A document retrieval method, device and medium based on multidimensional semantic indexing

Abstract

The embodiments of the present application provide a document retrieval method, device and medium based on multidimensional semantic indexing, the method comprising: obtaining data to be retrieved obtained through the input operation of the current user, and retrieving at least one relevant document based on the data to be retrieved; displaying at least one knowledge object corresponding to each relevant document, wherein the knowledge object is used to characterize the semantic entity in each relevant document; in response to the target knowledge object selected by the current user in the at least one knowledge object, searching for the paper fragments that co-occur in the document with the target knowledge object to obtain the retrieval results. Through some embodiments of the present application, knowledge objects and paper fragments related to the user input can be retrieved, thereby improving the accuracy and granularity of the retrieval.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of document retrieval, and specifically to a document retrieval method, device and medium based on multidimensional semantic indexing. Background Art

[0002] In related technologies, the document retrieval system supports the application scenario of the paper list retrieval service by building a basic information database of the paper, thereby helping researchers to quickly find the required papers and knowledge from large-scale, semi-structured scientific and technological documents, and returning the results in the form of relevant paper lists to users by providing general search, advanced search, secondary search and citation search functions. However, the above technical solution can only provide the overall documents related to the user input, resulting in reduced accuracy and granularity of document retrieval.

[0003] Therefore, how to improve the accuracy and granularity of literature retrieval becomes a problem that needs to be solved. Summary of the invention

[0004] Embodiments of the present application provide a document retrieval method, device and medium based on multidimensional semantic indexing. Through some embodiments of the present application, at least knowledge objects and paper fragments related to user input can be retrieved, thereby improving the accuracy and granularity of the retrieval.

[0005] In a first aspect, the present application provides a document retrieval method based on multidimensional semantic indexing, the method comprising: obtaining data to be retrieved obtained through the input operation of the current user, and retrieving at least one relevant document based on the data to be retrieved; displaying at least one knowledge object corresponding to each relevant document, wherein the knowledge object is used to characterize semantic entities in each relevant document; in response to a target knowledge object selected by the current user from the at least one knowledge object, searching for paper fragments that co-occur in the document with the target knowledge object to obtain retrieval results.

[0006] Therefore, unlike the related art that only searches for paper lists, the embodiment of the present application enriches the document retrieval function by searching for semantic entities in the documents and displaying paper fragments where the semantic entities co-occur, providing users with richer search results, thereby improving the accuracy and granularity of the search.

[0007] In combination with the first aspect, in one embodiment of the present application, the retrieving at least one relevant document based on the data to be retrieved includes: retrieving at least one relevant document corresponding to the data to be retrieved through basic information of the paper; before displaying at least one knowledge object corresponding to each relevant document, the method also includes: searching for at least one knowledge object of each relevant document in the abstract step information and / or knowledge object information, wherein the abstract step information includes the various components of the document abstract, and the knowledge object information includes the semantic entities contained in the sentences in each relevant document.

[0008] Therefore, the embodiment of the present application can extract fine-grained knowledge from the document for the user by searching for knowledge objects corresponding to the document, so that the user can quickly identify the key points from the document and quickly understand the semantic knowledge in the document.

[0009] In combination with the first aspect, in one embodiment of the present application, after the display of at least one knowledge object corresponding to each relevant document, the method also includes: searching for semantic knowledge that has an association relationship with each knowledge object in the association relationship information to obtain retrieval results, wherein the association relationship information includes the correspondence between the knowledge object and the semantic knowledge.

[0010] Therefore, the embodiments of the present application can provide a more fine-grained knowledge organization by searching for semantic knowledge corresponding to the knowledge object in the association relationship information.

[0011] In combination with the first aspect, in one embodiment of the present application, after retrieving at least one relevant document based on the data to be retrieved, the method further includes: clustering the relevant documents through the knowledge objects in the knowledge object information and the association relationship information to obtain a document portrait.

[0012] Therefore, the embodiment of the present application can provide the knowledge structure related to the research problem by determining the correlation relationship portrait of the research problem, thereby providing accurate retrieval of the research problem.

[0013] In combination with the first aspect, in one embodiment of the present application, the data to be retrieved is the question input by the current user; after obtaining the data to be retrieved obtained through the input operation of the current user, the method also includes: obtaining relevant documents corresponding to the question, and generating an answer based on the relevant documents corresponding to the question.

[0014] Therefore, the embodiment of the present application can improve the accuracy of answering questions by extracting answer text.

[0015] In combination with the first aspect, in one embodiment of the present application, before obtaining the data to be retrieved through the input operation of the current user, the method also includes: constructing a multidimensional semantic index, wherein the multidimensional semantic index is composed of basic information of the paper, paper fragment information, abstract language information, knowledge object information and association relationship information.

