, the second acquisition module is used to extract concept attributes from the term semantic network according to the candidate term extraction result;

[0014] The second generation module is used to construct a domain knowledge graph based on clustering the candidate term extraction results and the concept attributes.

[0015] The third aspect embodiment of the present application proposes an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the scientific research document information extraction method proposed in the first aspect embodiment of the present application.

[0016] The fourth aspect embodiment of the present application proposes a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by the processor of an electronic device, the electronic device can execute the method proposed in the first aspect embodiment of the present application.

[0017] The fifth aspect embodiment of the present application proposes a computer program product, including a computer program, which implements the method proposed in the first aspect embodiment of the present application when executed by a processor in a communication device.

[0018] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:

[0019] By constructing a semantic network of terms and ranking them by confidence, we can obtain the extraction results of candidate terms in scientific research documents, and then obtain the conceptual attributes of the semantic network of terms. We can cluster the extraction results of candidate terms and conceptual attributes to obtain a domain knowledge graph. We can automatically obtain key information such as domain terms, concepts, and attributes in different types of scientific research documents, build a description system for corresponding domain knowledge, accurately obtain the implicit logical structure in scientific research documents, and provide strong support for the management, analysis, and utilization of scientific research information.

[0020] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0022] Figure 1 A schematic diagram of a method for extracting scientific research document information provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of another method for extracting scientific research document information provided in an embodiment of the present application;

[0024] Figure 3 A schematic diagram of the structure of a scientific research document information extraction device provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of the structure of an electronic device provided according to an embodiment of the present application;

[0026] Figure 5 It is a schematic diagram of the structure of another electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.

[0028] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0029] It should be understood that, although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the embodiments of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at..." or "when..." or "in response to determination".

[0030] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0031] It should be noted that the scientific research document information extraction method provided in any embodiment of the present application can be executed alone, or in combination with possible implementation methods in other embodiments, or in combination with any technical solution in the relevant technology.

[0032] The following describes the scientific research document information extraction method and device of the embodiment of the present application with reference to the accompanying drawings.

[0033] Figure 1 A schematic diagram of a method for extracting scientific research document information provided in an embodiment of the present application. Figure 1 As shown, the method includes but is not limited to the following steps:

[0034] S101, constructs a term semantic network from scientific research documents.

[0035] In a feasible implementation, constructing a term semantic network involves an in-depth understanding and analysis of the document content, and using semantic relationships to connect terms to form a network. Terms are core words in scientific research documents, and they represent key concepts and topics in the documents. The first step in constructing a term semantic network by term extraction can be to automatically or manually extract terms from scientific research documents. For automatic extraction of terms, natural language processing techniques, such as word frequency statistics, word frequency-inverse document frequency, etc., can be used to extract high-frequency words from scientific research documents as candidate terms. Then, by training a word embedding model (such as Word2Vec, BERT, etc.) to calculate the semantic similarity between words (semantic similarity refers to the degree of proximity or similarity between two entities, such as words, sentences, documents, etc., at the semantic level. It focuses on whether the information or meaning conveyed by the entity is similar or consistent. Semantic similarity not only involves similarity at the vocabulary level, such as synonyms and near-synonyms, but also involves similarity at the sentence or document level, such as the same theme and similar content. In addition, it also considers the influence of context and context on semantics), and further screens out terms closely related to the topic. For manual extraction of terms, manual extraction may be more accurate for scientific research documents in some specific fields. This usually involves in-depth reading and understanding of scientific research documents, as well as selecting key terms based on domain knowledge and experience.

[0036] After determining the terms, the next step is to determine the relationships between these terms. These relationships can be direct (such as synonyms and antonyms) or indirect (such as hyponymy and relatedness). Existing semantic network resources such as WordNet and various encyclopedias can be used to determine the relationships between terms. These resources usually contain a large number of words and the relationships between words, which can be used as a reference for building a semantic network of terms. It is also possible to infer the correlation between terms by calculating the co-occurrence frequency of terms in documents. The higher the co-occurrence frequency of terms, the closer the relationship between them may be. It is also possible to use the semantic analysis function in natural language processing technology to further determine the specific relationship between terms. For example, through syntactic analysis and semantic role labeling, it is possible to identify the hyponymy and hyponymy relationships, subject-object relationships, etc. between terms.

