Atlas retrieval method and device, electronic equipment and storage medium

Abstract

The application provides a graph retrieval method and device, electronic equipment and a storage medium. The method comprises: obtaining question information requiring answer search; determining candidate entities corresponding to the question information and entity categories of the candidate entities; querying a candidate path containing at least one target node in a mode graph according to the entity categories, wherein the target node is a node used to indicate the entity category; generating a target query graph according to all candidate paths and candidate entities corresponding to the entity categories, wherein all target nodes are included in the target query graph; obtaining a query language by parsing the target query graph, and querying an answer corresponding to the question information based on the query language. The method in the embodiment overcomes the technical problems of high cost and poor generalization performance caused by manually defining templates for graph retrieval in the related art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data search, and in particular to a graph retrieval method and device, an electronic device and a storage medium. BACKGROUND

[0002] Graph retrieval refers to that, given a natural language question, an answer is obtained by performing semantic understanding and analysis on the question, and then performing query and reasoning on a knowledge base. Graph retrieval is a commonly used technical solution in search and intelligent question answering. The process of a method based on information retrieval is as follows: first, a plurality of candidate entities are obtained according to the question. Then, relationships connected with the candidate entities are extracted from the knowledge base as candidate query paths. Next, a text matching model is used to select a candidate query path with the highest similarity to the question. Finally, the answer is retrieved from the knowledge base. This method has high time consumption in this step because it needs to search from the database, and there are rich entities and relationships in the graph.

[0003] At present, the semantic analysis graph retrieval method based on related technologies uses a dictionary, rules and machine learning to directly analyze entities, relationships and logical combinations from the question, generates executable language, and directly searches from a graph database to obtain an answer. However, because the forms of the questions used for the analysis are various, the cost of manually defining templates for graph retrieval is high, and the generalization performance is poor.

[0004] Therefore, the related technologies have the problem that the cost of manually defining templates for graph retrieval is high, and the generalization performance is poor. SUMMARY

[0005] The present application provides a graph retrieval method and device, an electronic device and a storage medium to at least solve the problem that the cost of manually defining templates for graph retrieval is high, and the generalization performance is poor in the related technologies.

[0006] According to an aspect of an embodiment of the present application, a graph retrieval method is provided, comprising:

[0007] Obtaining question information requiring answer search;

[0008] Determining candidate entities corresponding to the question information, and entity categories of the candidate entities;

[0009] Querying candidate paths containing at least one target node in a schema graph according to the entity categories, wherein the target node is a node used to indicate the entity category;

[0010] Generating a target query graph according to all the candidate paths and the candidate entities corresponding to the entity categories, wherein the target query graph includes all the target nodes.

[0011] According to the target query graph analysis, a query language is obtained, and an answer corresponding to the problem information is queried based on the query language.

[0012] Optionally, in the method as described above, the querying, according to the entity category, for the candidate path containing the target node in the schema graph comprises:

[0013] The at least one candidate path containing the target node is queried by performing at least one round of traversal operation in the schema graph according to the entity category and a preset maximum jump number, wherein the maximum jump number indicates a number of nodes required to jump from the target node to a farthest node, the farthest node is a node in the candidate path that is most spaced apart from the target node by a number of other nodes, a number of jumps performed in each round of the traversal operation is less than or equal to the maximum jump number, and there is at least one different node between different candidate paths.

[0014] Optionally, in the method as described above, the querying, according to the entity category and the preset maximum jump number, for the at least one candidate path containing the target node in the schema graph by performing at least one round of traversal operation in the schema graph comprises:

[0015] The target node corresponding to the entity category is taken as a starting node of the traversal operation, and N times of jump operation are performed in the schema graph, wherein N is a value of the maximum jump number.

[0016] The following operations are repeatedly performed until N times of jump operation have been performed: in a case where a total number M of jump operations that have been performed in the N times of jump operation is less than N, and a post-jump node jumped to by an (M+1)th jump operation and a current node located at a time of an Mth jump operation satisfy a preset relationship, the post-jump node and the current node are taken as a complex entity, and it is determined that M times of jump operation have been performed, wherein M is an integer, and 0≤M<N, and the preset relationship comprises that there is a CVT node in the post-jump node and the current node, and the CVT node is used to describe a scenario of the post-jump node and another node in the current node other than the CVT node; and in a case where the total number M of jump operations that have been performed in the N times of jump operation is less than N, and the post-jump node jumped to by the (M+1)th jump operation and the current node located at the time of the Mth jump operation do not satisfy the preset relationship, it is determined that (M+1) times of jump operation have been performed.

[0017] In a case that N times of hopping operations have been performed, the starting node and nodes to which each of the N times of hopping operations hops are connected in a sequence of each of the N times of hopping operations, to obtain the candidate path.

[0018] Optionally, in the method as described above, the determining of the candidate entity corresponding to the question information and the entity category of the candidate entity comprises:

[0019] performing entity recognition on the question information to determine a first entity contained in the question information and an entity category of the first entity;

[0020] determining a second entity that satisfies a preset relevance with the first entity in a graph database according to the entity category, wherein a plurality of entities are preset in the graph database, and the entity category of the second entity is the entity category of the first entity;

[0021] determining all the candidate entities according to the first entity and the second entity, and determining an entity category corresponding to each of the candidate entities.