[0016] In a second aspect, the present application provides a device for obtaining search results, the device comprising:

[0017] A data acquisition module is configured to acquire the data to be retrieved obtained through the input operation of the current user, and retrieve at least one relevant document based on the data to be retrieved;

[0018] A data display module, configured to display at least one knowledge object corresponding to each relevant document, wherein the knowledge object is used to represent a semantic entity in each relevant document;

[0019] The data retrieval module is configured to respond to the target knowledge object selected by the current user from the at least one knowledge object, search for paper fragments in which the target knowledge object co-occurs in the literature, and obtain retrieval results.

[0020] In combination with the second aspect, in one embodiment of the present application, the data acquisition module is configured to: retrieve at least one relevant document corresponding to the data to be retrieved through the basic information of the paper; search for at least one knowledge object of each relevant document in the abstract step information and / or knowledge object information, wherein the abstract step information includes the various components of the document abstract, and the knowledge object information includes the semantic entities contained in the sentences in the relevant documents.

[0021] In combination with the second aspect, in one embodiment of the present application, the data retrieval module is further configured to: search for semantic knowledge that has an association relationship with each knowledge object in the association relationship information to obtain retrieval results, wherein the association relationship information includes the correspondence between the knowledge object and the semantic knowledge.

[0022] In combination with the second aspect, in one embodiment of the present application, the data retrieval module is further configured to: cluster the relevant documents through the knowledge object information and the knowledge objects in the association relationship information to obtain a document portrait.

[0023] In combination with the second aspect, in one embodiment of the present application, the data to be retrieved is the question input by the current user; the data retrieval module is also configured to: obtain relevant documents corresponding to the question, and generate an answer based on the relevant documents corresponding to the question.

[0024] In combination with the second aspect, in one embodiment of the present application, the data acquisition module is further configured to: construct a multidimensional semantic index, wherein the multidimensional semantic index is composed of basic information of the paper, paper fragment information, abstract language information, knowledge object information and association relationship information.

[0025] In a third aspect, the present application provides an electronic device comprising: a processor, a memory and a bus; the processor is connected to the memory via the bus, the memory stores a computer program, and when the computer program is executed by the processor, it can implement the method described in any embodiment of the first aspect.

[0026] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed, can implement the method described in any embodiment of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of a scenario of document retrieval based on multidimensional semantic indexing shown in an embodiment of the present application;

[0028] Figure 2 This is one of the schematic diagrams for establishing a database shown in an embodiment of the present application;

[0029] Figure 3 The second schematic diagram of establishing a database shown in the embodiment of the present application;

[0030] Figure 4 A flowchart of a document retrieval method based on multidimensional semantic indexing is shown in an embodiment of the present application;

[0031] Figure 5 A display interface shown in an embodiment of the present application;

[0032] Figure 6 Another display interface shown in the embodiment of the present application;

[0033] Figure 7 A schematic diagram of the composition of a document retrieval device based on multidimensional semantic indexing shown in an embodiment of the present application;

[0034] Figure 8 A schematic diagram of the composition of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0036] The embodiments of the present application can be applied to scenarios where fine-grained document retrieval is performed on search data input by a user. In order to improve the problems in the background technology, in some embodiments of the present application, information such as paper metadata, fine-grained information contained in the paper, semantic relationships of knowledge objects, etc. are organized into an association network, and the single-dimensional index is changed to realize a document retrieval system with rich semantic information, thereby helping users to quickly discover semantic knowledge of the paper and improving the fine-grainedness of the document retrieval.

[0037] The method steps in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0038] Figure 1 A schematic diagram of a scenario composition of document retrieval based on multidimensional semantic indexing in some embodiments of the present application is provided, and the scenario includes a current user 110, a client 120, and a server 130. Specifically, the current user 110 performs an input operation on the client 120 to input retrieval data, and then the client 120 sends the retrieval data to the server 130, and the server 130 performs a file retrieval operation according to the retrieval data to obtain a retrieval result, and finally, the server 130 returns the retrieval result to the client 120 for display.

[0039] The following uses a server as an example to illustrate a method for document retrieval based on multidimensional semantic indexing provided by some embodiments of the present application. It can be understood that the execution subject of the method for document retrieval based on multidimensional semantic indexing in the present application can be any electronic device that can execute the method. The above-mentioned server is only an example and the present application does not limit this.

[0040] With the development of semantic technologies such as text mining and information retrieval, the application scenarios of paper list retrieval in document retrieval systems have gradually expanded to semantic retrieval scenarios of fine-grained semantic knowledge disclosure and knowledge question-and-answer dialogue, showing users fine-grained information such as research questions and research methods in papers, and providing question-and-answer dialogue functions based on semantically related content texts. However, the existing paper basic information indexing system does not contain fine-grained semantic knowledge and cannot effectively support semantic retrieval applications.

