Knowledge graph construction, retrieval method and device

Abstract

The present invention relates to the field of computer technology, and provides a knowledge graph construction, retrieval method and device, wherein the knowledge graph construction method includes: obtaining literature data; classifying the literature data to obtain the document type of the literature data, and extracting the associated entity pairs contained in the literature data, the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the document types include genetics, non-clinical and clinical; using the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the document types as the attributes of the edges to construct a knowledge graph. The knowledge graph construction, retrieval method and device provided by the present invention can quickly and effectively obtain the relationships between the three types of biological entities of drugs, targets and diseases and the credibility of entity pairs, thereby helping users to explore potential drug research and development information, and then assist in new drug research and development.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a knowledge graph construction and retrieval method and device. Background Art

[0002] In the medical field, the publicly available knowledge graphs mainly provide information on biological entities and their relationships. If the knowledge graph is constructed using industry consensus, although the credibility is high, there are problems such as small data volume and incomplete coverage. If the knowledge graph is constructed using large medical literature libraries such as PubMed, although the data volume is large, the information on some entities is not credible.

[0003] At the same time, when researchers are developing new drugs, they hope to discover biological entities that have certain evidence support but for which competition for related drug development is not yet fierce, especially biological entities that have recently made major research progress that has not yet received widespread attention.

[0004] Therefore, providing a knowledge graph with large data volume, which can comprehensively cover the relationships among various biological entities and provide reliable information among various entity pairs, is an urgent problem to be solved. Summary of the invention

[0005] The present invention provides a knowledge graph construction and retrieval method and device to solve the defects of the prior art in providing knowledge graphs of biological entities and their relationships, such as small amount of data, incomplete coverage, and unreliable information on some entities.

[0006] The present invention provides a knowledge graph construction method, comprising:

[0007] Obtain literature data;

[0008] Classifying the literature data to obtain the literature type of the literature data, and extracting the associated entity pairs contained in the literature data, wherein the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the literature types include genetics, non-clinical and clinical;

[0009] A knowledge graph is constructed by taking the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the document type as attributes of the edges.

[0010] According to the knowledge graph construction method provided by the present invention, the credibility includes the evidence level and / or the evidence index, and the step of obtaining the credibility of the associated entity under the document type includes:

[0011] Determining the level of evidence of the associated entity pair under the document type based on the number of citations of the associated entity pair's document data and the document type of the associated document data; and / or,

[0012] Based on the number of citations of each document data in the document collection, the evidence index of the same associated entity pair under the document type corresponding to the document collection is determined. The document types of each document data in the document collection are the same, and each document data contains the same associated entity pair.

[0013] According to the knowledge graph construction method provided by the present invention, the method of determining the evidence level of the associated entity pair under the document type based on the number of citations of the associated entity pair's document data and the document type of the associated document data includes:

[0014] Determining the evidence level of the associated entity pair under the associated document data based on the number of citations of the associated entity pair's associated document data;

[0015] Based on the document type of the belonging document data, the evidence levels of the same associated entity pairs with the same document type are aggregated, and the evidence level of the same associated entity pair under the document type is determined based on the aggregation result.

[0016] According to the knowledge graph construction method provided by the present invention, the method of determining the evidence index of the same associated entity under the corresponding document type of the document collection based on the number of citations of each document data in the document collection includes:

[0017] Based on the number of citations of each document data in the document collection, the document data are sorted in descending order according to the number of citations to obtain the arrangement sequence number of each document data;

[0018] Based on the arrangement sequence number and the number of citations of each document data, the evidence index of the same associated entity pair under the corresponding document type of the document collection is determined.

[0019] According to the knowledge graph construction method provided by the present invention, the construction of the knowledge graph further includes:

[0020] Determine a comprehensive score of the level of evidence of the associated entity pair based on the level of evidence of the associated entity pair under each document type and the weight of the level of evidence under each document type;

[0021] The color of the knowledge graph edge is constructed based on the comprehensive score of the evidence level of the associated entity pairs.

[0022] According to the knowledge graph construction method provided by the present invention, the document data is classified to obtain the document type of the document data, including:

[0023] Based on the first text classification model, the document data is classified according to importance to obtain a target document;

[0024] Based on the second text classification model, the target document is classified to obtain the document type of the target document.

[0025] According to the knowledge graph construction method provided by the present invention, the extracting the associated entity pairs contained in the document data includes:

[0026] Extracting pronouns and entities of a preset type from the document data to obtain pronouns and entities contained in the document data;

[0027] Based on the relationship between the pronoun and the entity, performing entity recognition on the pronoun, and replacing the pronoun with the entity based on the recognition result;

[0028] Relationships between entities of the same type and relationships between entities of different types are identified respectively to obtain associated entity pairs contained in the document data.

[0029] The present invention also provides a retrieval method, comprising:

[0030] receiving a search keyword sent by a user terminal, wherein the search keyword includes at least one of an entity name, an associated entity pair, a document type, and a credibility of the associated entity pair under any document type;

[0031] Determine target document data matching the search keyword from a pre-built knowledge graph, and return the target document data to the user terminal;

[0032] The knowledge graph is determined based on the knowledge graph construction method.

[0033] The present invention also provides a knowledge graph construction device, comprising:

[0034] A data acquisition unit, used for acquiring literature data;

[0035] an entity pair extraction unit, configured to classify the document data, obtain the document type of the document data, and extract associated entity pairs contained in the document data, wherein the entities in the associated entity pairs include at least one of drugs, targets, and diseases, and the document types include genetics, non-clinical, and clinical;

[0036] The graph construction unit is used to construct a knowledge graph using the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the document types as attributes of the edges.

[0037] The present invention also provides a retrieval device, comprising:

[0038] A keyword receiving unit, configured to receive a search keyword sent by a user terminal, wherein the search keyword includes at least one of an entity name, an associated entity pair, a document type, and a credibility of an associated entity pair under any document type;

[0039] A data determination unit is used to determine target document data matching the search keyword from a pre-constructed knowledge graph and return the target document data to the user terminal; the knowledge graph is determined based on the knowledge graph construction method.

[0040] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the knowledge graph construction methods or retrieval methods described above.

[0041] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the knowledge graph construction methods or retrieval methods described above.

[0042] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the knowledge graph construction methods or retrieval methods described above.

