A method and device for determining target information based on knowledge graph

Abstract

According to an embodiment, a method and device for determining target information based on a knowledge graph are disclosed. The method includes: a similarity calculation step, in which, based on the TFIDF algorithm, the TFIDF value of the target name for each node entry is calculated as the similarity score between the node entry and the target, and a selection step, in which the node entries with similarity scores above a threshold are taken as candidate entries, and a determination step, in which the target information is determined from the candidate entries.

Description

Technical Field

[0001] The present invention relates to data mining technology, and in particular to a method and device for determining target information based on a knowledge graph. Background Art

[0002] With the development of e-commerce and mobile payment, more and more merchants or businesses need to maintain various relevant information to ensure that economic activities are carried out safely and orderly. Merchants' relevant information includes, for example, merchant industry information, store location information, business license information, etc. In the prior art, the filling, supplementation or verification of various relevant information of merchants or businesses is usually maintained by the person in charge of the merchants or businesses or by relevant market researchers. In this way, the relevant information of the merchants may be incomplete or inaccurate. Summary of the invention

[0003] According to one embodiment, a method for determining target information based on a knowledge graph is disclosed. The method includes: a similarity calculation step, in which, based on the TFIDF algorithm, the TFIDF value of the target name for each node entry is calculated as the similarity score between the node entry and the target, and a selection step, in which the node entries with similarity scores above a threshold are taken as candidate entries, and a determination step, in which the target information is determined from the candidate entries.

[0004] The knowledge graph contains multiple nodes and edges between nodes. Two nodes and the edges between the two nodes constitute a node entry. A node entry can represent the relationship between two nodes or a node and the attributes of the node.

[0005] In one example, the three elements in a node entry consist of two nodes and an edge between the two nodes. When the node entry represents two nodes and the relationship between the two nodes, the two nodes in the node entry refer to entities or concepts. When the node entry represents a node and the attributes of the node, one node in the node entry refers to an entity or concept, the other node refers to the attribute value of the entity or concept, and the edge refers to the attribute.

[0006] In one example, the method of this embodiment also includes: a word segmentation step, wherein a word segmentation operation is performed on the node description words and attribute values ​​in each node entry, a word segmentation operation is performed on the name of the target, and an inverted index establishment step, wherein, after the word segmentation step, for each node entry, an inverted index from sub-words to node entries is established for the words that are the result of the word segmentation operation.

[0007] In one example, in the similarity calculation step, based on the TFIDF algorithm, the TFIDF value of each subword of the target name for each node entry is calculated, and the sum of the TFIDF values ​​of each subword of the target name is taken as the similarity score of the target relative to a node entry.

[0008] In one example, in the determining step, a logistic regression function is used to calculate an estimated value of each candidate entry relative to the name of the target, and information about the target is determined from the candidate entries according to the estimated value.

[0009] In one example, the method of this embodiment further includes: a filtering step, in which node entries containing the relationship or attribute are filtered from the knowledge graph according to the relationship or attribute of the target to be determined.

[0010] According to one embodiment, a device for determining target information based on a knowledge graph is disclosed. The device includes: a computer processor, a computer-readable storage medium, and a program instruction stored on the computer-readable storage medium, wherein the program instruction is executed by the processor to perform the following steps: a similarity calculation step, wherein, based on the TFIDF algorithm, the TFIDF value of the name of the target for each node entry is calculated as the similarity score between the node entry and the target, and a selection step, wherein the node entries with similarity scores above a threshold are taken as candidate entries, and a determination step, wherein the target information is determined from the candidate entries.

[0011] In one example, the processor is configured to execute: a word segmentation step, wherein a word segmentation operation is performed on the node description words and attribute values ​​in each node entry, a word segmentation operation is performed on the name of the target, and an inverted index creation step, wherein, after the word segmentation step, for each node entry, an inverted index from sub-words to node entries is created for the words that are the result of the word segmentation operation.

[0012] In one example, in the similarity calculation step, based on the TFIDF algorithm, the TFIDF value of each subword of the target name for each node entry is calculated, and the sum of the TFIDF values ​​of each subword of the target name is taken as the similarity score of the target relative to a node entry.