[0022] Optionally, in the method as described above, the generating of the target query graph according to all the candidate paths and the candidate entities corresponding to the entity categories comprises:

[0023] determining an association relationship between any two candidate nodes according to all the candidate paths, wherein the candidate nodes are nodes in the candidate paths;

[0024] determining attribute information corresponding to each candidate node according to the candidate entities corresponding to the entity categories;

[0025] obtaining a plurality of candidate query graphs according to the association relationship, the candidate nodes and the attribute information corresponding to each candidate node, wherein for each of the candidate query graphs, the candidate query graph includes a plurality of candidate nodes, the plurality of candidate nodes are connected according to the association relationship, and each of the candidate nodes has corresponding attribute information;

[0026] determining the target query graph with the highest matching degree with the question information from the plurality of candidate query graphs according to a preset matching degree algorithm.

[0027] Optionally, in the method as described above, the obtaining of the query language according to the target query graph and the obtaining of the answer corresponding to the question information based on the query language comprise:

[0028] determining a semantic expression structure according to each first target node in the target query graph and an association relationship between each first target node.

[0029] generating a query language according to the semantic expression structure;

[0030] querying in a graph database through the query language to obtain the answer.

[0031] Optionally, the method as described above, the determining the semantic expression structure according to each first target node in the target query graph and the association relationship between each first target node comprises:

[0032] determining a main chain and a branch chain in the target query graph;

[0033] determining all second target nodes on the main chain, attribute information of each second target node in all the second target nodes, and a main chain association relationship between any two second target nodes connected to each other, wherein all the first target nodes comprise second target nodes; determining all third target nodes on the branch chain, attribute information of each third target node in all the third target nodes, and a branch chain association relationship between any two third target nodes connected to each other, wherein there is a second target node on the main chain among all the third target nodes on the branch chain, and all the first target nodes comprise third target nodes;

[0034] generating the semantic expression structure according to the attribute information of the second target node, the main chain association relationship, the attribute information of the third target node, and the branch chain association relationship.

[0035] According to another aspect of the embodiments of the present application, a graph retrieval device is also provided, comprising:

[0036] an acquisition module configured to acquire question information requiring answer search;

[0037] a determination module configured to determine candidate entities corresponding to the question information and entity categories of the candidate entities;

[0038] a query module configured to query candidate paths containing at least one target node in a schema graph according to the entity categories, wherein the target node is a node used to indicate the entity category;

[0039] a generation module configured to generate a target query graph according to all the candidate paths and the candidate entities corresponding to the entity categories, wherein the target query graph comprises all the target nodes;

[0040] an answer determination module configured to parse a query language according to the target query graph and query an answer corresponding to the question information based on the query language.

[0041] According to a further aspect of the embodiments of the present application, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete the communication among each other through the communication bus; wherein the memory is configured to store a computer program; and the processor is configured to execute the method steps in any of the above embodiments by running the computer program stored in the memory.

[0042] According to a further aspect of the embodiments of the present application, a computer readable storage medium is provided, which stores a computer program, wherein the computer program is configured to execute the method steps in any of the above embodiments when running.

[0043] The present solution can be applied to optimization in the field of data search technology. The above technical solution provided by the embodiments of the present application has the following advantages compared with the prior art: by querying the candidate paths containing at least one target node in the schema graph according to the entity category, and generating the target query graph according to all candidate paths and the candidate entities corresponding to the entity category, an implementation manner of generating a query graph based on a schema graph (i.e., scheme graph) is provided, and then a query language can be generated based on the target query graph, and the answer corresponding to the question information is queried based on the query language, so that the answer can be queried without defining the target query graph in advance. Therefore, the embodiment overcomes the technical problems in the related art that the template for graph retrieval needs to be manually defined, which results in high cost and poor generalization performance. BRIEF DESCRIPTION OF DRAWINGS

[0044] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and serve to explain the principles of the present application together with the specification.

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, those skilled in the art can obtain other drawings from these drawings without creative labor.

[0046] Figure 1 is a flow diagram of an optional graph retrieval method according to an embodiment of the present application;

[0047] Figure 2 is a structural diagram of a schema graph in an example according to the related art;

[0048] Figure 3is a structural schematic diagram of an optional schema graph according to an embodiment of the present application;

[0049] Figure 4 is a flow schematic diagram of another optional method for determining candidate entities and candidate query graphs according to an embodiment of the present application;

[0050] Figure 5 is a conversion schematic diagram of an optional conversion of a target query graph into a semantic expression structure according to an embodiment of the present application;

[0051] Figure 6 is a structural schematic diagram of an optional data according to an embodiment of the present application;

[0052] Figure 7 is a structural schematic diagram of an optional semantic expression structure according to an embodiment of the present application;

[0053] Figure 8 is a block diagram of an optional graph retrieval device according to an embodiment of the present application;

[0054] Figure 9 is a structural block diagram of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the personnel in the art better understand the present application scheme, the technical scheme in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor should be within the scope of protection of the present application.