[0043] The knowledge graph construction and retrieval method and device provided by the present invention classify the acquired literature data and extract the associated entity pairs contained in the literature data. On this basis, the knowledge graph is constructed with the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the literature type as the attributes of the edges. It is possible to extract associated entity pairs from a large amount of literature data, and construct a knowledge graph based on the credibility of the associated entity pairs under each literature type. Users can quickly and effectively obtain the relationships between the three types of biological entities, namely drugs, targets and diseases, and the credibility of the entity pairs, thereby helping users to explore potential drug development information and further assist in the development of new drugs. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 It is a schematic diagram of the process of the knowledge graph construction method provided by the present invention;

[0046] Figure 2It is a flow chart of the credibility acquisition method provided by the present invention;

[0047] Figure 3 It is a schematic diagram of the flow of the retrieval method provided by the present invention;

[0048] Figure 4 It is a schematic diagram of the structure of the knowledge graph construction device provided by the present invention;

[0049] Figure 5 It is a schematic diagram of the structure of the retrieval device provided by the present invention;

[0050] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Medical researchers need to find and evaluate new research directions from a vast amount of literature, and are particularly interested in the associations between three types of biological entities: "drugs", "targets" (including but not limited to genes and the proteins they encode), and "diseases" (hereinafter referred to as "entity pairs"). If an entity pair is supported by a large number of high-level literature, it means that the research on the entity pair is relatively popular or mature, and the competition for related drug development may be fierce, so it may not be easy to produce new and competitive research results; if an entity pair is only supported by low-level literature, it means that the research on the entity pair is not yet mature, and research on this technology may not be successful. Therefore, researchers hope to find entity pairs that have certain evidence support, but the competition for related drug development is not yet fierce, especially entity pairs that have made major research progress in the recent period that has not yet received widespread attention.

[0053] The publicly available knowledge graphs mainly provide information on biological entities and their relationships. If the knowledge graph is constructed using industry consensus, although the credibility is high, there are problems such as small data volume and incomplete coverage. If the knowledge graph is constructed using large medical literature libraries such as PubMed, although the data volume is large, the information on some entities is not credible.

[0054] At the same time, researchers also expect to understand the ratings of entity pairs on key dimensions of evidence. For example, researchers may expect to find entity pairs with high levels of genetic evidence, high levels of nonclinical evidence, but low levels of clinical evidence, so as to conduct further clinical research and obtain high-level clinical evidence.

[0055] In view of the above problems, the inventive concept of the present invention is to classify the acquired literature data and extract the associated entity pairs contained in the literature data. On this basis, a knowledge graph is constructed with the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the literature type as the attributes of the edges. It is possible to extract associated entity pairs from a large amount of literature data and construct a knowledge graph based on the credibility of the associated entity pairs under each literature type. Users can quickly and effectively obtain the relationships between the three types of biological entities, namely drugs, targets and diseases, and the credibility of the entity pairs, thereby helping users to explore potential drug research and development information and further assist in the development of new drugs.

[0056] Based on the above-mentioned inventive concept, the present invention provides a knowledge graph construction, retrieval method and device, which are applied to the knowledge graph construction and retrieval scenarios in the new generation of information technology, so as to realize the construction of a structured knowledge graph for disordered literature data, and obtain the relationship between three types of biological entities: drugs, targets and diseases, and the credibility of entity pairs.

[0057] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings. Figure 1 It is a flow chart of the knowledge graph construction method provided by the present invention. The executor of each step in the method can be a knowledge graph construction device, which can be implemented by software and / or hardware. The device can be integrated in an electronic device. The electronic device can be a terminal device (such as a smart phone, a personal computer, a wearable device, etc.), or a server (such as a local server or a cloud server, or a server cluster, etc.), or a processor, or a chip, etc.

[0058] like Figure 1 As shown, the method may include the following steps:

[0059] Step 110, obtaining document data.

[0060] Specifically, the literature data can be the source text from which various biological entities and their relationships can be mined and the knowledge graph can be constructed accordingly. The literature data can be part of the text or the entire text of the paper that describes the relationship between the three types of biological entities that focus on targets, drugs and diseases. The literature data can be obtained by crawling from relevant websites (such as PubMed database) using a web crawler, or by taking an image or scanning a paper version of the literature data, etc., and the embodiments of the present invention do not specifically limit this.

[0061] The acquisition of literature data can be done at a scheduled time or by real-time monitoring of various literature data publishing websites, and timely acquisition of new literature data after new literature data appears.

[0062] It should be noted that in order to improve the accuracy of knowledge graph construction, the literature data here usually refers to the original literature (ie "primary literature") created based on the author's own research results, and does not include "tertiary literature" such as reviews and guidelines processed based on primary literature.

[0063] Step 120, classify the literature data to obtain the document type of the literature data, and extract the associated entity pairs contained in the literature data, the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the document types include genetics, non-clinical and clinical.

[0064] Specifically, in the field of medicine, the research and development process of new drugs usually includes the following processes: first, genetic research, researchers have found that, for example, changes in certain gene sequences (such as gene mutations) can lead to certain diseases, and this gene can be called a target; researchers will then classify the molecules that can match this target based on the importance of some scientific methods, and then enter preclinical research, where researchers will verify at the cellular and animal levels whether the selected molecule is effective, what its toxicity is, etc.; then it will enter the clinical stage, and through several phases of clinical trials, it will gradually verify whether the drug is effective in treating the disease in humans, what its toxicity is, etc.

[0065] Considering that the construction of knowledge graphs is used to help users explore some potential associated entity pairs, especially those that hope to find associated entity pairs with less intense competition in drug development, which have certain correlations with the disease in "genetics" and "non-clinical" aspects but have not yet been confirmed in "clinical" and have significant research progress recently.

[0066] Therefore, after obtaining the literature data, the literature data can be classified according to the drug development history to obtain the literature type of the literature data. The literature type here can at least include genetics, non-clinical, clinical and other types.

[0067] Among them, the genetics type indicates that the research object of the literature data is in the genetics research stage; the non-clinical type indicates that the research object of the literature data is in the non-clinical research stage; the clinical type indicates that the research object of the literature data is in the clinical research stage; other types indicate that it does not belong to any of genetics, non-clinical and clinical, that is, the literature data does not temporarily meet the requirements of the knowledge graph construction project.

[0068] The type of document data can be realized through a pre-trained text classification model. The document data is input into the pre-trained text classification model to obtain the document type of the document data output by the text classification model.