[0013] In one example, in the determining step, a logistic regression function is used to calculate an estimated value of each candidate entry relative to the name of the target, and information about the target is determined from the candidate entries according to the estimated value.

[0014] In one example, the processor is configured to perform: a filtering step, in which node entries containing the relationship or attribute are filtered from the knowledge graph according to the relationship or attribute of the target to be determined.

[0015] According to one embodiment, a device for determining target information based on a knowledge graph is disclosed. The device includes: the device for determining target information based on a knowledge graph includes a word segmentation unit, which is used to perform word segmentation operations on node description words and attribute values ​​in each node entry, and to perform word segmentation operations on the name of the target; a similarity calculation unit, which is used to calculate the TFIDF value of the name of the target for each node entry based on the TFIDF algorithm as the similarity score between the node entry and the target; a selection unit, which is used to take the node entries with similarity scores above a threshold as candidate entries, and a determination unit, which is used to determine the target information from the candidate entries.

[0016] The similarity calculation unit can also be configured to calculate the TFIDF value of each subword of the target name for each node entry based on the TFIDF algorithm, and take the sum of the TFIDF values ​​of each subword of the target name as the similarity score of the target relative to a node entry. The determination unit can also be configured to use a logistic regression function to calculate the estimated value of each candidate entry relative to the name of the target, and determine the information of the target from the candidate entries based on the estimated value.

[0017] Optionally, the device may further include an inverted index establishment unit, which is used to establish an inverted index from subwords to node entries for each node entry after the word segmentation unit. Optionally, the device may further include a filtering unit, which is used to filter node entries containing the relationship or attribute of the target to be determined from the knowledge graph.

[0018] According to other embodiments, a computer-readable storage medium is provided having instructions located thereon that, when executed, cause a computing device to perform the steps of the methods described herein.

[0019] One or more embodiments described in this application can actively maintain target information, supplement or correct key information of the target, and avoid false information. Other advantages of some embodiments include timely updating of target data, improving target data management capabilities, and being able to cross-verify data reported autonomously by the target to improve the accuracy of target information. One or more embodiments described in this application can be targeted at an enterprise, a merchant, an individual, or any other entity or concept. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The foregoing and other features and advantages of various embodiments of the present invention will become apparent from the following detailed description taken in conjunction with the accompanying drawings.

[0021] Figure 1 It is a schematic diagram showing how to determine target information based on a knowledge graph according to one embodiment.

[0022] Figure 2 is a schematic diagram illustrating triple classification in a knowledge graph according to one embodiment.

[0023] Figure 3 It is a flow chart showing a method for determining target information based on a knowledge graph according to one embodiment.

[0024] Figure 4 is a schematic diagram showing word segmentation of triples and suggestion of inverted index according to one embodiment.

[0025] Figure 5 It is a block diagram showing the system structure for determining target information based on a knowledge graph according to one embodiment.

[0026] Figure 6 It is a block diagram showing a program module unit of an apparatus for determining target information based on a knowledge graph according to one embodiment. DETAILED DESCRIPTION

[0027] In the following description, many specific details are set forth in order to provide a thorough understanding of the exemplary embodiments of the present invention. However, it will be appreciated by those of ordinary skill in the art that the embodiments described herein may be practiced without these specific details. In other cases, the present disclosure does not describe in detail known algorithms, processes, and components, so as not to obscure the embodiments described herein. The phrases "in an embodiment," "in an example," and "in an implementation" are used in the description herein, which may each indicate one or more of the same or different embodiments. The terms used herein are intended to better explain the principles of the various embodiments, practical applications, improvements to the technology, or to enable those of ordinary skill in the art to understand and implement the embodiments disclosed herein.

[0028] Embodiments of the present application relate to a method and apparatus for determining merchant information based on a knowledge graph.

[0029] As those skilled in the art already know, a knowledge graph is a semantic network that aims to describe various entities or concepts and their relationships that exist in the real world, thereby forming a semantic network graph. Nodes can represent entities or concepts, and edges represent relationships. Here, entities refer to specific things that are distinguishable and exist independently, such as specific people, specific cities, specific businesses, etc. Concepts refer to a collection of entities with common characteristics, such as ethnic groups, countries, industries, etc. There are different relationships between different entities / concepts. Nodes can also represent attribute values ​​of entities / concepts, and edges represent attributes. For example, the attributes of a country can be area, population, etc. At this time, another node pointed to by the node representing the country through the edge as an attribute can be the attribute value.