[0037] After determining the terms and the relationships between them, you can start building a term semantic network. The term semantic network can be represented in different ways, such as directed graphs, undirected graphs, tree structures, etc. The appropriate expression method should be selected according to specific needs and domain characteristics. In a term semantic network, nodes usually represent terms and edges represent the relationships between terms. Therefore, a node can be created for each term and these nodes can be connected according to the relationships between them. Then, visualization tools such as Gephi, Cytoscape, etc. can be used to present the term semantic network in a graphical way to better analyze and understand the network structure.

[0038] After building the term semantic network, it is usually necessary to verify and optimize the term semantic network to ensure its accuracy and usability. For example, the accuracy and completeness of the network can be verified by comparing it with the existing knowledge base or the opinions of domain experts. If errors or omissions in term relationships are found, they can be corrected and supplemented in a timely manner. Then, based on the verification results, the term semantic network is optimized, such as adjusting the layout of nodes and edges, adding or deleting nodes or edges, etc., to improve the clarity and readability of the term semantic network.

[0039] S102, based on the confidence ranking of the term semantic network, obtaining the extraction results of candidate terms in the scientific research document.

[0040] In a feasible implementation, in the term semantic network, the confidence of each term can reflect its reliability and importance as a domain term. Select a suitable domain, and use association rules (such as FT-tree, full name Frequent Template Tree, which is an extended prefix tree structure that can identify the longest combination of frequently occurring words from log messages) to calculate the strength of association between terms, thereby obtaining the confidence of the terms. The popularity of the terms can also be calculated based on factors such as the number of times the terms are used in scientific research documents, the amount of promotion, and the importance of the context (calculating the popularity of terms usually involves multiple considerations, including the frequency of occurrence of the terms in documents or corpora, distribution range, time trends, etc.), and then the confidence of the terms can be obtained. The semantic confidence of the terms can also be calculated by analyzing the position and relationship of the terms in the term semantic network. This method can comprehensively consider the context information and semantic relationship of the terms.

[0041] After calculating the confidence of each term, they can be sorted from high to low according to the confidence. The sorted term list is the confidence-sorted candidate term extraction result. This result can provide strong support for subsequent scientific research document analysis, knowledge extraction and other tasks.

[0042] It should be noted that when calculating the confidence of a term, multiple aspects are usually considered, such as the linguistic features, domain features, contextual information, etc. of the term, in order to improve the accuracy and reliability of the confidence calculation. For confidence calculation methods based on association rules, popularity or semantic dependencies, the calculation efficiency and accuracy can be improved by optimizing the input parameters. Although the candidate term extraction results based on confidence ranking can be automatically generated, manual verification and correction are still required to ensure the accuracy and reliability of the results.

[0043] S103, extracting concept attributes from the term semantic network according to the candidate term extraction results.

[0044] In a feasible implementation, the process of extracting concept attributes from the term semantic network can be regarded as a task of further mining and refining domain knowledge based on the constructed semantic network. Based on the candidate term extraction results, the attributes associated with a specific term (i.e., concept) are identified. The attributes can be direct features of the term or features implied by the relationship with other terms. Specifically, the formation and definition of concepts are based on the attributes of things, and the attributes of things are the specific manifestations of concepts. When analyzing the relationships in the semantic network, determine which relationships can be regarded as representations of attributes. For example, if there is a relationship "A is an attribute of B" in the network, B can be regarded as an attribute of A; the contextual information of the term in the term semantic network can also be used to infer attributes. For example, if a term is often associated with another term in a specific way (such as "X's Y"), Y may be regarded as an attribute of X.

[0045] S104, constructing a domain knowledge graph based on clustering the candidate term extraction results and concept attributes.

[0046] In a feasible implementation, according to the characteristics of the scientific research document field and the data scale of the candidate term extraction results and attributes, a suitable clustering algorithm is selected, such as K-means, hierarchical clustering, DBSCAN, etc. The candidate term extraction results and attributes are used as input for clustering operations; according to the clustering results, similar candidate term extraction results and attributes are grouped into one category to form concept clusters; then, evaluation indicators such as silhouette coefficient and Calinski-Harabasz index can be used to quantitatively evaluate the concept clusters; according to the evaluation results, the clustering parameters or algorithms are adjusted to obtain concept clusters with better clustering effects.