[0069] The construction of the knowledge graph can be achieved by extracting the associated entity pairs contained in the literature data. The entities in the associated entity pairs here can include at least one of the three biological entities of focus, namely, drugs, targets and diseases. The targets include but are not limited to "genes" and "proteins" encoded by genes.

[0070] An associated entity pair means that there is a certain relationship between the two entities contained in the entity pair, such as an equal relationship, a containing relationship, a contained relationship, a treatment relationship, an indication relationship, etc., which are not listed here one by one.

[0071] For example, the associated entity pairs may include six types: "drug-drug", "target-target", "disease-disease", "drug-disease", "target-disease", and "drug-target". Each type contains associated entity pairs composed of various entity names. For example, the target-disease entity pair may include an associated entity pair consisting of the target EGFR and the disease colorectal tumor; it may also include an associated entity pair consisting of the target HER2 and the disease breast tumor.

[0072] It should be noted that the classification of the document data and the extraction of the associated entity pairs contained in the document data in step 120 may be performed simultaneously or sequentially, and the execution order of the two is not limited.

[0073] For the extraction of related entity pairs, the entities contained in the document data can be firstly extracted, and then the relationship between the entities can be identified, so as to obtain the related entity pairs in the document data.

[0074] Step 130, constructing a knowledge graph with entities in the associated entity pairs as nodes, relationships between the associated entity pairs as edges, and credibility of the associated entity pairs under the document type as attributes of the edges.

[0075] Specifically, based on the associated entity pairs contained in the document data, the credibility of the associated entity pairs in each document type can be further obtained. The credibility can reflect the credibility of any associated entity pair in the three types of "genetics", "non-clinical" and "clinical".

[0076] Then, the knowledge graph is constructed with the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the document type as the attributes of the edges. For example, the entity name of the entity corresponding to the node can be marked at each node in the knowledge graph, and the credibility of the associated entity pair corresponding to the edge in the three dimensions of "genetics", "non-clinical" and "clinical" can be marked on each edge.

[0077] Furthermore, the size of a node in a knowledge graph can also be used to represent the number of documents related to the node. For example, the larger the node, the more documents the entity corresponding to the node involves; conversely, the smaller the node, the fewer documents the entity corresponding to the node involves.

[0078] The thickness of each edge in the knowledge graph can also be used to represent the number of documents involved in the associated entity pair corresponding to the edge. For example, the thicker the edge line, the more documents involved in the associated entity pair corresponding to the edge; conversely, the thinner the edge line, the fewer documents involved in the associated entity pair corresponding to the edge.

[0079] It should be noted that the knowledge graph constructed by the method provided by the present invention can also be combined with knowledge graphs from other sources to become part of a larger knowledge graph.

[0080] The method provided by the embodiment of the present invention classifies the acquired literature data and extracts the associated entity pairs contained in the literature data. On this basis, a knowledge graph is constructed with the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the literature type as the attributes of the edges. It is possible to extract associated entity pairs from a large amount of literature data and construct a knowledge graph based on the credibility of the associated entity pairs under each literature type. Users can quickly and effectively obtain the relationships between the three types of biological entities, namely drugs, targets and diseases, and the credibility of the entity pairs, thereby helping users to explore potential drug research and development information and further assist in the development of new drugs.

[0081] Based on the above embodiments, Figure 2 is a flow chart of the credibility acquisition method provided by the present invention, such as Figure 2 As shown, the credibility includes the evidence level and / or the evidence index, and the steps for obtaining the credibility of the associated entity under the document type include:

[0082] Step 210, determining the evidence level of the associated entity pair under the document type based on the number of citations of the document data to which the associated entity pair belongs and the document type of the document data to which the associated entity pair belongs; and / or,

[0083] Step 220, based on the number of citations of each document data in the document collection, determine the evidence index of the same associated entity pair under the corresponding document type in the document collection, the document type of each document data in the document collection is the same, and each document data contains the same associated entity pair.

[0084] Specifically, the credibility can reflect the credibility of any associated entity pair, such as whether it has been confirmed to be completely credible, or whether it has a certain degree of credibility and needs to be further confirmed. The credibility can include the evidence level and / or the evidence index.

[0085] According to the definition of evidence level, a single document can convert the number of citations into evidence level and summarize the evidence level of each document. Considering that the same associated entity pair may belong to multiple documents, for example, if there are 5 documents including the same associated entity pair, the evidence level of the associated entity pair under each type of document data is determined based on the type of the 5 documents and the maximum number of citations of the same type of documents.

[0086] For example, the types of the five documents are: 2 "genetics" type, 1 "non-clinical" type, and 2 "clinical" type. The evidence level of the associated entity pair under the genetics type can be determined based on the maximum number of citations of the two "genetics" type documents; the evidence level of the associated entity pair under the non-clinical type can be determined based on the number of citations of the one "non-clinical" type document; and the evidence level of the associated entity pair under the clinical type can be determined based on the maximum number of citations of the two "clinical" type documents. The corresponding relationship between the number of citations and the evidence level can be pre-set, and the evidence level of the associated entity pair contained in the document data can be obtained based on the number of citations of the document data.

[0087] The evidence index can be determined by the number of citations of each document data in the document collection. Among them, the document type of each document data in the document collection is the same, and each document data contains the same associated entity pair. The larger the evidence index, the higher the credibility of the entity pair; conversely, the smaller the evidence index, the lower the credibility of the entity pair.

[0088] It is understandable that the level of evidence can be used to evaluate the data of a single document, while the evidence index can only be used to evaluate the data of multiple documents, that is, a collection of documents.

[0089] The evidence level can be used to evaluate a single document data, while the evidence index can only be used to evaluate multiple document data. For any associated entity pair, the credibility of the associated entity pair can be represented by the evidence level under each type of document data, or by the evidence index under each type of document data. The credibility can also be represented by both the evidence level and the evidence index, which is not specifically limited in the embodiments of the present invention.

[0090] The method provided in the embodiment of the present invention characterizes the credibility of each associated entity under various types of literature data through evidence level and / or evidence index, and adopts indicators that are easier to understand and quantify to characterize the credibility, thereby providing a basis for the subsequent construction of the credibility attribute of the knowledge graph.

[0091] Based on any of the above embodiments, step 210 specifically includes:

[0092] Step 211, determining the evidence level of the associated entity pair under the associated document data based on the number of citations of the associated entity pair's associated document data;

[0093] Step 212, based on the document type of the corresponding document data, aggregate the evidence levels of the same associated entity pairs with the same document type, and determine the evidence level of the same associated entity pair under the document type based on the aggregation result.