[0030] As those skilled in the art already know, the construction of a knowledge graph may include processes such as data acquisition, information acquisition, and knowledge fusion. The data acquisition process may include extracting structured data from the business data of an existing system and extracting unstructured data from the Internet. The information acquisition process may include identifying named entities from text data, extracting associations between entities from related corpora, linking entities through relationships, and extracting attribute information of entities from data sources. In the knowledge fusion process, entity linking and knowledge merging are performed.

[0031] In this application, the knowledge graph that has been constructed is used to represent the relevant information of the entity. For example, a merchant (entity) has a subordinate relationship (relationship) with an industry (concept). The knowledge graph can also contain the attribute information of the entity, for example, a merchant (entity) is located in (descriptive word) a geographical location (attribute).

[0032] Figure 1 It is a schematic diagram showing how to determine target information based on a knowledge graph according to one embodiment.

[0033] Figure 1 The left side of the diagram schematically shows the knowledge graph that has been constructed. The knowledge graph contains multiple nodes and edges between nodes. The edge between two nodes can represent two entities and the relationship between the two entities. For example, Merchant A - belongs to Industry B. The edge between two nodes can also represent an entity and the attribute value of this entity pointed to by the edge as an attribute. For example, Merchant A - Location - City X, Road Y, No. Z

[0034] In an embodiment of the present application, two nodes and the edge between the two nodes may constitute a node entry. The three elements of a node entry, such as (entity 1, relationship, entity 2), (entity, attribute, attribute value), may form a triple. A knowledge graph may be viewed as a collection of triples. The Resource Description Framework (RDF) may be used as a data model to store RDF-based triples, or a graph database may be used to store triples.

[0035] like Figure 1As shown on the right side of , when a node entry represents two nodes and the relationship between the two nodes, the two nodes in the node entry refer to entities or concepts. Such a node entry can be called a relationship entry, and the triple of the relationship entry can be composed of (node ​​descriptor, relationship descriptor, node descriptor), and the three elements in the triple can refer to (entity, relationship, entity) respectively. When a node entry represents a node and the attribute of the node, one node in the node entry refers to an entity or concept, the other node refers to the attribute value of the entity or concept, and the edge refers to the attribute. Such a node entry can be called an attribute entry, and the triple of the attribute entry can be composed of (node ​​descriptor, attribute descriptor, attribute value), and the three elements in the triple can refer to (entity, attribute, attribute value) respectively.

[0036] Figure 1 The concept of filtering specific node entries from the knowledge graph and using the triple information of the node entries as the information of the merchant is schematically shown. It can be understood by those skilled in the art that one or more embodiments described in this application can also be applied to enterprises, individuals, or any other entities or concepts. For example, the target information to be determined can be the credit information of an individual, as well as the relationship between individuals, or the relationship between an individual and other entities (e.g., an enterprise).

[0037] According to one embodiment, first, the similarity score of each node entry relative to the merchant is obtained by calculating the similarity between the merchant and the node entry from the n node entries in the knowledge graph, and then k candidate entries are selected according to the similarity score. For example, the similarity scores can be sorted from high to low, and the top k node entries are selected from the n node entries as candidate entries. The content of the triples from these k candidate entries can be selected as the merchant's information or the merchant's credibility can be verified based on the content of the triples of these k candidate entries. For example, the attribute value of the entity in the triple can be used as the attribute value of the merchant's changed attribute, or it can be confirmed that the entity in the triple has a relationship in the triple with the merchant.

[0038] Figure 2 is a schematic diagram illustrating triple classification in a knowledge graph according to one embodiment.

[0039] As mentioned above, the knowledge graph contains multiple nodes and edges between nodes. Two nodes and the edges between the two nodes constitute a node entry. A node entry can represent the relationship between two nodes or a node and the attributes of the node. Figure 2As shown, node entries can be divided into relationship entries and attribute entries. A relationship entry is, for example, a (entity, relationship, entity) triple, and an attribute entry can be, for example, a (entity, attribute, attribute value) triple. For a relationship entry, (node ​​descriptor, relationship descriptor, node descriptor) respectively describes (entity, relationship, entity). For an attribute entry, (node ​​descriptor, attribute descriptor, attribute value) respectively describes (entity, attribute, attribute value).