[0056] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0057] According to an aspect of the embodiments of the present application, a graph retrieval method is provided. Optionally, the graph retrieval method can be applied in a hardware environment composed of a terminal and a server in the embodiments. The server is connected with the terminal through a network and can provide services (such as advertisement pushing service, application service, etc.) for the terminal or a client installed on the terminal. A database can be set on the server or independently of the server to provide data storage service for the server.

[0058] The network can include, but is not limited to, at least one of a wired network and a wireless network. The wired network can include, but is not limited to, at least one of a wide area network, a metropolitan area network, and a local area network. The wireless network can include, but is not limited to, at least one of WIFI (Wireless Fidelity) and Bluetooth. The terminal can be, but is not limited to, a PC, a mobile phone, a tablet computer, etc.

[0059] The graph retrieval method of the embodiments of the present application can be executed by the server, or by the terminal, or by both the server and the terminal. The terminal can execute the graph retrieval method of the embodiments of the present application by a client installed thereon.

[0060] Taking the server as an example, Figure 1 The graph retrieval method provided by the embodiments of the present application includes the following steps:

[0061] In step S101, question information requiring answer search is obtained.

[0062] The graph retrieval method of the embodiments of the present application can be applied in a scenario where an answer corresponding to question information to be searched by a user is given, such as a scenario where an answer is given according to question sentence information, a scenario where an answer is given according to voice question information, etc. The graph retrieval method of the embodiments of the present application is described by taking the scenario where an answer is given according to question sentence information as an example. The graph retrieval method is also applicable to other types of question information without contradiction.

[0063] Taking the scenario where an answer is given according to question sentence information as an example, all question information is processed to determine the answer to the question to be queried by the question information.

[0064] When the user needs to search for a question, the user can input question sentence on an operable client to generate corresponding query (i.e., question information), and then send the query to the server through the client. Then, the server can obtain the question information requiring answer search.

[0065] Step S102, determine the candidate entity corresponding to the problem information, and the entity category of the candidate entity.

[0066] After the server obtains the problem information, the entity contained in the problem information can be questioned to obtain the candidate entity corresponding to the problem information.

[0067] The candidate entity can be an entity included in the problem information or other entities similar to the entity included in the problem information (satisfying a preset similarity, for example, the similarity between word vectors is greater than or equal to a preset value, and the method of calculating the similarity can be: Euclidean distance, cosine similarity, Jaccard distance, edit distance, etc.).

[0068] After determining each candidate entity, the entity category of the candidate entity can be determined. The entity category can be information indicating the category to which the candidate entity belongs. For example, the entity category can include but is not limited to: country, province, city, time, activity, etc.; when the candidate entity is "Nanjing City", the corresponding entity category is province, and when the candidate entity is "June 1st", the corresponding entity category is time.

[0069] Step S103, according to the entity category, query the candidate path containing at least one target node in the schema graph, wherein the target node is a node indicating the entity category.

[0070] After determining the entity category of each candidate entity, the query can be performed in all paths in the schema graph according to the entity category. The schema graph is a schema graph. The schema graph can include a plurality of preset paths, each preset path including a plurality of preset nodes, each two preset nodes being connected by a directed edge, indicating the relationship between the two preset nodes.

[0071] Each preset node in the preset path can be a node indicating a preset entity category. Therefore, by determining the target node indicating the entity category in all preset nodes, the preset path including the target node can be determined as the candidate path.

[0072] Step S104, according to all candidate paths and the candidate entity corresponding to the entity category, generate a target query graph, wherein the target query graph includes all target nodes.

[0073] After determining the candidate path, all candidate paths can be integrated to obtain the target query graph. So that the target query graph obtained can include all target nodes.

[0074] And, in order to enable the later query based on the target query graph, it is also required to record the candidate entity corresponding to each entity category in the target query graph, which is equivalent to determining the attribute information corresponding to each node.

[0075] In general, the target query graph contains other nodes in addition to all target nodes, and the entity category corresponding to the other nodes can be identified as the entity category of the answer that the user wants to obtain.

[0076] Step S105, the query language is parsed according to the target query graph, and the answer corresponding to the question information is queried based on the query language.

[0077] After obtaining the target query graph, the candidate entity corresponding to each entity category and the association relationship between the entity categories can be obtained by parsing the target query graph, and then the query language can be obtained.

[0078] After obtaining the query language, the corresponding graph database can be searched based on the query language, and the search result is the answer corresponding to the question information.

[0079] The method in the embodiment provides an implementation manner of generating a query graph based on a scheme graph (i.e., scheme graph) by querying a candidate path containing at least one target node in a mode graph according to an entity category, and generating a target query graph according to all candidate paths and candidate entities corresponding to the entity categories, and then a query language can be generated based on the target query graph, and the answer corresponding to the question information can be queried based on the query language, so that the answer can be queried without defining the target query graph in advance. Therefore, the embodiment overcomes the technical problems of high cost and poor generalization performance caused by manually defining templates for graph retrieval in the related art.