[0094] Specifically, when calculating the evidence level of the associated entity pairs under each document type, it can be achieved by aggregating the evidence levels of the same associated entity pairs with the same document type.

[0095] First, step 211 is executed to determine, for any associated entity pair, the evidence level of the associated entity pair under the associated document data based on the number of citations of any document data to which the associated entity pair belongs.

[0096] The level of evidence can be divided into four levels, which can be marked as ABCD respectively, among which:

[0097] A: Confirmed, which can be manually annotated, indicating that the entity pair has been confirmed. The confirmation evidence level indicates that it is impossible to overturn the relationship of the entity pair;

[0098] B: Highly cited, usually means the number of citations is greater than or equal to the preset threshold a. High citation evidence level indicates that the entity has high credibility;

[0099] C: Potential, usually refers to the number of citations greater than or equal to the preset threshold b, but less than the preset threshold a, and the average daily citation number of literature data is greater than or equal to the preset threshold c. The potential evidence level indicates that the entity pair has been published recently and will soon be applied by a large number of papers;

[0100] D: Ordinary, usually means that the number of citations is less than the preset threshold b. The ordinary evidence level means that the entity is not very credible.

[0101] It should be noted that the preset threshold can be determined based on the ranking of the number of citations of papers in the field of medicine and health. For example, the top 0.05% are generally considered to be highly cited papers, and the top 15% are generally considered to be highly cited papers. For example, the number of citations of pubmed papers and all papers can be obtained, and the number of citations ranked in the top 0.05% based on the number of citations of the papers is taken as the preset threshold a, and the number of citations ranked in the top 15% is taken as the preset threshold b.

[0102] After obtaining the evidence level of any associated entity pair under the corresponding document data, the evidence levels of the same associated entity pairs with the same document type can be aggregated based on the document type of the corresponding document data, and the evidence level of the same associated entity pair under the document type can be determined based on the aggregation result. The aggregation result here can be the highest evidence level of the same associated entity pair with the same document type. In other words, the highest level of evidence of entity pairs of the same document type can be used as the evidence level of the same entity pair under the document type.

[0103] For example, for the same target-disease entity pair, there are two literature data of the "genetics" type to which the target-disease entity pair belongs, and the corresponding evidence levels are "highly cited" and "potential", respectively. In this case, the "highly cited" and "potential" evidence levels are aggregated, and the highest level "highly cited" is used as the evidence level of the target-disease entity pair under the genetics type.

[0104] In addition, considering that popular research fields have a higher number of citations, while unpopular research fields have a lower number of citations, the evidence levels marked as B / C / D based on the above scheme can also be adjusted in combination with expert opinions.

[0105] The method provided by the embodiment of the present invention aggregates the evidence levels of the same associated entity pairs with the same document type, and determines the evidence level of the same associated entity pair under the document type based on the aggregation result, thereby further improving the calculation accuracy of the evidence level of each associated entity pair.

[0106] Based on any of the above embodiments, step 220 specifically includes:

[0107] Step 221, based on the number of citations of each document data in the document collection, sorting each document data in descending order of the number of citations to obtain a ranking number of each document data;

[0108] Step 222, based on the arrangement sequence number and the number of citations of each document data, determine the evidence index of the same associated entity pair under the corresponding document type in the document collection.

[0109] Specifically, the evidence index can be determined by the number of citations of each document data in the document collection of the same type and containing the same associated entity pair. The evidence index is usually used to evaluate a document collection, in which multiple document data contained in the document collection are of the same type and contain the same associated entity pair.

[0110] The multiple document data may be arranged in descending order according to the number of citations, and at the same time, a serial number is assigned to each document data according to the above order, that is, an arrangement serial number of each document data is obtained.

[0111] On this basis, when the ranking number of a document is greater than the number of citations, the number of the document is the termination number, and the evidence index = termination number - 1. That is, based on the ranking number and the number of citations of each document data, the evidence index of the same associated entity pair under the corresponding document type in the document collection can be determined.

[0112] For example, assume that the document type corresponding to the document collection is "genetics", and there are 10 document data in the document collection, and the 10 document data all contain the same target disease entity pair. Arrange the 10 document data in descending order according to the number of citations, for example, the number of citations of sequence number 1 is 11, the number of citations of sequence number 2 is 11, ..., the number of citations of sequence number 6 is 7, and the number of citations of sequence number 7 is 6. At this time, the arrangement number of the document> the number of citations of the document (7 is greater than 6), and sequence number 7 is the termination number, then the evidence index of the target disease entity pair in the 10 document data of the "genetics" type is 7-1=6. It can be seen that the larger the evidence index, the more credible the entity pair.

[0113] The method provided by the embodiment of the present invention determines the evidence index of the same associated entity under the corresponding document type in the document collection based on the arrangement sequence and the number of citations of each document data, by sorting the document data from high to low, thereby realizing the characterization of credibility by an indicator that can be quantified by the evidence index, and providing a data basis for the construction of the knowledge graph.

[0114] Based on any of the above embodiments, constructing a knowledge graph further includes:

[0115] Based on the evidence level of the associated entity pairs under each document type and the weight of the evidence level under each document type, determine the comprehensive score of the evidence level of the associated entity pairs;

[0116] The color of the knowledge graph edge is constructed based on the comprehensive score of the evidence level of the associated entity pairs.

[0117] Specifically, based on the evidence level of each associated entity pair under each document type, considering that the evidence level includes three dimensions, namely, the three document types of "genetics", "non-clinical" and "clinical", the comprehensive score of the evidence level of the associated entity pair can be further determined. The comprehensive score here can integrate the credibility under the three dimensions, so as to obtain a comprehensive score of the evidence level of each associated entity pair in the comprehensive dimension.

[0118] The comprehensive score of the level of evidence can be achieved by the level of evidence of the associated entity pair under each document type and the weight of the level of evidence under each document type. For example, the level of evidence (number of citations) of a certain associated entity pair under the three types of document data, "genetics", "non-clinical" and "clinical" are "10", "8" and "6" respectively, and the weights of the level of evidence under each type of document data are 0.4, 0.3 and 0.3 respectively. The calculated comprehensive number of citations for the target disease entity pair is 8.2, thus obtaining a comprehensive score of the level of evidence.