[0040] Figure 3 FIG. 1 is a flow chart showing a method for determining target information based on a knowledge graph according to an embodiment. Figure 3 As shown, the method for determining merchant information based on the knowledge graph includes steps S310 to S340.

[0041] In the word segmentation step S310, the elements of the node entry are segmented. Among them, the node description words and attribute values ​​in each node entry are segmented, and the name of the merchant is segmented. Here, the name of the merchant can be any character or symbol used to describe the merchant. The result of the word segmentation operation is to divide the node description words into several sub-words according to semantics, and to divide the name of the merchant into several sub-words. Those skilled in the art can understand that a variety of tools can be used for Chinese word segmentation, such as Ansj, jieba, ICTCLAS, etc. For example, for a node entry, such as the triple (Yonghui Supermarket Tangzhen Store, Location, No. 777, Gaoke East Road, Pudong New Area, Shanghai), after word segmentation, it becomes (Yonghui Supermarket / Tangzhen / Store, Location, Shanghai / Pudong New Area / Gaoke East Road / No. 777). The node description word "Yonghui Supermarket Tangzhen Store" is segmented into three sub-words: "Yonghui Supermarket", "Tangzhen", and "Store". The attribute value "No. 777, Gaoke East Road, Pudong New Area, Shanghai" is segmented into four sub-words: "Shanghai", "Pudong New Area", "Gaoke East Road", and "No. 777". For another example, for the name of the merchant "Yonghui Supermarket on Gaoke East Road", after segmentation, two sub-words "Gaoke East Road" and "Yonghui Supermarket" can be obtained.

[0042] In another embodiment, after the word segmentation step, for each node entry, an inverted index from subwords to node entries is established for the words that are the result of the word segmentation operation. For triples 1 to triples n, each triple node description word and attribute value is segmented, and then each subword obtained is mapped to all triples containing the subword. The efficiency of the similarity calculation in step S320 can be improved by the inverted index. Those skilled in the art will appreciate that a variety of tools can be used to establish an inverted index, such as Lucene, Elasticsearch, Solr, etc. Figure 4As shown in an example, in the word segmentation step, the entity description words 1-n contained in each of the entity entries (in the form of triples) 1-n are segmented to obtain subwords 1-m for each entity entry, and then a reverse index to the entity entry is established for each subword 1-m.

[0043] In the similarity calculation step S320, the similarity between the merchant and the node is calculated, that is, the similarity between the merchant and each triple is calculated. In an embodiment of the present application, based on the TFIDF algorithm, the TFIDF value of the merchant's name for each node entry is calculated as the similarity score between the node entry and the merchant. TFIDF (term frequency-inverse document frequency) can be used to evaluate the importance of a word to a file. The importance of a word increases in direct proportion to the number of times it appears in the file, but at the same time it decreases in inverse proportion to the frequency of its appearance in the corpus. Term frequency (TF) indicates the frequency of an entry (keyword) in a text. Inverse document frequency (IDF): The IDF of a particular word can be obtained by dividing the total number of files by the number of files containing the word, and then taking the logarithm of the quotient. TFIDF = TF*IDF. In one example, the TFIDF value of each subword of the merchant's name for each node entry is calculated, and the sum of the TFIDF values ​​of each subword of the merchant's name is taken as the similarity score of the merchant relative to a node entry.

[0044] In one example, the TFIDF value of the merchant name for each node entry is calculated by the following formula (1):

[0045]

[0046] Here M i is the name of the merchant, Tr i is a triple. By segmenting the name of the merchant, the subwords of the merchant's name (M i1 , M i2 , ...M in ). Among them, N t (M ij ,Tr i ) represents the subword M ij In the triple Tr i The number of occurrences in, Ns(M ij ) indicates M ij The number of occurrences in all triples, |D| represents the total number of triples in the knowledge graph used, N d (M ij ) means it contains the subword M ijThe number of triples of the merchant name and the triple Tr is obtained by summing the TFIDF values ​​of each subword. i Here, k and b are smoothing coefficients. In a preferred embodiment, k is 0.75 and b is 0.5. In the similarity calculation step S320, the similarity between the merchant and the node is calculated, that is, the similarity between the merchant and each triple is calculated. In the embodiment of the present application, in addition to the TFIDF algorithm, candidate triples can also be screened based on similarity algorithms such as BM25 and Jaccard coefficient.