[0080] As an optional embodiment, the step S103 of querying a candidate path containing at least one target node in the mode graph according to the entity category includes the following steps:

[0081] Step S201, at least one round of traversal operation is performed in the mode graph according to the entity category and a preset maximum jump number, to query at least one candidate path containing the target node, wherein the maximum jump number is used to indicate the number of nodes required to jump from the target node to the farthest node, the farthest node is a node in the candidate path that is separated from the target node by the largest number of other nodes, the number of jumps performed in each round of traversal operation is less than or equal to the maximum jump number, and there is at least one different node between different candidate paths.

[0082] In order to avoid too many irrelevant candidate paths and other nodes with low relevance, the maximum number of hops can be preset to ensure that the nodes in the candidate paths have high relevance to the target node. In addition, the nodes in the maximum number of hops that are used to indicate the number of nodes that need to be jumped from the target node to the farthest node can be limited to nodes other than the CVT node.

[0083] The maximum number of hops can be 2 or 3, etc., which can be selected according to the actual application scenario.

[0084] After determining the entity category and the maximum number of hops, the optional path containing the target node corresponding to the entity category can be found in the schema graph. Then, the target node is taken as the starting node, and the hops between different nodes in the optional path are performed according to the maximum number of hops. The node after each hop is recorded. After performing the maximum number of hops, the target node and the node reached by each hop are connected to obtain the candidate path.

[0085] In general, at least one node in any two candidate paths in all candidate paths is different.

[0086] For example, in the case of a maximum number of hops of 2 and a target node of a province-city node, after performing the first hop, the node is a time node, and after performing the second hop, the node is a brand node. The candidate path obtained can be a province-city node-time node-brand node.

[0087] As an optional embodiment, the method described above, step S201 performs at least one round of traversal operation in the schema graph according to the entity category and the preset maximum number of hops, and queries at least one candidate path containing the target node, including the following steps:

[0088] Step S301 takes the target node corresponding to the entity category as the starting node of the traversal operation, and performs N times of hop operation in the schema graph, where N is the value of the maximum number of hops.

[0089] After determining the entity category of the candidate entity and the target node corresponding to the entity category, the N times of hop operation can be performed in the schema graph with the target node as the starting node.

[0090] Step S302, loop and perform the following operations until N jump operations have been executed: When the total number M of jump operations already executed in the N jump operations is less than N, and the jumped-to node after the (M + 1)-th jump operation and the current node where the M-th jump operation is located satisfy a preset relationship, use the jumped-to node and the current node as a complex entity, and determine that M jump operations have been executed, where M is an integer and 0 ≤ M < N, and the preset relationship includes: There is a CVT node in the jumped-to node and the current node, and the CVT node is used to describe the scenario of the other node in the jumped-to node and the current node except the CVT node; When the total number M of jump operations already executed in the N jump operations is less than N, and the jumped-to node after the (M + 1)-th jump operation and the current node where the M-th jump operation is located do not satisfy the preset relationship, determine that (M + 1) jump operations have been executed.

[0091] Before each jump, the total number of jump operations already executed in this round of traversal operations can be determined, so M can be determined, and when M is less than N, the next jump operation can be continued.

[0092] When it is determined that the jumped-to node after the (M + 1)-th jump operation and the current node where the M-th jump operation is located satisfy the preset relationship, that is, there is a CVT node in the jumped-to node and the current node, and the CVT node is a node used to associate other nodes constituting a certain scenario to the other remaining node in the jumped-to node and the current node except the CVT node.

[0093] Such as Figure 3 The CVT - activity mechanism node shown is a CVT node in one scenario, and the activity node is a node that satisfies the preset relationship with the CVT - activity mechanism node, and the time node, province - city node, food node, and coupon node therein are the other nodes constituting the scenario; Therefore, the activity node and the CVT - activity mechanism node can form a complex entity (as shown by the gray area including the activity node and the CVT - activity mechanism node in Figure 3 . Among them, the activity node is the real part of the complex entity, and the CVT activity - mechanism node is the virtual part of the complex entity. And when jumping from the activity node to the CVT - activity mechanism node, this jump operation is not recorded in the N jump operations. That is, assuming that the total number of jump operations already executed before jumping to the CVT - activity mechanism node is M times, after jumping to the CVT - activity mechanism node, the total number of jump operations already executed is still recorded as M times.

[0094] Conversely, when the jump-after node jumped to by the M+1th jump operation does not satisfy the preset relationship with the current node located at when the Mth jump operation, it is determined that M+1 jump operations have been performed. And when M+1 is less than N, step S302 is continuously executed; conversely, it is jumped to step S303.

[0095] Step S303, when N jump operations have been performed, the starting node and the node jumped to by each jump operation in the N jump operations are connected according to the order of each jump operation in the N jump operations, to obtain a candidate path.

[0096] When N jump operations have been performed, the node jumped to by each jump operation can be determined, and then the nodes jumped to by each jump operation are connected to each other according to the order of each jump operation in the N jump operations, to obtain a candidate path.

[0097] The method in the embodiment can effectively reduce the distance between each node in the presence of a CVT node by determining the jump-after node and the current node satisfying the preset relationship and determining the jump-after node and the current node as a complex entity. Moreover, when the preset relationship is satisfied, the jump operation is not recorded in the maximum jump number, so that the path under the presence of a CVT node can be screened even when the maximum jump number is reduced once. Since the jump operation is not recorded in the maximum jump number only when the preset relationship is satisfied, the number of candidate paths under the absence of a CVT node can also be effectively reduced.