[0119] After obtaining the comprehensive score of the evidence level of each associated entity pair, when constructing the knowledge graph, the color of the edge corresponding to each associated entity pair in the knowledge graph can be determined according to the comprehensive score of the evidence level of each associated entity pair. For example, the comprehensive score interval corresponding to each color can be pre-set, and the color of the corresponding edge can be determined according to the comprehensive score.

[0120] The method provided by the embodiment of the present invention determines the comprehensive score of the level of evidence of the associated entity pairs by the level of evidence of the associated entity pairs under each document type and the weight of the level of evidence under each document type, and constructs the color of the edge of the knowledge graph based on the comprehensive score, which can enrich the knowledge expression of the knowledge graph and analyze each associated entity pair from more dimensions.

[0121] Based on any of the above embodiments, the document data is classified in step 120 to obtain the document type of the document data, which specifically includes:

[0122] Step 121, based on the first text classification model, classify the document data by importance to obtain the target document;

[0123] Step 122: classify the target document based on the second text classification model to obtain the document type of the target document.

[0124] Specifically, in order to obtain the document type of the document data, that is, to distinguish whether each document belongs to one of "genetics", "non-clinical" and "clinical", the document data can be classified through a text classification model.

[0125] After the literature data is obtained, it can be screened to eliminate "tertiary literature" such as reviews and guidelines that are processed on the basis of primary literature. However, the literature after the initial screening is still messy and diverse in types. Based on the project requirements, many of the literature screened out are not involved in the project. Therefore, combined with the experience and knowledge of experts and artificial intelligence technology, the literature screened out is further classified to screen out literature that meets the project requirements. This will greatly reduce the cost of labor and the amount of data to be processed in practical applications.

[0126] Based on the experts' experience and knowledge, it is believed that the literature can be divided into "necessary literature" and "unnecessary literature". "Necessary literature" is the literature that needs to be used in accordance with the project requirements, including literature related to genetics, literature related to clinical medicine, and non-clinical but related literature. "Unnecessary literature" refers to other categories, which may be related to the mechanical field, non-medical field, etc.

[0127] First of all, based on project requirements, the first goal is: "needed literature" must never be classified into the "other" category. For the "needed literature" category, there will be periodic manual verification of data to ensure the quality of the project, while the "other" category will be directly discarded.

[0128] Therefore, in the first stage, the document data can be classified by importance based on the first text classification model to obtain the target document data. Here, the importance classification can be understood as the category used to distinguish between the "need document" category and the "other" category, and the target document data is the document data of the "need document" category.

[0129] The first text classification model can be a text binary classification model, which is used to distinguish the categories of documents, namely, "documents needed" and "others". The training and test data used are all expert-annotated data, and the label of "documents needed" is 1, and the label of "others" is 0. The last layer of the text binary classification model is designed as a sigmoid layer, and the output of this layer is a value between 0 and 1. If the output value is close to 1, it means that the input document data category is closer to the "documents needed" category; if the output value is close to 0, it means that the input document data category is closer to the "others" category. In the actual use of the model, a threshold can be set. If the output result is greater than this threshold, it can be indicated that the output value is closer to 1, otherwise, it is closer to 0. The threshold can be adjusted to make the model recall "documents needed" with an accuracy of 100%, thereby meeting the project requirements: "documents needed" will never be classified as "others".

[0130] Secondly, based on the result of the text binary classification model in step 121, it can be determined whether the document is definitely not the required document. If it is definitely not the required document, it will not be considered. If it may be the required document, that is, the target document data, a second classification is performed on the target document data to perform a more refined single-label classification and discrimination of the text, and the target document is more finely labeled as one of: "genetics", "non-clinical", and "clinical", that is, the document types of the target document data include genetics, non-clinical, and clinical.

[0131] The classification of target documents can be achieved through the second text classification model. The data used is still the data annotated by experts. The model input is the document text. The output layer of the model is designed as a softmax layer with three units. The value of each unit represents the probability that the document belongs to the corresponding label above. Finally, the label with the maximum probability is selected as the final output. For example: if a document is input into the second text classification model, the final softmax layer output of the model is [0.2, 0.3, 0.5], which means that the probability that the document belongs to "genetics" is 0.2, the probability that it belongs to "non-clinical" is 0.3, and the probability that it belongs to "clinical" is 0.5, so the final output of the model is: "This document belongs to the clinical type". Therefore, through the second text classification model, it is possible to more accurately determine which category the document belongs to.

[0132] Therefore, through the first text classification model and the second text classification model, it is possible to quickly and accurately identify whether each document is a target document and the type of the target document.

[0133] The method provided by the embodiment of the present invention uses a first text classification model to classify the importance of document data to obtain target document data, which can greatly reduce labor costs and reduce the amount of subsequent data processing. Then, based on the second text classification model, the target document data is quickly and accurately classified to obtain the type of the target document data.

[0134] Based on any of the above embodiments, extracting the associated entity pairs contained in the document data in step 120 includes:

[0135] Step 123, extracting pronouns and entities of a preset type from the document data to obtain the pronouns and entities contained in the document data;

[0136] Step 124, based on the relationship between the pronoun and the entity, perform entity recognition on the pronoun, and replace the pronoun with the entity based on the recognition result;

[0137] Step 125 , respectively identifying the relationships between entities of the same type and the relationships between entities of different types, and obtaining the associated entity pairs contained in the document data.

[0138] Specifically, considering the need to extract desired entities, namely, target, disease and drug entities, from literature data, and the overall literature text is long and most of it is redundant information, the text extraction scope can be set to the title and / or abstract of the literature data.

[0139] The drug dictionary, target dictionary and disease dictionary can be pre-set to include the standard entity names of drugs, targets and diseases.

[0140] The extraction of associated entity pairs is to extract which entities are mentioned in the literature data text and what are the relationships between these entities. The main solution adopted can be pipeline-based relationship extraction, which is mainly divided into three steps: the first is to extract entities from the text, the second is to process the entities and determine the relationships between the extracted entities, and the third is to link the determined entities to the dictionary.