[0041] In response to the above shortcomings in the relevant technology, this application constructs a multidimensional semantic index consisting of a basic paper information index, a paper fragment information index, an abstract language information index, a knowledge object information index and an association relationship information index, and organizes the paper metadata, the fine-grained information at the chapter, paragraph, sentence and phrase levels contained in the paper, the semantic relationship between knowledge objects, and the co-occurrence relationship information between knowledge objects and papers into an association network to support fine-grained semantic knowledge discovery in literature retrieval, thereby expanding the application scenarios of paper list retrieval, better revealing semantic knowledge to users, and providing a more fine-grained and more targeted literature retrieval system.

[0042] It can be understood that the documents in this application can be papers, patent documents, etc., and the basic information of papers and paper fragment information mentioned in this application can be expanded to basic information of patents and patent fragment information, that is, this application can be applied to all document searches, not limited to the search of papers.

[0043] In one embodiment of the present application, before performing a search operation on user input, a database (ie, a multi-dimensional semantic index) capable of supporting fine-grained document search is established.

[0044] First, a semantic knowledge organization model is constructed. That is, each document in the database is split, extracted, and associated to obtain fine-grained knowledge information of each document.

[0045] Specifically, taking a paper as an example, the basic information of the paper is split to obtain the title, author, abstract, keywords, subject, source information, full text in HTML format, pdf file link, etc. The chapters, paragraphs and sentences in the paper are structured to obtain the paper fragment information. The abstract part of the basic information of the paper is extracted and refined to obtain the sentence-level abstract language step information involved in the six research processes: research background, research purpose, research question, research conclusion, research result and research method.

[0046] The knowledge objects included in the above-mentioned paper fragment information are extracted to obtain knowledge object information. It can be understood that knowledge objects are used to represent semantic entities in documents. For example, knowledge objects can be research objectives, research questions and research methods in abstracts, or data, instruments and equipment, theoretical principles, etc. in paper fragments. Knowledge object information can include general knowledge objects and domain knowledge objects. General knowledge objects are knowledge objects applicable to all fields, such as research questions, research methods, etc. Domain knowledge objects are knowledge objects in the field in the current document. For example, the field of the current document is the agricultural field, and the corresponding domain knowledge objects are growth environment, agronomic operations, etc. The field of the current document is the field of science and technology policy, and the corresponding domain knowledge objects are policy themes, policy scope of application, etc.

[0047] Furthermore, as a specific embodiment of the present application, the general knowledge objects in the knowledge object information include nine categories: concept definitions, research problems, research methods, instruments and equipment, theoretical principles, software systems, measurement indicators, data, scientists' names and place names, and the field knowledge objects are classified into eight categories: science and technology policy, agriculture, resources and environment, computers, chemistry, medicine, industry, and physics.

[0048] The knowledge objects in the field of science and technology policies include policy themes, policy scope of application, existing issues, policy objectives, policy basis, policy guiding ideology, policy issuing agencies, policy issuing time, policy tools, policy implementers, policy audiences and replaced policies.

[0049] The knowledge objects in the agricultural field include growing environment, crop quality, agronomic operations, pests and diseases, new varieties, agricultural inputs, agricultural disasters and emergency response.

[0050] The knowledge objects in the field of resources and environment include natural resources, natural ecological risks, sudden environmental events, pollution sources, natural disasters and safety accidents.

[0051] Computer field knowledge objects include database systems, computer networks, software engineering, algorithms and data structures, operating systems and artificial intelligence.

[0052] Knowledge objects in the field of chemistry include elements, molecules, ions, chemical reactions, chemical bonds and catalysts.

[0053] Medical domain knowledge objects include diseases, tests, symptoms, drugs, viruses, genes, amino acids, proteins, and body parts.

[0054] Industrial domain knowledge objects include fault location, fault characteristics, fault causes, treatment measures, aircraft, aircraft manufacturers, engines, engine components and engine sub-components.

[0055] The knowledge objects in the field of physics include measurement, instrumentation and data analysis, models, method theory, mathematics, as well as phenomena and laws.

[0056] After extracting the knowledge object information of each paper, the association relationship information is constructed to enable the knowledge object, paper and paper fragment to be associated with each other. Specifically, the association relationship information includes the paper fragment co-occurrence relationship, the knowledge object fragment co-occurrence relationship and the knowledge object semantic relationship.

[0057] Among them, the paper fragment co-occurrence relationship is used to characterize the co-occurrence of papers and paper fragments. For example, when a user inputs paper fragment A, it can be displayed in which paper the paper fragment A appears.