[0047] Next, we select a suitable knowledge representation model, such as resource description framework, network text language, etc., to represent the domain knowledge graph. Specifically, we use the candidate term extraction results and attributes in the concept cluster as entities and attributes, and the relationships formed in the clustering process as edges in the graph; then we optimize the constructed domain knowledge graph, such as removing redundant information, merging duplicate entities, etc., to ensure the accuracy and completeness of the domain knowledge graph.

[0048] In summary, the scientific research document information extraction method of this embodiment obtains the candidate term extraction results of the scientific research document by constructing a term semantic network and performing confidence sorting, and then obtains the concept attributes of the term semantic network, clusters the candidate term extraction results and the concept attributes, and thus obtains the domain knowledge graph. It realizes the automatic acquisition of key information such as domain terms, concepts and attributes in different types of scientific research documents, constructs a description system of corresponding domain knowledge, accurately obtains the implicit logical structure in the scientific research document, and provides strong support for the management, analysis and utilization of scientific research information.

[0049] Figure 2 This is a flow chart of another method for extracting scientific research document information provided in the embodiment of the present application. Figure 2 As shown, the method includes but is not limited to the following steps:

[0050] S201, obtaining a candidate term set from scientific research documents.

[0051] Optionally, as an example, based on preprocessing the scientific research document (for example, removing unnecessary characters such as punctuation marks, numbers, line breaks, etc.; and then performing word segmentation), the word segmentation result is obtained; according to the preset word frequency threshold and term features, the candidate term set is extracted from the word segmentation result. Specifically, according to the field-related corpus corresponding to the scientific research document (the corpus refers to the text corpus. The text corpus refers to a set of text data that can be used to train a natural language processing model to solve various natural language processing tasks, such as text classification, sentiment analysis, machine translation, intelligent customer service, automatic summarization, text generation, etc.), the noun phrases existing in the corpus can be automatically extracted as a candidate term set through a template matching method. Specifically, the word sequence obtained after the input text corpus is processed by word segmentation and part-of-speech tagging is W=(W1, W2,…, W i …,W n ), where W i represents the i-th word, then the candidate term set T can be expressed as:

[0052] T={c1,c2,…,c i ,…c m}

[0053] Among them, C i represents the i-th candidate term, which consists of several adjacent words W j Composition, and C i The part-of-speech pattern satisfies the predefined term part-of-speech template (part-of-speech pattern usually refers to the categories in a language based on the grammatical function and morphological characteristics of words. These categories, namely parts of speech, are an important concept in linguistics, which help us understand and analyze the structure of sentences. Part of speech, full name Part of Speech, is the basic grammatical attribute of vocabulary, which is divided according to the function and role of vocabulary in a sentence. Different languages ​​have different part-of-speech classifications, but generally include basic categories such as nouns, verbs, adjectives, and adverbs).

[0054] For further introduction to step S201, please refer to the relevant contents in the above embodiment, which will not be repeated here.

[0055] S202, constructing a term semantic network based on fitting the elements in the candidate term set.

[0056] Optionally, as an example, the word grouping and semantic relevance of the candidate term set are determined. Specifically, the candidate term C can be calculated based on the point mutual information method. i The phrase P(C i ), phrase P(C i ) quantified C i The probability of being a grammatically and semantically correct word or phrase. i It consists of k words, namely C i =(W1,W2,…,W k ), then the phrase P(C i ) is calculated as:

[0057]

[0058] Among them, p(W j ,W l ) represents word W j and W l co-occurrence probability (co-occurrence probability measures the probability of another word or entity appearing given a word or entity. Suppose there are two words E and F, the co-occurrence probability can be defined as the probability of word F appearing in the context or environment of word E. This is usually calculated by counting the number of times words E and F appear together in the corpus and dividing it by the total number of times word E appears or the total number of times within the context window); p(W j ) and p(W l ) represent W j and W l The marginal probability (marginal probability is the probability of a random variable or event occurring. It only focuses on the random variable or event itself and is not affected by other random variables or events. In the joint probability distribution of multidimensional random variables, the marginal probability is obtained by summing (for discrete variables) or integrating (for continuous variables) all possible values ​​of other variables in the joint probability distribution), where the co-occurrence probability and marginal probability can be obtained by performing word frequency statistics in the corpus.