[0141] The first step is to extract entities from the text. There are many categories of entities, including but not limited to: targets, drugs, diseases, etc. First, based on the many pre-built standard dictionaries, rule matching solutions can be used to extract entities, extract all entities that appear in all dictionaries, and ensure the accuracy of dictionary-based entity extraction; secondly, since there are still many entities or entity aliases that are not included in the dictionary, the named entity recognition model in artificial intelligence can be used for extraction, and the required entities can be extracted by parsing the semantics of the sentence. Since there are many entities to be extracted, a multi-task architecture is used to extract multiple entities on one model. The input of the model is the title and abstract text of the paper, and the output of the model is the result of multiple tasks, each task is the extraction of an entity. In addition, since the text contains a large number of pronouns, such as "it", these pronouns are also likely to represent certain entities, so a new task is added to the multi-task model: extracting pronouns in the text.

[0142] Through the first step, we can get many entities and related pronouns from the text. The next thing to do is to associate these entities. First, for all pronouns and entities, use a binary classification model to determine whether each pronoun has an inclusion relationship with each entity in the text. The input is the pronouns and entities identified in the first step and the original text, and the output is whether the input pronouns and entities are related. In the case that there is a relationship between the pronoun and the entity, each pronoun can be replaced with a related entity; in the case that there is no relationship, the pronoun can be directly discarded.

[0143] Secondly, the task of entity reasoning for many entities is to determine the relationship between different entities of the same entity category. This is because there is an entity but there are many different representations in the article, such as full name and abbreviation. A multi-classification model can be used to achieve recognition. Two entities A and B of the same category and the original text are input into the multi-classification model to obtain the relationship between the two entities A and B output by the multi-classification model. The relationship between the two entities A and B can specifically include the following four types: A equals B, A contains B, B contains A, and A and B are unrelated.

[0144] Finally, we need to determine the entity relationship between different entity categories, such as the relationship between an entity in a disease and an entity in a drug. The token-classification model can be used to identify the entity relationship between different entity categories. Two entities of different categories and the original text are input into the token-classification model, and the token-classification model outputs the text that represents the relationship between the two entities in the original text, thereby obtaining several entity pairs and their relationships.

[0145] For example, the literature data includes the following text: "A is a medicine, referred to as A', which can treat Class B diseases, which include B1, B2, and B3".

[0146] First, entities can be extracted through the entity recognition task: drugs: {A, A'}, diseases: {B, B1, B2, B3}, pronouns: {it}.

[0147] Secondly, by eliminating pronouns through the binary classification model, we can get A = it, then the current entities only include drugs: {A = it, A'}, diseases: {B, B1, B2, B3}.

[0148] Then, through entity reasoning, we can determine the relationship between different entities in the same entity type. We can get A=A'=it in the drug type, and B includes B1, B2 and B3 in the disease type, and B1, B2 and B3 are unrelated to each other. The result is drug: {A=A'=it}, disease: {B(B1, B2, B3)}.

[0149] Finally, by inputting entities of different categories into the token-classification model, we can get a set of relations: A-treatment-B. Therefore, we can finally get drug {A=A'=it}-treatment-disease {B(B1, B2, B3)}.

[0150] It should be noted that after obtaining all entities in the text and their relationships, all entities can be mapped to the dictionary. Since most of the identified entities are proper nouns, the rule matching + fuzzy matching method can be used to map the entities to the dictionary. For example, the drug predicted in the second step is: {A=A'=it}, and the entry C in the dictionary has an alias A', so the drug {A=A'=it} can be linked to the entry C in the dictionary. For some that cannot be linked at all, manual review and verification are used to determine whether to put the entry into the dictionary.

[0151] The method provided in the embodiment of the present invention realizes the extraction of associated entity pairs in document data through artificial intelligence technology, thereby obtaining the associated entity pairs contained in the document data, providing a basis and convenience for the construction of a knowledge graph.

[0152] Based on any of the above embodiments, Figure 3 It is a schematic diagram of the process of the retrieval method provided by the present invention, such as Figure 3 As shown, a retrieval method is provided, comprising:

[0153] Step 310, receiving a search keyword sent by a user terminal, the search keyword including at least one of an entity name, an associated entity pair, a document type, and the credibility of the associated entity pair under any document type;

[0154] Step 320, determine the target document data matching the search keyword from the pre-constructed knowledge graph, and return the target document data to the user terminal; the knowledge graph is determined based on the knowledge graph construction method.

[0155] Specifically, literature data from various sources can be used to extract related entity pairs and calculate credibility through the methods described in the above embodiments, thereby constructing a knowledge graph, which in turn provides convenience for information search and positioning. Moreover, on this basis, a retrieval system and / or a literature subscription system can be constructed to facilitate the rapid search and positioning of target information.

[0156] The user can input the search keyword through a user terminal in the form of a smart phone, computer, tablet computer, etc., and send the search keyword to the server for retrieval. The search keyword here is the keyword for finding relevant information from the knowledge graph. The search keyword can specifically include at least one of the entity name, associated entity pair, document type, and the credibility of the associated entity pair under any document type.

[0157] After receiving the search keywords, the target document data matching the search keywords can be found in the predetermined knowledge graph, and the found target document data can be returned to the user terminal for the user to view and analyze.

[0158] It should be noted that the knowledge graph constructed by the method provided by the present invention can also be combined with knowledge graphs from other sources to become part of a larger knowledge graph. The search keywords include but are not limited to: drug name, target name, disease name, target type, disease type, associated entity pair, document type, number of citations, credibility information, paper publication date, etc.

[0159] Among them, the target type and disease type are obtained from the pre-constructed target type dictionary and disease type dictionary.

[0160] The retrieval method provided in the embodiment of the present invention realizes rapid retrieval of target document data matching the search keywords through the knowledge graph, which helps to improve the efficiency of information query and helps R&D personnel to quickly obtain the relationship between the three types of biological entities and the credibility of the target entity pairs.

[0161] Based on any of the above embodiments, a knowledge graph construction and retrieval method is provided, including:

[0162] S1, obtain literature data.

[0163] S2, classifying the literature data to obtain the document type of the literature data, and extracting the associated entity pairs contained in the literature data, the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the document types include genetics, non-clinical and clinical.

[0164] S3, obtain the credibility of the associated entity pair under the document type. The credibility includes the evidence level and / or evidence index. S3 specifically includes:

[0165] Determine the evidence level of the associated entity pair under the document type based on the number of citations of the associated entity pair's document data and the document type of the associated document data; and / or,

[0166] Based on the number of citations of each document data in the document collection, the evidence index of the same associated entity pair under the corresponding document type in the document collection is determined. The document type of each document data in the document collection is the same, and each document data contains the same associated entity pair.