[0047] In the selection step S330, the candidate node entries are screened according to the similarity scores, wherein the node entries (triplets) with similarity scores above the threshold are taken as candidate entries (candidate triplets). For example, the similarity scores can be sorted from high to low, and the top k node entries are selected from the n node entries as candidate entries. In one example, the value of k is 10.

[0048] In the determination step S340, the information of the merchant is determined from the candidate items. For example, the contents of the triples of the k candidate items can be selected as the merchant information or the merchant information can be verified based on the contents of the triples of the k candidate items. For example, the attribute value of the entity in the triple can be used as the attribute value of the merchant's attribute, or it can be confirmed that the entity in the triple has a relationship in the triple with the merchant.

[0049] For step S340, in another embodiment, k candidate items are further optimized, matched and sorted. In the determination step, a logistic regression function can be used to calculate the estimated value of each candidate item relative to the name of the merchant, and the merchant information is determined from the candidate items according to the estimated value.

[0050] In one example, the estimated value of the merchant name for each candidate entry is calculated by the following formula (2):

[0051]

[0052] Here F(M i ',Tr i ')" can use the Sigmoid function and the logistic regression classification algorithm, or other machine learning algorithms such as SVM, GBDT, DNN, LambdaRank, etc.

[0053] M i ' and Tr i 'Indicates the name of the merchant M i and triple Tr i The eigenvector of M can be calculated using various tools. iand triple Tr i For example, one-hot encoding can be used, or embedded vector feature calculation methods based on contextual information such as word2vec and BERT can be used, or knowledge graph random walk algorithm methods such as PRA can be combined.

[0054] In formula (2), [M i ',Tr i '] represents the feature vector M i ' and Tr i ', where w is the vector [M i ',Tr i '] weight, b is a constant coefficient. Those skilled in the art will appreciate that the arithmetic values ​​of w and b can be calculated by a stochastic gradient descent algorithm.

[0055] The estimated value of each candidate entry relative to the name of the merchant can be calculated by equation (2). The element of the entry with the highest estimated value can be used as the merchant information.

[0056] In one example, the method further includes a filtering step, in which, according to the relationship or attribute of the merchant to be determined, the node entries containing the relationship or attribute are filtered from the knowledge graph. Thus, the amount of data used to find merchant information can be reduced, and only the node entries in the knowledge graph containing the relationship or attribute are used for operations such as word segmentation, sorting, and similarity calculation.

[0057] Other embodiments of the present invention may be a computer system and a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions may be stored for causing a processor to execute various embodiments of the present invention.

[0058] Figure 5 It is a block diagram showing the system structure for determining target information based on a knowledge graph according to one embodiment. Figure 5 The illustrated system 10 includes a processor 20 , a memory 30 , an I / O unit 40 , and a communication unit 50 .

[0059] The processor 20 determines the merchant information according to the program and data read from the memory 30. The memory 30 may include one or more of a volatile memory (such as a random access memory) and a non-volatile memory (such as a read-only memory), and a network memory. The memory 30 stores a program for executing the method described in the present application and data (knowledge graph) to be used for executing the program. The memory 30 stores a program module, which is used to implement Figure 3The process of the method shown in FIG. 4 is a flow chart of the method shown in FIG. The program modules in the memory 30 can be distributed as a collection of various programming instructions in a computer program product. The I / O unit 40 may include various devices that interact with the processor 30 via an I / O interface, such as a display device, a keyboard for receiving user instructions, a touch screen, etc. The communication unit 50 provides communication with other data processing systems. The communication unit 50 can communicate through a physical channel and a wireless communication channel. The programs and data for practicing various embodiments of the present invention can be transmitted through the communication unit 50 and stored in the memory 30.