[0098] As an optional embodiment, the step S102 of determining the candidate entity corresponding to the question information and the entity category of the candidate entity includes the following steps:

[0099] Step S401, performing entity recognition on the question information to determine a first entity contained in the question information and an entity category of the first entity.

[0100] After obtaining the question information, entity recognition can be directly performed on the question information to determine a first entity in the question information and determine an entity category of the first entity.

[0101] For example, when the question information is “What is the discount of Crazy Thursday in Nanjing?”, entity recognition can be performed to obtain the first entities: Nanjing and Crazy Thursday; and determine that the entity category of Nanjing is a province and city, and the entity category of Crazy Thursday is an activity.

[0102] Step S402, determining a second entity satisfying a preset correlation with the first entity according to the entity category in the graph database, wherein a plurality of entities are preset in the graph database, and the entity category of the second entity is the entity category of the first entity.

[0103] After the first entity is determined, the first entity can be entity linked, and by determining the relevance information of each preset entity in the graph database and the first entity, a second entity meeting a preset relevance with the first entity is determined.

[0104] Optionally, the method of determining the relevance information can include, but is not limited to, Euclidean distance, cosine similarity, Jaccard distance, edit distance, and the like.

[0105] The preset relevance can be a preset relevance threshold indicating that the second entity can be set according to an actual application scenario, which is not limited herein.

[0106] And the entity category of the second entity meeting the preset relevance with the first entity is the entity category of the first entity.

[0107] For example, when the first entity is "Nanjing" (the entity category is "province / city"), the second entity obtained through entity linking can be "Nanjing City" (the entity category is "province / city"); when the first entity is "Crazy Thursday" (the entity category is "event"), the second entity obtained through entity linking can be "XXX Crazy Thursday in June" (the entity category is "event"), and the second entity can be "YYY Crazy Thursday" (the entity category is "event").

[0108] In step S402, all candidate entities are determined according to the first entity and the second entity, and the entity category corresponding to each candidate entity is determined.

[0109] After the first entity and the second entity are determined, all the first entities and all the second entities can be used as all candidate entities, and the entity category of each candidate entity is determined.

[0110] Through the method in this embodiment, candidate entities can be determined in multiple ways, and the entity category corresponding to each candidate entity is obtained.

[0111] As an optional embodiment, the method as described above, the step S104 generates a target query graph according to all candidate paths and the candidate entities corresponding to the entity category, including the following steps:

[0112] In step S501, the association relationship between any two candidate nodes is determined according to all candidate paths, wherein the candidate nodes are nodes in the candidate paths.

[0113] After all candidate paths are determined, since each candidate path includes multiple candidate nodes, the association relationship between any two candidate nodes can be determined.

[0114] Step S502, according to the candidate entity corresponding to the entity category, the attribute information of each node is determined.

[0115] After determining the specified candidate query graph, since the entity category of the candidate entity has been determined in the foregoing embodiment, the candidate entity corresponding to the entity category can be determined, and each node in the candidate query graph has a corresponding entity category, and then the attribute information of the corresponding node can be obtained based on the candidate entity, for example, the value (attribute information includes value) of the activity node is the candidate entity whose entity category is activity, and the value of the province and city node is the candidate entity whose entity category is province and city.

[0116] Step S503, according to the association relationship, the candidate node and the attribute information corresponding to each candidate node, a plurality of candidate query graphs are obtained, wherein for each candidate query graph, the plurality of candidate nodes are included in the query graph, the plurality of candidate nodes are connected according to the association relationship, and each candidate node has corresponding attribute information;

[0117] After determining the association relationship, a plurality of candidate nodes in all candidate nodes can be connected according to the association relationship to obtain a plurality of candidate query graphs, wherein each candidate query graph includes at least one candidate node other than the CVT node in addition to all target nodes. The attribute information of the candidate node in the candidate query graph is the original attribute information of the candidate node.

[0118] For example, when the association relationship 1: province and city node-CVT node-time node, activity node-CVT node-time node is obtained, the attribute information of the province and city node is "the province and city is Nanjing city", the attribute information of the time node is "June 1", and the attribute information of the activity node is "crazy Thursday", a candidate query graph A can be obtained, and in the candidate query graph A, the province and city node, the time node and the activity node are connected to the CVT node, and the attribute information of the province and city node in the candidate query graph A is still "the province and city is Nanjing city", the attribute information of the time node is still "6 / 1", and the attribute information of the activity node is still "crazy Thursday".

[0119] Step S504, according to a preset matching degree algorithm, a specified candidate query graph with the highest matching degree with the problem information is determined from the plurality of candidate query graphs.

[0120] After obtaining the plurality of candidate query graphs, the matching degree between each candidate query graph and the problem information can be determined according to a preset matching degree algorithm. Optionally, the preset matching degree algorithm includes but is not limited to heuristic rules and text similarity algorithms.