[0058] The knowledge object-fragment co-occurrence relationship is used to characterize the co-occurrence of knowledge objects and paper fragments. For example, the knowledge object of machine learning appears in paper fragment B. In actual operation, the page displays the current paper and multiple knowledge objects in the paper. When the user clicks on machine learning among the multiple knowledge objects, the page jumps to paper fragment B where machine learning appears in the current paper.

[0059] The semantic relationship of knowledge objects is used to characterize the relationship between knowledge objects. For example, the relationship between research problems and research methods is a solution relationship (that is, research methods solve research problems, research problems and research methods are knowledge objects, and solution is the semantic relationship between them), the relationship between research methods and instruments and equipment is a use relationship, the relationship between research methods and theoretical principles is a basis relationship, the relationship between research methods and measurement indicators is a standard relationship, and the relationship between research methods and data is a reference relationship.

[0060] Then, a multidimensional semantic index is constructed. In other words, a database supporting fine-grained document retrieval is established through the fine-grained knowledge information of the above-mentioned documents.

[0061] That is to say, based on the knowledge organization in the previous step, a multidimensional semantic index is constructed. The multidimensional semantic index consists of basic information of the paper, information of paper fragments, information of abstract steps, information of knowledge objects and information of association relationships, including fine-grained information at the chapter, paragraph, sentence and phrase levels, as well as the co-occurrence relationship and semantic relationship between them.

[0062] The relationship between the various parts of the multidimensional semantic index is as follows Figure 2 As shown, the basic information of the paper, the information of the paper fragments, the information of the abstract steps, the information of the knowledge objects and the information of the association relationship are associated with each other through the paper ID number.

[0063] It should be noted that the abstract step information is the research step sentence contained in the abstract part of the paper basic information index. The knowledge object information is the semantic knowledge entity contained in the sentence in the paper fragment information index. The association relationship information is composed of the semantic relationship between knowledge objects, the co-occurrence relationship between paper fragments, and the co-occurrence relationship between knowledge objects and paper fragments. The paper fragment information includes the implication relationship between chapters, paragraphs, and sentences.

[0064] After the database is built, the document retrieval system can perform document retrieval operations based on user input. The following describes the steps of using the multidimensional semantic index built above to perform document retrieval.

[0065] like Figure 3 As shown, the database (i.e., multidimensional semantic index) provided by this application supports the realization of eight main functions: user intent understanding, heuristic knowledge navigation, paper list retrieval, semantic knowledge disclosure, knowledge question and answer dialogue, full-text highlight disclosure, problem technology portrait, and semantic content recommendation. Specifically, in the process of obtaining retrieval results provided by this application, the results in three forms of document list, semantic knowledge, and question and answer dialogue are displayed to users through document retrieval, knowledge retrieval, and question and answer dialogue engines, respectively. Then, when the user clicks on a specific research question or paper, the problem technology portrait, full-text highlight disclosure, and semantic content recommendation results are presented.

[0066] That is to say, the database stores basic information of the paper, information of the paper fragment, abstract step information, knowledge object information and association relationship information. The abstract step information is obtained by extracting the information of the abstract in the basic information of the paper, and the chapters, paragraphs and sentences of the paper are extracted to obtain the information of the paper fragment, wherein the information about the abstract in the chapter is stored in the abstract step information. The knowledge objects related to the paper fragment information and the abstract step information are extracted to obtain the knowledge object information, and in the process of extracting the knowledge object information, all the knowledge objects of the current document are classified according to the general knowledge objects and the domain knowledge objects. The semantics with association relationship in the basic information of the paper, the paper fragment information, the abstract step information and the knowledge object information are associated to obtain the association relationship information, wherein the knowledge objects with association relationship are associated to obtain the knowledge object semantic relationship. The co-occurring papers and paper fragments are extracted and associated to obtain the paper fragment co-occurrence relationship (i.e. the paper and paper fragment co-occurrence relationship). The knowledge objects appearing in the paper fragment are associated (i.e. the co-occurring paper fragments and knowledge objects are associated to obtain the knowledge object and paper fragment co-occurrence relationship (i.e. the object fragment co-occurrence relationship).

[0067] like Figure 4As shown, the implementation steps of a document retrieval method based on multidimensional semantic indexing provided by the present application (i.e., the implementation steps of using the above multidimensional semantic indexing to support the retrieval operation and obtain the retrieval results) include:

[0068] S410, obtaining the data to be searched obtained through the input operation of the current user, and searching for at least one relevant document based on the data to be searched.

[0069] That is to say, after the current user inputs the data to be retrieved, the user intention understanding function is first used to automatically understand the intentions of names and qualifiers, subject terms, subject terms and names, etc. according to the different data to be retrieved input by the current user. Specifically, when the current user inputs the data to be retrieved, the basic information of the paper and the knowledge object information are matched to identify the fine-grained information in the data to be retrieved, and then the current user's query intention is accurately judged.