[0060] In some embodiments, a device for determining target information based on a knowledge graph is provided, such device comprising: a computer processor, a computer-readable storage medium, and program instructions stored on the computer-readable storage medium, wherein the program instructions perform the following steps when executed by the processor: a similarity calculation step, wherein, based on the TFIDF algorithm, the TFIDF value of the target name for each node entry is calculated as the similarity score between the node entry and the target, and a selection step, wherein node entries with similarity scores above a threshold are taken as candidate entries, and a determination step, wherein the target information is determined from the candidate entries.

[0061] In other embodiments, the processor may be further configured to perform: a word segmentation step, wherein a word segmentation operation is performed on the node description words and attribute values ​​in each node entry, a word segmentation operation is performed on the name of the target, and an inverted index establishment step, wherein, after the word segmentation step, an inverted index from subwords to node entries is established for the words that are the result of the word segmentation operation for each node entry. Optionally, in the similarity calculation step, based on the TFIDF algorithm, the TFIDF value of each subword of the name of the target for each node entry is calculated, and the sum of the TFIDF values ​​of each subword of the name of the target is used as the similarity score of the target relative to a node entry. Optionally, in the determination step, wherein a logistic regression function is used to calculate the estimated value of each candidate entry relative to the name of the target, and the information of the target is determined from the candidate entries based on the estimated value.

[0062] In other embodiments, the processor may be further configured to perform: a filtering step, in which node entries containing the relationship or attribute are filtered from the knowledge graph according to the relationship or attribute of the target to be determined.

[0063] Figure 6 is a block diagram showing a program module unit of an apparatus for determining target information based on a knowledge graph according to an embodiment. The apparatus may include various program module units, which are stored in Figure 5 In the memory 30 shown, each unit is used to complete the function in the method steps of the embodiment.

[0064] According to some embodiments, the device for determining target information based on the knowledge graph includes a word segmentation unit, which is used to perform word segmentation operations on the node description words and attribute values ​​in each node entry, and to perform word segmentation operations on the name of the target; a similarity calculation unit, which is used to calculate the TFIDF value of the name of the target for each node entry as the similarity score between the node entry and the target based on the TFIDF algorithm; a selection unit, which is used to take the node entries with similarity scores above the threshold as candidate entries, and a determination unit, which is used to determine the target information from the candidate entries. The similarity calculation unit can also be configured to calculate the TFIDF value of each subword of the name of the target for each node entry based on the TFIDF algorithm, and take the sum of the TFIDF values ​​of each subword of the name of the target as the similarity score of the target relative to a node entry. The determination unit can also be configured to use a logistic regression function to calculate the estimated value of each candidate entry relative to the name of the target, and determine the information of the target from the candidate entries based on the estimated value.

[0065] Optionally, the device may further include an inverted index establishment unit, which is used to establish an inverted index from subwords to node entries for each node entry after the word segmentation unit. Optionally, the device may further include a filtering unit, which is used to filter node entries containing the relationship or attribute of the target to be determined from the knowledge graph.

[0066] According to other embodiments, a computer-readable storage medium is provided having instructions located thereon that, when executed, cause a computing device to perform the steps of the methods described herein.

[0067] The description of one or more embodiments of the present invention is not intended to be exhaustive or to limit the scope of the embodiments to the precise forms disclosed or claimed herein. Modifications and variations are possible according to the above teachings, or can be obtained from the practice of various implementations of the various embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the present invention.