[0059] Next, the semantic relevance between candidate terms can be calculated from the candidate term set based on cosine similarity. Specifically, assuming that candidate term C i The word vector is V i (A word vector is a mathematical representation of a word, usually a high-dimensional vector (e.g., several hundred dimensions). Each dimension can represent a certain feature or attribute of the word, although the specific meaning of these features is unknown in most cases (called "latent features" or "implicit features"), then the candidate term C i and candidate term C j The semantic relevance r(C i , C j) is:

[0060]

[0061] Next, based on fitting the word group and semantic relevance, the term semantic network of the scientific research document is constructed. Specifically, based on the word group, the dependency relationship between the candidate terms is identified, and based on the dependency relationship, the network nodes are determined; based on the preset relevance threshold, the weights of each semantic relevance are compared to determine the network edges; based on clustering the network nodes and network edges, the term semantic network G = (V, E) of the scientific research document is constructed, where V is the candidate term node set and E is the semantic relevance edge set between the candidate terms (the semantic relevance edge set usually refers to a set of edges constructed based on the semantic relationship between words, phrases or sentences in a semantic network or knowledge graph. These edges represent the semantic connection between them, and the weight of the edge may indicate the strength of the semantic relevance).

[0062] For further introduction to step S202, please refer to the relevant contents in the above embodiment, which will not be repeated here.

[0063] S203, based on the confidence ranking of the term semantic network, obtaining the extraction results of candidate terms in the scientific research document.

[0064] Optionally, as an example, the confidence of each candidate term in the term semantic network is calculated (the confidence can be obtained based on statistical methods, such as calculating the co-occurrence frequency of the candidate term and other candidate terms in the term semantic network. The higher the co-occurrence frequency, the stronger the correlation between the candidate term and the other candidate terms, and thus the higher its confidence may be; the conditional probability formula can also be used to calculate the probability of occurrence of the candidate term in a given context. The higher the conditional probability, the higher the confidence of the candidate term in the specific context; the semantic similarity between the candidate term and other candidate terms in the term semantic network can also be calculated. The higher the semantic similarity, the closer the semantic relationship between the candidate term and the other terms, and thus the higher its confidence may be; the confidence of the candidate term can also be evaluated based on the position of the candidate term in the ontology, the inheritance relationship or attribute relationship with other terms, etc.).

[0065] After obtaining the confidence of each candidate term, the candidate terms are sorted according to the confidence, and the sorted candidate terms are output as the extraction results. Specifically, assuming S is a seed term set, the candidate term C i The confidence value conf at the kth iteration k (c i ) is calculated as:

[0066]

[0067] Among them, N(Ci ) indicates that in the semantic network G, i The adjacent candidate term node set is calculated iteratively until the confidence value converges or the maximum number of iterations is reached, and finally the candidate terms are sorted according to their confidence values, and the Top-K candidate terms are selected as the final term extraction results.

[0068] For further introduction to step S203, please refer to the relevant contents in the above embodiment, which will not be repeated here.

[0069] S204, extracting concept attributes from the term semantic network according to the candidate term extraction results.

[0070] Optionally, as an example, based on the final term extraction result and based on semantic similarity, the concept attribute corresponding to each candidate term is extracted from the term semantic network. Further, other terms similar to the candidate term can be found based on vector similarity, ontology-based similarity, etc., and concept attributes can be extracted from these terms. It is also possible to analyze the relationship path between the candidate terms to derive the concept attributes related to the candidate terms.