[0070] For example, the data to be retrieved input by the user is "Li Yi 1947 Journal of Edible Fungi Self-driven vehicle", and after identifying the user's intention, it is obtained that the type of the data to be retrieved is "name and qualifier", and the type of entity is subject word, name, time and journal.

[0071] After understanding the user's intention for the data to be retrieved, the heuristic knowledge navigation is entered based on the user's intention recognition results. Specifically, the relevant fine-grained semantic knowledge is presented in the drop-down box according to the user's input information to help users navigate and discover important semantic knowledge. When the user inputs, the fine-grained information contained in the input is obtained by matching the basic information of the paper and the knowledge object information, and then the recommended fine-grained navigation content is obtained by matching the association relationship information.

[0072] For example, the title of the paper is matched from the current user's query intention, and then the knowledge object in the title is matched in the knowledge object information to include the research method. The research method is then matched in the association information to obtain research questions, instruments, theoretical principles, measurement indicators and data that have related information to the research method. Finally, the above research questions, instruments, theoretical principles, measurement indicators and data are displayed in the navigation drop-down box to obtain fine-grained navigation content.

[0073] After displaying the drop-down box of the heuristic knowledge navigation, the search data clicked by the user in the drop-down box is obtained, and then the search data originally input by the user and the search data selected by the user in the drop-down box (hereinafter referred to as the selected search data) are rewritten and converted. Specifically, the data in the form of sentences in the data to be searched and the selected search data are converted into structured search formulas, and the data in the form of words in the data to be searched and the selected search data are expanded into data in the form of sentences, that is, any input will have corresponding information in both structured and sentence forms.

[0074] After the user input is converted and expanded through the above methods, the search data is retrieved. Specific search functions include: paper list search, semantic knowledge disclosure, knowledge question and answer dialogue, full text highlight disclosure, problem technology portrait and semantic content recommendation. The implementation methods of these functions will be described in detail below.

[0075] In one embodiment of the present application, in a knowledge question-answering dialogue scenario, when the data to be retrieved is a question input by the current user, relevant documents corresponding to the question are obtained, and an answer is generated based on the relevant documents corresponding to the question. Specifically, the answer text corresponding to the question input by the current user can be obtained from the title, abstract in the basic information of the paper, and the sentences, paragraphs, and chapters in the paper fragment information.

[0076] That is to say, the question entered by the current user in the dialog box is obtained, and the answer to the question raised by the user is returned. The answer comes from the factual content in the paper abstract and chapter, or the retrieved content is generated and processed. Specifically, for the question raised by the user, the knowledge question and answer dialogue engine obtains the relevant text from the title, abstract in the basic information of the paper, and the sentence, paragraph and chapter fields in the paper fragment information, and directly returns the answer or obtains the answer through generation processing.

[0077] It is understandable that the generation process may be to concatenate or rewrite the searched sentences to generate a more readable answer.

[0078] In one embodiment of the present application, in the scenario of paper list retrieval, a matching paper list is returned based on the data to be retrieved, such as title, abstract content, etc., input and selected by the user. Specifically, the paper list retrieval engine matches the retrieval data with the fields in the basic information of the paper, and then displays them in a list form.

[0079] It should be noted that, while returning the document list, semantic knowledge and semantic content recommendations can also be returned at the same time.

[0080] In one embodiment of the present application, in the scenario where the full-text highlights are revealed, first, at least one relevant document corresponding to the data to be retrieved is searched, and then, in response to the target document selected by the current user from at least one relevant document, the basic information, fragment information, abstract step information, knowledge object information and association relationship information involved in the target document are searched to obtain the search results.

[0081] Specifically, Figure 4 As shown, S420 may be executed first to display at least one knowledge object corresponding to each relevant document, and then S430 may be executed to search for co-occurring paper fragments of the target knowledge object in the document in response to a target knowledge object selected by the current user from at least one knowledge object, to obtain search results.

[0082] That is to say, searching the search data entered by the current user not only returns the list of retrieved documents, but also returns the fine-grained knowledge in each document. Furthermore, on the details page of each document, by matching the basic information of the paper, the language information of the abstract, the information of the paper fragment and the information of the knowledge object, the important information in the document is revealed to the current user, for example, revealing at least one knowledge object in the document and the paper fragment where each knowledge object is located.

[0083] Specifically, after finding at least one document corresponding to the user input through the basic information of the paper, the abstract information of each document is obtained in the abstract step information through the document ID, and the paper fragment of each document is obtained in the paper fragment information, and then the abstract information and at least one knowledge object included in the paper fragment, as well as the paper fragments where each knowledge object appears (i.e., the co-occurrence fragments of each knowledge object) are searched in the knowledge object information, and at least one knowledge object and the corresponding document are displayed in the document list. After the current user sees at least one knowledge object, he selects the target knowledge object from the at least one knowledge object, and then the server searches the knowledge object information for the paper fragment that co-occurs with the target knowledge object, and then returns it to the client for display, so that the current user can clearly understand the original paper fragment corresponding to the target knowledge object.