[0121] After determining the matching degrees corresponding to each candidate query graph, the target query graph with the highest matching degree to the question information can be determined from the plurality of candidate query graphs.

[0122] As an optional embodiment, the step S105 according to the target query graph analyzes the query language, and queries the answer corresponding to the question information based on the query language, including the following steps:

[0123] In step S601, the semantic expression structure is determined according to each first target node in the target query graph and the association relationship between each first target node.

[0124] After determining the target query graph as shown in Figure 5 , in order to generate the query language for querying, the semantic expression structure as shown in Figure 7 may be generated according to the following steps.

[0125] As an optional embodiment, the step S601 according to the target query graph analyzes the query language, and queries the answer corresponding to the question information based on the query language, including the following steps:

[0126] In step S701, the main chain and the branch chain are determined in the target query graph.

[0127] After determining the target query graph, the main chain and the branch chain can be determined in the target query graph in a preset manner.

[0128] For example, the longest chain in the target query graph is determined as the main chain, and the remaining chains connected to the main chain are determined as the branch chain. The chain including a preset number of nodes in the target query graph can also be randomly determined as the main chain, and the remaining chains connected to the main chain are determined as the branch chain.

[0129] In step S702, all second target nodes on the main chain, the attribute information of each second target node in all second target nodes, and the main chain association relationship between any two mutually connected second target nodes are determined, wherein all first target nodes include second target nodes; all third target nodes on the branch chain, the attribute information of each third target node in all third target nodes, and the branch chain association relationship between any two mutually connected third target nodes are determined, wherein there is a second target node on the main chain among all third target nodes on the branch chain, and all first target nodes include third target nodes.

[0130] After the main chain and the branch chain are determined, all the nodes on the main chain are determined, and the second target node is obtained. The branch chain connected to the main chain is determined, and the second target node on the branch chain is determined. For each branch chain, one of the second target nodes exists in all the third target nodes on the branch chain. As shown in Figure 5 the active node, the CVT node, and the time node constitute the main chain, and are the second target nodes on the main chain respectively. The province-city node and the CVT node constitute the branch chain and are connected to the main chain, so the province-city node is the third target node, and the CVT node is also the third target node. Therefore, the CVT node is both the second target node and the third target node.

[0131] After the second target node is determined, the main chain association relationship between any two connected second target nodes can be determined. For example, as shown in Figure 5 the main chain association relationship between the active node and the CVT node is the activity mechanism indicated by the connected edge, the main chain association relationship between the CVT node and the time node is the activity mechanism time indicated by the connected edge, and the branch chain association relationship between the province-city node and the CVT node is the activity mechanism province-city indicated by the connected edge.

[0132] In step S703, the semantic expression structure is generated according to the attribute information of the second target node, the main chain association relationship, the attribute information of the third target node, and the branch chain association relationship.

[0133] After the attribute information of the second target node, the main chain association relationship, the attribute information of the third target node, and the branch chain association relationship are determined, the semantic expression structure is generated based on the attribute information of the second target node, the main chain association relationship, the attribute information of the third target node, and the branch chain association relationship.

[0134] For example, the entity and the related data on the branch chain are stored by data, and the edge (i.e., the main chain association relationship) on the main chain is stored by edge. The structure of data is shown in Figure 6 Each term is explained as follows: tag-entity type, properties-all attributes of the entity, field-attribute field, value-attribute value (i.e., candidate entity), sign-symbol (which can include but is not limited to: greater than, less than, belongs to, contains, equal to, not equal to, maximum, minimum, total amount), relations-all relationships of the connected entity, edge tag-relationship type, intent field-intent field, which represents the attribute of the last node on the main chain. The attribute information can include properties-all attributes of the entity, field-attribute field, value-attribute value, and sign-symbol.

[0135] Then, based on the structure of each data and the backbone association relationship indicated by the edge between the data, a semantic expression structure as shown in Figure 7 is obtained.

[0136] Step S602: generating a query language according to the semantic expression structure.

[0137] After the semantic expression structure is determined, a corresponding query language can be generated based on the semantic expression structure. Optionally, the method for generating the query language includes one or more of the following methods:

[0138] Gremlin-a graph search language originally developed for the Apache Tinkerpop project, which can implement procedural or declarative queries;

[0139] Cypher-a declarative language originally created by Neo4J and later popularized in the form of OpenCypher, which can be used to search for vertices and edges matching specific properties;

[0140] GQL-a standard initiative to unify Cyher, GSQL, and PSQL;

[0141] SPARQL-a standard established for querying RDF format knowledge graphs;

[0142] PGQL-a language originally developed by Oracle for searching and collecting information from standardized vertices;

[0143] GSQL-a procedural language originally developed by TigerGraph;

[0144] AQL-a procedural language originally developed by ArangoDB.

[0145] Further, based on the type of query language supported by the graph database, a corresponding query language generation method can be used to generate a query language corresponding to the type of query language.

[0146] Step S603: querying the graph database through the query language to obtain an answer.

[0147] After the query language is generated, the graph database can be queried based on the query language to obtain an answer.