[0167] S4, constructs a knowledge graph with entities in associated entity pairs as nodes, relationships between associated entity pairs as edges, and credibility of associated entity pairs under document types as attributes of edges.

[0168] S5, receiving a search keyword sent by a user terminal, where the search keyword includes at least one of an entity name, an associated entity pair, a document type, and a credibility of the associated entity pair under any document type.

[0169] S6, determining the target document data matching the search keyword from the knowledge graph constructed in S4, and returning the target document data to the user terminal.

[0170] The embodiment of the present invention provides a solution for obtaining entity pairs and the credibility of entity pairs from massive documents. In addition, the knowledge graph is a more structured knowledge base that effectively organizes scattered knowledge and constructs a knowledge graph with "drugs", "targets" and "diseases", three types of biological entities that are key to medical researchers, as nodes, which helps R&D personnel quickly obtain the relationship between the three types of biological entities and the credibility of target entity pairs. In addition, it can also help R&D personnel to explore potential drug development information, thereby facilitating new drug development.

[0171] The knowledge graph construction device provided by the present invention is described below. The knowledge graph construction device described below and the knowledge graph construction method described above can be referenced to each other.

[0172] Based on any of the above embodiments, Figure 4 is a schematic diagram of the structure of the knowledge graph construction device provided by the present invention, such as Figure 4 As shown, the knowledge graph construction device includes a data acquisition unit 410, an entity pair extraction unit 420 and a graph construction unit 430, wherein:

[0173] A data acquisition unit 410 is used to acquire document data;

[0174] An entity pair extraction unit 420 is used to classify the document data, obtain the document type of the document data, and extract the associated entity pairs contained in the document data, wherein the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the document types include genetics, non-clinical and clinical;

[0175] The graph construction unit 430 is used to construct a knowledge graph using the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the document type as attributes of the edges.

[0176] The knowledge graph construction device provided by the embodiment of the present invention classifies the acquired document data and extracts the associated entity pairs contained in the document data. On this basis, the knowledge graph is constructed with the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the document type as the attributes of the edges. It is possible to extract associated entity pairs from a large amount of document data and construct a knowledge graph based on the credibility of the associated entity pairs under each document type. Users can quickly and effectively obtain the relationships between the three types of biological entities, namely drugs, targets and diseases, and the credibility of the entity pairs, thereby helping users to explore potential drug development information and further assist in the development of new drugs.

[0177] Based on any of the above embodiments, it further includes a credibility acquisition unit, which is used to:

[0178] Determining the level of evidence of the associated entity pair under the document type based on the number of citations of the associated entity pair's document data and the document type of the associated document data; and / or,

[0179] Based on the number of citations of each document data in the document collection, the evidence index of the same associated entity pair under the document type corresponding to the document collection is determined. The document types of each document data in the document collection are the same, and each document data contains the same associated entity pair.

[0180] Based on any of the above embodiments, the credibility acquisition unit is specifically used to:

[0181] Determining the evidence level of the associated entity pair under the associated document data based on the number of citations of the associated entity pair's associated document data;

[0182] Based on the document type of the belonging document data, the evidence levels of the same associated entity pairs with the same document type are aggregated, and the evidence level of the same associated entity pair under the document type is determined based on the aggregation result.

[0183] Based on any of the above embodiments, the credibility acquisition unit is specifically used to:

[0184] Based on the number of citations of each document data in the document collection, the document data are sorted in descending order according to the number of citations to obtain the arrangement sequence number of each document data;

[0185] Based on the arrangement sequence number and the number of citations of each document data, the evidence index of the same associated entity pair under the corresponding document type of the document collection is determined.

[0186] Based on any of the above embodiments, the graph construction unit is further used for:

[0187] Determine a comprehensive score of the level of evidence of the associated entity pair based on the level of evidence of the associated entity pair under each document type and the weight of the level of evidence under each document type;

[0188] The color of the knowledge graph edge is constructed based on the comprehensive score of the evidence level of the associated entity pairs.

[0189] Based on any of the above embodiments, the entity pair extraction unit is further specifically configured to:

[0190] Based on the first text classification model, the document data is classified according to importance to obtain a target document;

[0191] Based on the second text classification model, the target document is classified to obtain the document type of the target document.

[0192] Based on any of the above embodiments, the entity pair extraction unit is further specifically configured to:

[0193] Extracting pronouns and entities of a preset type from the document data to obtain pronouns and entities contained in the document data;

[0194] Based on the relationship between the pronoun and the entity, performing entity recognition on the pronoun, and replacing the pronoun with the entity based on the recognition result;

[0195] Relationships between entities of the same type and relationships between entities of different types are identified respectively to obtain associated entity pairs contained in the document data.

[0196] Based on any of the above embodiments, Figure 5 is a schematic diagram of the structure of the retrieval device provided by the present invention, such as Figure 5 As shown, the search device includes a keyword receiving unit 510 and a data determining unit 520, wherein:

[0197] The keyword receiving unit 510 is used to receive a search keyword sent by a user terminal, wherein the search keyword includes at least one of an entity name, an associated entity pair, a document type, and a credibility of an associated entity pair under any document type;

[0198] The data determination unit 520 is used to determine the target document data matching the search keyword from a pre-constructed knowledge graph and return the target document data to the user terminal; the knowledge graph is determined based on the knowledge graph construction method.

[0199] The retrieval device provided in the embodiment of the present invention realizes rapid retrieval of target document data matching the search keywords through the knowledge graph, which helps to improve the efficiency of information query and helps R&D personnel to quickly obtain the relationship between the three types of biological entities and the credibility of the target entity pairs.

[0200] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the knowledge graph construction method, which includes: obtaining document data; classifying the document data to obtain the document type of the document data, and extracting the associated entity pairs contained in the document data, wherein the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the document types include genetics, non-clinical and clinical; constructing the knowledge graph with the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the document type as the attributes of the edges.

[0201] The processor may call the logic instructions in the memory to execute the retrieval method, the method comprising: receiving a retrieval keyword sent by a user terminal, the retrieval keyword comprising at least one of an entity name, an associated entity pair, a document type, and a credibility of the associated entity pair under any document type;

[0202] Determine target document data matching the search keyword from a pre-built knowledge graph, and return the target document data to the user terminal;

[0203] The knowledge graph is determined based on the knowledge graph construction method.