[0084] For example, Figure 5 As shown, at least one paper corresponding to the user input is displayed, wherein the title and abstract of the paper can be displayed, and at least one knowledge object related to the current paper (i.e. Figure 5 Then, when the user puts the mouse on the research question, the specific content of the research question will appear, including machine learning, large language models and natural language processing. If the user clicks on machine learning, the paper snippets that co-occur with machine learning in the current paper will be displayed.

[0085] In one embodiment of the present application, in a scenario of semantic knowledge disclosure, at least one knowledge object corresponding to the data to be retrieved is first searched, and then semantic knowledge that has an association relationship with each knowledge object is searched in the association relationship information to obtain a search result.

[0086] That is to say, when searching the search data currently input by the user, in addition to returning the retrieved literature list, it also returns the fine-grained semantic knowledge related to the current search data. Furthermore, when the user enters a subject word, by searching for knowledge objects such as research method phrases, research question phrases, and software system words in the knowledge object information, combined with the semantic related information in the association relationship information, the fine-grained knowledge related to the subject word is displayed.

[0087] Specifically, the association relationship information stores the association relationships of all knowledge objects. For example, if the knowledge object is an organic solar cell, then the association relationship information stores knowledge objects of different semantic relationship types related to organic solar energy (for example, the basic principles of organic solar cells, the photoelectric conversion process, and the light absorption mechanism all have a core concept relationship with the knowledge object of organic solar cells). After obtaining the search data input by the user, the knowledge object in the search data is searched in the knowledge object information, and then the semantic knowledge that has an association relationship with the knowledge object is searched in the association relationship information, and the semantic knowledge is displayed in the interface.

[0088] For example, Figure 6 As shown, the query statement input by the user is "organic solar cell", and the associated relationships of "organic solar cell" found in the associated relationship information are: core concept relationship, performance indicator relationship, material and device relationship, and spectral characteristic relationship, and the specific content with the above relationships is returned and displayed in the interface, so that the user can select the content of interest. For example, if the user selects "The basic principle of A solar cell", the page can jump to the page showing the specific content of "The basic principle of A solar cell".

[0089] In one embodiment of the present application, in the scenario of problem technology portrait, the specific steps of determining the association relationship portrait include: clustering the relevant documents through the knowledge object information and the knowledge objects in the association relationship information to obtain the document portrait.

[0090] That is to say, when searching the search data entered by the current user, in addition to returning the retrieved literature list, it also returns the clustering results of research questions, research methods and research progress. Furthermore, when the user clicks on a research question, by searching the knowledge object information and the association relationship information, the research question, research method, and research progress association relationship portrait results are displayed.

[0091] Specifically, the knowledge objects included in the question input by the user are matched in the knowledge object information. For example, the question input by the user includes a research question and a scientist's name. Then, the content that has a semantic relationship and a co-occurrence relationship with the research question and the scientist's name is searched in the association relationship information and displayed. For example, the knowledge objects that have a semantic relationship with the research question include: research methods and research progress, among which the research method has a semantic relationship with other knowledge objects. Their relationships are connected to obtain an association relationship portrait result, and important literature related to the scientist's name and other core scholars can also be found.

[0092] In one embodiment of the present application, in the scenario of semantic content recommendation, the present application makes recommendations based on fine-grained semantic content. By calculating the similarity between fine-grained semantic information such as research question words and research method words in knowledge object information, the semantic content recommendation results are displayed to users.

[0093] For example, if the user inputs organic solar cells, related recommended questions such as "Design of efficient organic solar cells" and "New strategy mechanism to improve photoelectric conversion efficiency" can also be returned.

[0094] Therefore, based on the basic information of the paper, this application extracts the chapters, paragraphs, sentences, phrase-level fragments, language steps, knowledge object information, and the semantic association relationship information between them in the paper, and constructs a multidimensional semantic index composed of the basic information of the paper, paper fragment information, abstract language step information, knowledge object information and association relationship information. The paper metadata, the fine-grained information contained in the paper, the semantic relationship of knowledge objects and other information are organized into an association network, changing the single-dimensional index, thereby supporting semantic functions such as heuristic knowledge navigation and semantic knowledge revelation.