[0148] For example, based on the examples given in the aforementioned embodiments, when the question information is "What are the special offers for Crazy Thursday in Nanjing?" and the obtained target query graph includes provincial and municipal nodes, time nodes, activity nodes, and CVT nodes, the query language used for time queries can be determined, and the resulting answer is also used to indicate the time of the "Crazy Thursday" activity in "Nanjing", for example, "June 1st to June 30th".

[0149] The following is an example of applying any of the foregoing embodiments:

[0150] 1.1 Constructing the schema graph (i.e., the pattern graph)

[0151] In graph databases, a schema represents the links between various types of entities, as well as the attribute fields carried by the entities themselves. Storing the links between different entities using a graph structure results in a schema graph. Figure 2 This is a partial view of the retail schema graph. Circles represent entities, and the text inside the circles indicates the type of that entity. Lines represent relationships between entities. The "CVT-Activity Mechanism" is the CVT node for "Activities".

[0152] Depend on Figure 3 It is known that due to the introduction of CVT nodes, the distance from entities such as "time" and "province / city" to the activity changes from 1 hop to 2 hops, and the distance to the brand changes from 2 hops to 3 hops. If the path for generating the candidate query graph increases from 2 hops to 3 hops, the number of candidate query graphs will increase exponentially. To solve this problem, this application sets up a complex entity. That is, there exists a complex entity, where the real part of the node corresponding to the complex entity is the entity node, and the imaginary part is the CVT entity node it connects to. For example... Figure 3 The schema graph after introducing complex entities is shown, where gray represents complex entities. The real part of a complex entity represents "Activity," and the imaginary part represents "CVT-Activity Mechanism." Entities connected to the real and imaginary parts are all connected to complex entity nodes. Therefore, the distance from "Brand" to "Province / City" changes from 3 hops to 2 hops, effectively reducing the number of candidate paths.

[0153] 1.2 Obtaining Candidate Entities

[0154] Obtaining candidate entities involves two parts. First, a pre-defined model is used to identify the first entity mentioned in the query (i.e., the question information) and assign it an entity label (i.e., the entity category of the first entity). Then, entity links are performed on the mentioned first entity to obtain its corresponding second entity in the graph database. For example... Figure 4Step 1 in FIG. 1. First, the first entity contained in the query and the corresponding entity label (entity: entity label) are extracted by the entity recognition model: (Nanjing: province and city), (Crazy Thursday: activity). Then, the above first entity is mapped to the second entity in the graph database by entity linking related technology. For example, "Nanjing" corresponds to the second entity "Nanjing Province" in the graph database. The second entity corresponding to "Crazy Thursday" in the graph database includes: "XXX June Crazy Thursday" and "YYY Crazy Thursday activity".

[0155] 1.3 Generating query graph

[0156] The entity category of the candidate entity is the entity in the schema graph. Taking the candidate entity as the core entity, the path within two hops (i.e., the maximum number of hops is 2) in the schema graph is queried. A number of candidate paths are generated. Except for the core entity, other entities on the path are bundled to form a candidate query graph. For example, Figure 4 Step 2 in FIG. 1. Taking the types of entities, province and city, and activity as core entities, respectively, the related paths within two hops in the schema graph are obtained as the main chain of the query graph. For each series of related paths, a query graph is generated. For example, "(activity)-(CVT)-(time)" is a path in the related path generated by taking the activity as the core entity (i.e., the activity node as the starting node). According to the schema graph, it can be known that the candidate entity type province and city can be connected to the CVT node. Therefore, the CVT node on the path is stretched to link to the province entity. The value of the activity in the candidate query graph is the candidate entity of the type activity. The value of the province is the candidate entity of the type province. Then, the similarity between the candidate query graph and the problem information is calculated based on the heuristic rules and the text similarity algorithm, and the target query graph with the highest similarity (i.e., the highest similarity) is taken to generate the query language.

[0157] 1.4 Generating query language

[0158] The above target query graph needs to be translated into an executable language suitable for a specific graph database to search for an answer in the graph database. In order to facilitate the translation into the query language of different graph databases, the present application designs a query graph language expression system as an intermediate structure of the query graph to the executable language.

[0159] The query graph is parsed in the order of the main chain of the query graph and stored in a certain structure. For example, the target query graph in section 1.3 can be stored as a structure indicated by data and edge in Figure 5 . The entity and the related data on the branch are stored by data, and the edge is stored by edge. The structure of data is as follows: Figure 6The shown. Each explanation: tag-entity type, properties-entity all attributes, field-attribute field, value-attribute value, sign-symbol (including: greater than, less than, belong to, contain, equal, not equal, maximum, minimum, total), relations-all entity connections, edge tag-relation type, intent field-intention field, indicating the attribute of the last node on the main chain. Figure 7 For Figure 5 The semantic expression structure of the query graph generation.

[0160] Then generate an executable query language according to the semantic expression structure of the target query graph. Search in the corresponding graph database, and the search result is the answer to the query.

[0161] It should be noted that for the foregoing method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited by the action order described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily necessary for the present application.

[0162] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be realized by means of software and the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk, an optical disk), and includes a plurality of instructions for making a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) execute the method described in each embodiment of the present application.