[0204] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0205] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer can execute the knowledge graph construction method provided by the above methods, the method including:

[0206] Acquire literature data; classify the literature data to obtain the literature type of the literature data, and extract associated entity pairs contained in the literature data, wherein the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the literature types include genetics, non-clinical and clinical; construct a knowledge graph using the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the literature type as the attributes of the edges.

[0207] When the computer program is executed by a processor, the computer can execute the retrieval method provided by the above methods, which method includes:

[0208] receiving a search keyword sent by a user terminal, wherein the search keyword includes at least one of an entity name, an associated entity pair, a document type, and a credibility of the associated entity pair under any document type;

[0209] Determine target document data matching the search keyword from a pre-built knowledge graph, and return the target document data to the user terminal;

[0210] The knowledge graph is determined based on the knowledge graph construction method.

[0211] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the knowledge graph construction method provided by the above methods, the method comprising:

[0212] Acquire literature data; classify the literature data to obtain the literature type of the literature data, and extract associated entity pairs contained in the literature data, wherein the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the literature types include genetics, non-clinical and clinical; construct a knowledge graph using the entities in the associated entity pairs as nodes, the relationships between the associated entity pairs as edges, and the credibility of the associated entity pairs under the literature type as the attributes of the edges.

[0213] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0214] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0215] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A knowledge graph construction method, characterized in that: include: Obtain literature data; Classifying the literature data to obtain the literature type of the literature data, and extracting the associated entity pairs contained in the literature data, wherein the entities in the associated entity pairs include at least one of drugs, targets and diseases, and the literature types include genetics, non-clinical and clinical; Constructing a knowledge graph by taking entities in the associated entity pairs as nodes, relationships between the associated entity pairs as edges, and credibility of the associated entity pairs under the document types as attributes of the edges; The credibility includes the evidence level and / or the evidence index, and the steps of obtaining the credibility of the associated entity under the document type include: Determining the evidence level of the associated entity pair under the document type based on the number of citations of the associated entity pair's document data and the document type of the associated document data; and / or, Based on the number of citations of each document data in the document collection, the evidence index of the same associated entity pair under the document type corresponding to the document collection is determined. The document type of each document data in the document collection is the same, and the each document data contains the same associated entity pair. The evidence index represents the credibility of the same associated entity pair in the document collection. The larger the evidence index, the higher the credibility.

2. The knowledge graph construction method according to claim 1, characterized in that: The determining the evidence level of the associated entity pair under the document type based on the number of citations of the associated entity pair's document data and the document type of the associated document data comprises: Determining the evidence level of the associated entity pair under the associated document data based on the number of citations of the associated entity pair's associated document data; Based on the document type of the belonging document data, the evidence levels of the same associated entity pairs with the same document type are aggregated, and the evidence level of the same associated entity pair under the document type is determined based on the aggregation result.

3. The knowledge graph construction method according to claim 1, characterized in that: The step of determining the evidence index of the same associated entity under the corresponding document type in the document collection based on the number of citations of each document data in the document collection includes: Based on the number of citations of each document data in the document collection, the document data are sorted in descending order according to the number of citations to obtain the arrangement sequence number of each document data; Based on the arrangement sequence number and the number of citations of each document data, the evidence index of the same associated entity pair under the corresponding document type of the document collection is determined.

4. The knowledge graph construction method according to claim 1, characterized in that: The constructing of the knowledge graph further includes: Determine a comprehensive score of the level of evidence of the associated entity pair based on the level of evidence of the associated entity pair under each document type and the weight of the level of evidence under each document type; The color of the knowledge graph edge is constructed based on the comprehensive score of the evidence level of the associated entity pairs.

5. The method for constructing a knowledge graph according to any one of claims 1 to 4, characterized in that: The classifying the document data to obtain the document type of the document data includes: Based on the first text classification model, the document data is classified according to importance to obtain a target document; Based on the second text classification model, the target document is classified to obtain the document type of the target document.

6. The knowledge graph construction method according to any one of claims 1 to 4, characterized in that: The extracting the associated entity pairs contained in the document data includes: Extracting pronouns and entities of a preset type from the document data to obtain pronouns and entities contained in the document data; Based on the relationship between the pronoun and the entity, performing entity recognition on the pronoun, and replacing the pronoun with the entity based on the recognition result; Relationships between entities of the same type and relationships between entities of different types are identified respectively to obtain associated entity pairs contained in the document data.

7. A search method, characterized in that: include: receiving a search keyword sent by a user terminal, wherein the search keyword includes at least one of an entity name, an associated entity pair, a document type, and a credibility of the associated entity pair under any document type; Determine target document data matching the search keyword from a pre-built knowledge graph, and return the target document data to the user terminal; The knowledge graph is determined based on the knowledge graph construction method as described in any one of claims 1 to 6.

8. A knowledge graph construction device, characterized in that: include: A data acquisition unit, used for acquiring literature data; an entity pair extraction unit, configured to classify the document data, obtain the document type of the document data, and extract associated entity pairs contained in the document data, wherein the entities in the associated entity pairs include at least one of drugs, targets, and diseases, and the document types include genetics, non-clinical, and clinical; A graph construction unit, configured to construct a knowledge graph using entities in the associated entity pairs as nodes, relationships between the associated entity pairs as edges, and credibility of the associated entity pairs under the document types as attributes of the edges; The credibility includes the evidence level and / or the evidence index, and the knowledge graph construction device further includes a credibility acquisition unit for: Determining the level of evidence of the associated entity pair under the document type based on the number of citations of the associated entity pair's document data and the document type of the associated document data; and / or, Based on the number of citations of each document data in the document collection, the evidence index of the same associated entity pair under the document type corresponding to the document collection is determined. The document type of each document data in the document collection is the same, and the each document data contains the same associated entity pair. The evidence index represents the credibility of the same associated entity pair in the document collection. The larger the evidence index, the higher the credibility.

9. A search device, characterized in that: include: A keyword receiving unit, configured to receive a search keyword sent by a user terminal, wherein the search keyword includes at least one of an entity name, an associated entity pair, a document type, and a credibility of an associated entity pair under any document type; A data determination unit is used to determine target document data matching the search keyword from a pre-constructed knowledge graph and return the target document data to the user terminal; the knowledge graph is determined based on the knowledge graph construction method described in any one of claims 1 to 6.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the knowledge graph construction method as described in any one of claims 1 to 6, or the retrieval method as described in claim 7.

Images