[0095] In other words, this application proposes a multi-dimensional semantic indexing method for supporting paper knowledge discovery. Different from a single paper basic information index, this method organizes a knowledge network by constructing finer-grained information. Furthermore, the basic information of the paper, the information of the paper fragments, and the information of the knowledge object are used to support the user intention understanding, heuristic knowledge navigation, paper list retrieval, knowledge question and answer dialogue, and semantic content recommendation functions of the paper knowledge discovery system. The semantic knowledge disclosure, full text highlight disclosure, and problem technology portrait functions are realized by combining the abstract step information and the association relationship information, thus expanding the traditional document list retrieval effect to the application scenarios of semantic knowledge disclosure and dialogue question and answer assistant.

[0096] Therefore, the key technical points of this application are: First, effectively organize and store the fragmented knowledge of chapters, paragraphs, sentences and words in the document, as well as the co-occurrence and semantic relationships between them. Second, construct a multidimensional semantic index, which consists of basic information of the paper, information of paper fragments, abstract language information, knowledge object information and association relationship information, and clarifies the specific content of each type of information and the association relationship between them. Third, explain the mode of multidimensional semantic index supporting the application of paper knowledge discovery system with semantic function, and explain that the specific information in the index supports the realization of various semantic functions in the document knowledge discovery system.

[0097] The above describes an implementation of a method for document retrieval based on multidimensional semantic indexing. The following will describe an apparatus for document retrieval based on multidimensional semantic indexing.

[0098] like Figure 7 As shown, some embodiments of the present application provide a device 700 for document retrieval based on multi-dimensional semantic indexing, and the device includes: a data acquisition module 710, a data display module 720 and a data retrieval module 730.

[0099] The data acquisition module 710 is configured to acquire the data to be retrieved obtained through the input operation of the current user, and retrieve at least one relevant document based on the data to be retrieved;

[0100] A data display module 720 is configured to display at least one knowledge object corresponding to each relevant document, wherein the knowledge object is used to represent a semantic entity in each relevant document;

[0101] The data retrieval module 730 is configured to, in response to a target knowledge object selected by the current user from among the at least one knowledge object, search for paper fragments in which the target knowledge object co-occurs in the literature to obtain a retrieval result.

[0102] In one embodiment of the present application, the data acquisition module 710 is further configured to: retrieve at least one relevant document corresponding to the data to be retrieved through the basic information of the paper; search for at least one knowledge object of each relevant document in the abstract step information and / or knowledge object information, wherein the abstract step information includes the various components of the document abstract, and the knowledge object information includes the semantic entities contained in the sentences in the relevant documents.

[0103] In one embodiment of the present application, the data retrieval module 730 is further configured to: search for semantic knowledge that has an association relationship with each knowledge object in the association relationship information to obtain retrieval results, wherein the association relationship information includes the correspondence between the knowledge object and the semantic knowledge.

[0104] In one embodiment of the present application, the target knowledge object is a research question; the data retrieval module 730 is also configured to: cluster the relevant documents through the knowledge object information and the knowledge objects in the association relationship information to obtain a document portrait.

[0105] In one embodiment of the present application, the data to be retrieved is a question input by the current user; the data retrieval module 730 is further configured to: obtain relevant documents corresponding to the question, and generate an answer based on the relevant documents corresponding to the question.

[0106] In one embodiment of the present application, the data acquisition module 710 is further configured to: construct a multidimensional semantic index, wherein the multidimensional semantic index is composed of basic information of the paper, paper fragment information, abstract language information, knowledge object information and association relationship information.

[0107] In the embodiments of the present application, Figure 7 The modules shown can achieve Figures 1 to 6 Each process in the method embodiment. Figure 7 The operations and / or functions of each module in Figures 1 to 6 For details, please refer to the description in the above method embodiment. To avoid repetition, detailed description is appropriately omitted here.

[0108] like Figure 8 As shown, an embodiment of the present application provides an electronic device 800, including: a processor 810, a memory 820 and a bus 830, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement a method as described in any one of the above embodiments. For details, please refer to the description in the above method embodiments. To avoid repetition, the detailed description is appropriately omitted here.

[0109] Among them, the bus is used to realize the direct connection and communication of these components. Among them, the processor in the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0110] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method described in the above embodiment may be executed.

[0111] Understandably, Figure 8 The structure shown is for illustration only and may also include Figure 8 More or fewer components as shown, or with Figure 8 Different configurations are shown. Figure 8 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0112] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a server, any method described in any of the above-mentioned embodiments is implemented. For details, please refer to the description in the above-mentioned method embodiments. To avoid repetition, the detailed description is appropriately omitted here.