[0071] Furthermore, for structured and semi-structured scientific research documents, such as relational databases, structured tables, and XML / HTML scientific research documents, a rule-based mapping method can be used to extract concept attributes from the term semantic network. As an example, tools such as D2R can be used to convert database schemas, table column names, XML / HTML tags, etc. into a set of attribute candidates A, where A = {a1, a2, …, a L}, and then use the association rule mining method to calculate each attribute candidate a i The concept H in the concept hierarchy system j The correlation degree q(ai,H j ), then q(ai,H j ) is calculated as:

[0072]

[0073] Among them, sup(a i ∪H j ) means a i and H j The frequency of co-occurrence in the data source, sup(a i ) means a i The frequency of occurrence of a single occurrence. Select (ai,H j )Yes, then use ai as H j Candidate attributes of .

[0074] For unstructured and relatively semantically complete texts, such as encyclopedia knowledge bases and domain description texts, a statistical learning-based method is used. On the one hand, the online encyclopedia knowledge base is used to obtain relevant entries through keyword query, backtracking and induction, and the high-frequency attributes in the entry information box are counted as candidates; on the other hand, the dependency syntactic analysis technology is used to extract the semantic dependency relationship between concept terms and attribute terms in the sentence, and the candidate attributes are screened out by combining the concept hierarchy system and seed instances. Furthermore, assuming that after the text corpus is analyzed by dependency syntax, the dependency relationship set L is obtained, and each dependency relationship L in L i It can be represented as a triple:

[0075] L i =(w h ,w t ,r)

[0076] Among them, W h W is the first word t is the last word, r is W h To W t The dependency type.

[0077] Assuming R is a predefined attribute dependency set, the dependency path p(m,n) between concept term m and attribute term n can be expressed as:

[0078] P(m,n)={L1,L2,…Li,…Lz}

[0079] Among them, the last word W of L1 t Equal to m, L z The first word W h equal to n, for 1≤i<z, L i The first word is equal to L i+1 The last word of , and all L i The relationship type r∈R.

[0080] For a given concept m, count its dependency paths p(m,n) with all attribute terms n, and then calculate the frequency f(p(m,n)) of each path. Select p(m,n) with a frequency greater than a preset threshold, and use the corresponding attribute term n as a candidate attribute of concept m.

[0081] For further introduction to step S204, please refer to the relevant contents in the above embodiment, which will not be repeated here.

[0082] S205, constructing a domain knowledge graph based on clustering the candidate term extraction results and concept attributes.

[0083] Optionally, the association relationship between the candidate term extraction results and the concept attributes is obtained; based on the association relationship, the candidate term extraction results and the concept attributes are clustered to determine the domain knowledge graph corresponding to the scientific and technological document.

[0084] For further introduction to step S205, please refer to the relevant contents in the above embodiment, which will not be repeated here.

[0085] In summary, the scientific research document information extraction method of this embodiment obtains the candidate term extraction results of the scientific research document by constructing a term semantic network and performing confidence sorting, and then obtains the concept attributes of the term semantic network, clusters the candidate term extraction results and the concept attributes, and thus obtains the domain knowledge graph. It realizes the automatic acquisition of key information such as domain terms, concepts and attributes in different types of scientific research documents, constructs a description system of corresponding domain knowledge, accurately obtains the implicit logical structure in the scientific research document, and provides strong support for the management, analysis and utilization of scientific research information.

[0086] Figure 3 This is a schematic diagram of the structure of a scientific research document information extraction device provided in an embodiment of the present application. Figure 3 As shown, the scientific research document information extraction device 300 includes:

[0087] A first generating module 301, the first generating module 301 is used to construct a term semantic network from scientific research documents;

[0088] A first acquisition module 302, the first acquisition module 302 is used to obtain candidate term extraction results in scientific research documents based on confidence ranking of term semantic networks;

[0089] The second acquisition module 303 is used to extract concept attributes from the term semantic network according to the candidate term extraction result;

[0090] The second generation module 304 is used to construct a domain knowledge graph based on clustering the candidate term extraction results and concept attributes.

[0091] Figure 4 The present invention is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. Figure 4 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0092] like Figure 4As shown, the electronic device 400 includes a processor 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a memory 406 to a random access memory (RAM) 403. In the RAM 403, various programs and data required for the operation of the electronic device 400 are also stored. The processor 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0093] The following components are connected to the I / O interface 405: a memory 406 including a hard disk, etc.; and a communication part 407 including a network interface card such as a LAN (Local Area Network) card, a modem, etc., and the communication part 407 performs communication processing via a network such as the Internet; a drive 408 is also connected to the I / O interface 405 as needed.