Claims

1. A method for determining target information based on a knowledge graph, characterized in that: The method comprises: A word segmentation step, performing a word segmentation operation on the node description words and attribute values ​​in each node entry, performing a word segmentation operation on the name of the target, and establishing an inverted index from the subword to the node entry for each word whose node entry is the result of the word segmentation operation, wherein the node entry is composed of three elements: two nodes and an edge between the two nodes, the three elements forming a triple, and the knowledge graph is a set of triples; A similarity calculation step, wherein, based on the TFIDF algorithm, the TFIDF value of the target name character for each triple node entry is calculated as the similarity score between the node entry and the target, and, A selection step, in which node entries with similarity scores above a threshold are selected as candidate entries, and a determination step, wherein target information is determined from candidate entries, The step of establishing an inverted index from a subword to a node entry for each word that is a result of a word segmentation operation on each node entry comprises: performing a word segmentation operation on each entity description word 1-n contained in each entity entry 1-n to obtain a subword 1-m for each entity entry, and then establishing an inverted index from each subword 1-m to the entity entry. In the similarity calculation step, the TFIDF value of the target name for each triple node entry is calculated by the following formula: Here M i is the name of the target, Tr i is a triple, and the subwords of the target name are obtained by segmenting the target name (M i1 , M i2 , …M in ), where N t (M ij ,Tr i ) represents the subword M ij In the triple Tr i The number of occurrences in, Ns(M ij ) indicates M ij The number of occurrences in all triples, |D| represents the total number of triples in the knowledge graph used, N d (M ij ) means it contains the subword M ij The number of triples is obtained by summing the TFIDF values ​​of each subword to obtain the target name and triple Tr i Here, k and b are smoothing coefficients, A logistic regression function is further used to calculate an estimated value of each candidate entry relative to the name of the target, and information about the target is determined from the candidate entries based on the estimated value to optimize matching and ranking of the candidate entries.

2. The method according to claim 1, characterized in that In the similarity calculation step, based on the TFIDF algorithm, the TFIDF value of each subword of the target name for each node entry is calculated, and the sum of the TFIDF values ​​of each subword of the target name is taken as the similarity score of the target relative to a node entry.

3. The method according to claim 1, characterized in that The method comprises: A filtering step, in which node entries containing the relationship or attribute of the target to be determined are filtered from the knowledge graph.

4. A device for determining target information based on a knowledge graph, characterized in that: The device comprises: Computer processors, a computer readable storage medium, The program instructions stored on the computer readable storage medium, when executed by the processor, perform the following steps: A word segmentation step, performing a word segmentation operation on the node description words and attribute values ​​in each node entry, performing a word segmentation operation on the name of the target, and establishing an inverted index from the subword to the node entry for each word whose node entry is the result of the word segmentation operation, wherein the node entry is composed of three elements: two nodes and an edge between the two nodes, the three elements forming a triple, and the knowledge graph is a set of triples; A similarity calculation step, wherein, based on the TFIDF algorithm, the TFIDF value of the node entry of each triple of the target name is calculated as the similarity score between the node entry and the target, and A selection step, in which node entries with similarity scores above a threshold are selected as candidate entries, and a determination step, wherein target information is determined from candidate entries, The step of establishing an inverted index from a subword to a node entry for each word that is a result of a word segmentation operation on each node entry comprises: performing a word segmentation operation on each entity description word 1-n contained in each entity entry 1-n to obtain a subword 1-m for each entity entry, and then establishing an inverted index from each subword 1-m to the entity entry. In the similarity calculation step, the TFIDF value of the target name for each triple node entry is calculated by the following formula: Here M i is the name of the target, Tr i is a triple, and the subwords of the target name are obtained by segmenting the target name (M i1 , M i2 , …M in ), where N t (M ij ,Tr i ) represents the subword M ij In the triple Tr i The number of occurrences in, Ns(M ij ) indicates M ij The number of occurrences in all triples, |D| represents the total number of triples in the knowledge graph used, N d (M ij ) means it contains the subword M ij The number of triples is obtained by summing the TFIDF values ​​of each subword to obtain the target name and triple Tr i Here, k and b are smoothing coefficients, A logistic regression function is further used to calculate an estimated value of each candidate entry relative to the name of the target, and information about the target is determined from the candidate entries based on the estimated value to optimize matching and ranking of the candidate entries.

5. The device according to claim 4, characterized in that In the similarity calculation step, based on the TFIDF algorithm, the TFIDF value of each subword of the target name for each node entry is calculated, and the sum of the TFIDF values ​​of each subword of the target name is taken as the similarity score of the target relative to a node entry.

6. The device according to claim 4, characterized in that The processor is configured to perform: A filtering step, in which node entries containing the relationship or attribute of the target to be determined are filtered from the knowledge graph.

7. A computer-readable storage medium having instructions located thereon, which when executed cause a computing device to perform the method according to any one of claims 1 to 3.

8. A computer program product, comprising a computer program, wherein when the computer program is executed, the method according to any one of claims 1 to 3 is implemented.

Images