[0113] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0114] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A document retrieval method based on multidimensional semantic indexing, characterized in that: The method comprises: Acquire the data to be retrieved obtained through the input operation of the current user, and retrieve at least one relevant document based on the data to be retrieved, and return the clustering results of research questions, research methods and research progress, semantic knowledge, semantic content recommendations and answers to the questions input by the current user; wherein, the data to be retrieved is obtained by converting the original input data of the current user and the data in the form of sentences in the selected retrieval data of the drop-down box into structured retrieval formulas, and expanding the data in the form of words in the original input data and the selected retrieval data into data in the form of sentences; the selected retrieval data is obtained by understanding and identifying the user intention of the data to be retrieved, obtaining the user intention recognition result of the type of the data to be retrieved and the type of the entity, and selecting from the recommended fine-grained navigation content after entering the heuristic knowledge navigation through the user intention recognition result; Displaying at least one knowledge object corresponding to each relevant document, wherein the knowledge object is used to represent a semantic entity in each relevant document; the knowledge object is classified into general knowledge objects and domain knowledge objects; In response to the target knowledge object selected by the current user from the at least one knowledge object, searching for paper fragments that co-occur in the target knowledge object in the document to obtain a search result; wherein the paper fragment is to obtain the abstract information of each document in the abstract step information through the document ID, obtain the paper fragment of each document in the paper fragment information, and then search the knowledge object information for at least one knowledge object included in the abstract information and the paper fragment, as well as the paper fragment in which each knowledge object appears; Before obtaining the data to be retrieved obtained through the input operation of the current user, the method further includes: obtaining fine-grained knowledge information of each document by splitting, extracting and associating each document in the database, so as to realize the construction of a semantic knowledge organization model; constructing a multidimensional semantic index, wherein the multidimensional semantic index is composed of basic information of the paper, information of the paper fragment, information of the abstract step, knowledge object information and association relationship information; The clustering result is obtained by clustering the relevant documents through the knowledge object information and the knowledge objects in the association relationship information; the knowledge object is obtained by matching the data to be retrieved with the knowledge object information; the semantic content recommendation is obtained by calculating the similarity between the fine-grained semantic information of the research question words and the research method words in the knowledge object information; The method further comprises: When the data to be retrieved is a question input by the current user, an answer corresponding to the at least one relevant document is generated; the answer is the factual content in the abstract or chapter of the at least one relevant document, or is generated based on the content in the at least one relevant document after processing, and the types of processing include: splicing or rewriting.

2. The method according to claim 1, characterized in that The retrieving at least one relevant document based on the data to be retrieved comprises: Retrieve at least one relevant document corresponding to the data to be retrieved through the basic information of the paper; Before presenting at least one knowledge object corresponding to each relevant document, the method further includes: At least one knowledge object of each of the relevant documents is searched in the abstract step information and / or knowledge object information, wherein the abstract step information includes the components of the document abstract, and the knowledge object information includes the semantic entities contained in the sentences in each of the relevant documents.

3. The method according to claim 1 or 2, characterized in that: After presenting at least one knowledge object corresponding to each relevant document, the method further includes: The semantic knowledge that is associated with each knowledge object is searched in the associated relationship information to obtain a search result, wherein the associated relationship information includes a corresponding relationship between the knowledge object and the semantic knowledge.

4. The method according to claim 1 or 2, characterized in that: After retrieving at least one relevant document based on the data to be retrieved, the method further includes: The relevant documents are clustered through the knowledge object information and the knowledge objects in the association relationship information to obtain a document portrait.

5. The method according to claim 1 or 2, characterized in that: The data to be retrieved is the question input by the current user; After obtaining the data to be retrieved through the input operation of the current user, the method further includes: Relevant documents corresponding to the question are obtained, and an answer is generated based on the relevant documents corresponding to the question.

6. A document retrieval device based on multidimensional semantic indexing, characterized in that: The device is used to perform the method according to claim 1, comprising: A data acquisition module is configured to acquire the data to be retrieved obtained through the input operation of the current user, and retrieve at least one relevant document based on the data to be retrieved; A data display module, configured to display at least one knowledge object corresponding to each relevant document, wherein the knowledge object is used to represent a semantic entity in each relevant document; The data retrieval module is configured to respond to the target knowledge object selected by the current user from the at least one knowledge object, search for paper fragments in which the target knowledge object co-occurs in the literature, and obtain retrieval results.

7. The device according to claim 6, characterized in that The data acquisition module is configured as follows: Retrieve at least one relevant document corresponding to the data to be retrieved through the basic information of the paper; At least one knowledge object of each of the relevant documents is searched in the abstract step information and / or knowledge object information, wherein the abstract step information includes the components of the document abstract, and the knowledge object information includes the semantic entities contained in the sentences in each of the relevant documents.

8. An electronic device, characterized in that: include: processor, memory, and bus; The processor is connected to the memory via the bus, the memory stores a computer program, and the computer program, when executed by the processor, can implement the method as claimed in any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed, can implement the method as claimed in any one of claims 1 to 5.

Images