[0094] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 407. When the computer program is executed by the processor 401, the above-mentioned functions defined in the method of the present application are executed.

[0095] In an exemplary embodiment, a storage medium including instructions is also provided, such as a memory including instructions, and the instructions can be executed by the processor 401 of the electronic device 400 to complete the above method. Optionally, the storage medium can be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0096] In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0097] Figure 5 It is a schematic diagram of the structure of another electronic device provided according to an embodiment of the present application. Figure 5 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present application. Figure 5 As shown, the electronic device 500 includes a processor 501 and a memory 502. The memory 502 is used to store program codes, and the processor 501 is connected to the memory 502 to read program codes from the memory 502 to implement the scientific research document information extraction method in the above embodiment.

[0098] Optionally, the number of processors 501 may be one or more.

[0099] Optionally, the electronic device may further include an interface 503, and the number of the interface 503 may be multiple. The interface 503 may be connected to an application program and may receive data from an external device such as a sensor.

[0100] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0101] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for extracting scientific research document information, characterized in that: include: Build a semantic network of terms from scientific research documents; Based on ranking the confidence of the semantic network of the terms, obtaining extraction results of candidate terms in the scientific research document; Extracting concept attributes from the term semantic network according to the candidate term extraction result; Based on clustering the candidate term extraction results and the concept attributes, a domain knowledge graph is constructed.

2. The method according to claim 1, characterized in that The term semantic network is constructed from scientific research documents, including: Obtaining a candidate term set from scientific research documents, and determining the phraseology and semantic relevance of the candidate term set; Based on fitting the phraseology and the semantic relevance, a term semantic network of the scientific research document is constructed.

3. The method according to claim 2, characterized in that The step of obtaining a candidate term set from the scientific research document includes: Based on preprocessing of scientific research documents, word segmentation results are obtained; A candidate term set is extracted from the word segmentation results according to a preset word frequency threshold and term features.

4. The method according to claim 2, characterized in that: Determining the phraseology and semantic relevance of the candidate term set includes: Based on the point mutual information, calculating the phrase nature of each candidate term from the candidate term set; Based on the cosine similarity, the semantic relevance between the candidate terms is calculated from the candidate term set.

5. The method according to any one of claims 2 to 4, characterized in that: The step of constructing a term semantic network of the scientific research document based on fitting the phraseology and the semantic relevance includes: According to the phraseology, identifying dependency relationships between candidate terms, and determining network nodes according to the dependency relationships; Comparing the weights of the semantic relevances based on a preset relevance threshold to determine a network edge; Based on clustering the network nodes and the network edges, a term semantic network of the scientific research document is constructed.

6. The method according to claim 1, characterized in that The step of obtaining the extraction result of candidate terms in the scientific research document based on ranking the confidence of the term semantic network includes: Calculating the confidence of each candidate term in the term semantic network; The candidate terms are sorted according to the confidence level, and the sorted candidate terms are output as extraction results.

7. The method according to claim 1, characterized in that The step of extracting concept attributes from the term semantic network according to the candidate term extraction result includes: According to the candidate term extraction results and based on semantic similarity, a concept attribute corresponding to each candidate term is extracted from the term semantic network.

8. The method according to claim 1, characterized in that: The step of clustering the candidate term extraction results and the concept attributes to construct a domain knowledge graph includes: Obtaining the association relationship between the candidate term extraction result and the concept attribute; According to the association relationship, the candidate term extraction results and the concept attributes are clustered to determine the domain knowledge graph corresponding to the scientific and technological document.

9. A scientific research document information extraction device, characterized in that: include: A first generation module, the first generation module is used to construct a term semantic network from scientific research documents; A first acquisition module, the first acquisition module is used to obtain the extraction results of candidate terms in the scientific research document based on the confidence ranking of the term semantic network; A second acquisition module, the second acquisition module is used to extract concept attributes from the term semantic network according to the candidate term extraction result; The second generation module is used to construct a domain knowledge graph based on clustering the candidate term extraction results and the concept attributes.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method according to any one of claims 1 to 8.

Images