[0163] According to another aspect of the embodiments of the present application, a graph retrieval device for implementing the above graph retrieval method is also provided. Figure 8 The structure block diagram of an optional graph retrieval device according to an embodiment of the present application is shown in FIG. 1, which can include: Figure 8 As shown in the figure, the device can include:

[0164] The acquisition module 1 is used to acquire problem information that needs to be searched for an answer;

[0165] A determining module 2 is configured to determine a candidate entity corresponding to the problem information and an entity category of the candidate entity.

[0166] A querying module 3 is configured to query, according to the entity category, a candidate path containing at least one target node in the schema graph, where the target node is a node used to indicate the entity category.

[0167] A generating module 4 is configured to generate a target query graph according to all the candidate paths and the candidate entity corresponding to the entity category, where the target query graph includes all the target nodes.

[0168] An answer determining module 5 is configured to parse the target query graph to obtain a query language and query an answer corresponding to the problem information based on the query language.

[0169] It should be noted that the obtaining module 1 in this embodiment can be configured to perform the step S101, the determining module 2 in this embodiment can be configured to perform the step S102, the querying module 3 in this embodiment can be configured to perform the step S103, the generating module 4 in this embodiment can be configured to perform the step S104, and the answer determining module 5 in this embodiment can be configured to perform the step S105.

[0170] It should be noted that the modules described above can be implemented in the hardware environment shown in FIG. 15 as a part of the apparatus, and can be implemented by software or hardware, where the hardware environment includes a network environment. Figure 1

[0171] According to another aspect of the embodiments of the present application, an electronic device for implementing the above-mentioned schema retrieval method is provided, which can be a server, a terminal, or a combination thereof.

[0172] According to another embodiment of the present application, an electronic device is provided, which includes the above-mentioned schema retrieval method. Figure 9 As shown in FIG. 15, the electronic device can include a processor 1501, a communication interface 1502, a memory 1503, and a communication bus 1504, where the processor 1501, the communication interface 1502, and the memory 1503 can communicate with each other through the communication bus 1504.

[0173] The memory 1503 is configured to store a computer program.

[0174] The processor 1501 is configured to execute the program stored in the memory 1503 to implement the following steps.

[0175] In step S101, problem information requiring answer search is obtained.​

[0176] In step S102, the candidate entity corresponding to the problem information is determined, and the entity category of the candidate entity is determined.

[0177] In step S103, a candidate path containing at least one target node is queried in the schema graph according to the entity category, where the target node is a node used to indicate the entity category.

[0178] In step S104, a target query graph is generated according to all candidate paths and the candidate entity corresponding to the entity category, where the target query graph contains all target nodes.

[0179] In step S105, a query language is parsed according to the target query graph, and an answer corresponding to the problem information is queried based on the query language.

[0180] Optionally, in the embodiment, the communication bus can be a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The communication bus can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus. The communication interface is used for communication between the electronic device and other devices.

[0181] The memory can include a random access memory (RAM) and can also include a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0182] The processor can be a general-purpose processor, which can include but is not limited to a CPU (Central Processing Unit), an NP (Network Processor), and the like; and can also be a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0183] The embodiment of the present application further provides a computer readable storage medium, the storage medium comprising a stored program, wherein the program performs the method steps of the method embodiment when running.

[0184] Optionally, in the embodiment, the storage medium can include but is not limited to a U disk, a ROM, a RAM, a mobile hard disk, a magnetic disk or an optical disk and various storage program codes.

[0185] The serial numbers of the embodiments of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0186] The integrated units in the above embodiments, if realized in the form of software function units and sold or used as independent products, can be stored in the above computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or the whole or part of the technical solutions can be embodied in the form of software products, and the computer software products are stored in the storage medium, including a plurality of instructions for causing one or more computer devices (which can be personal computers, servers or network devices, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application.

[0187] In the above embodiments of the present application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0188] In the several embodiments provided by the present application, it should be understood that the disclosed client can be implemented in other ways. Of course, the above device embodiment is only illustrative, for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed units can be indirect coupling or communication connection through some interfaces, units or modules, and can be electrical or other forms.

[0189] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place or distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the scheme provided in the embodiment.

[0190] In addition, each of the functional units in the various embodiments of the present application can be integrated in one processing unit, or each of the units can exist physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0191] The above only describes the preferred embodiments of the present application. It should be noted that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A map retrieval method characterized by comprising: The method comprises the following steps: acquiring question information requiring an answer search; determining candidate entities corresponding to the question information and entity categories of the candidate entities; querying a candidate path containing at least one target node in a schema graph according to the entity categories, wherein the target node is a node used to indicate the entity category; generating a target query graph according to all the candidate paths and the candidate entities corresponding to the entity categories, wherein the target query graph contains all the target nodes; parsing a query language according to the target query graph and querying an answer corresponding to the question information based on the query language, which comprises the following steps: determining a semantic expression structure according to each first target node in the target query graph and an association relationship between each first target node, which comprises the following steps: determining a main chain and a branch chain in the target query graph; determining all second target nodes on the main chain, attribute information of each second target node in the all second target nodes, and a main chain association relationship between any two second target nodes connected to each other, wherein all the first target nodes include the second target nodes; determining all third target nodes on the branch chain, attribute information of each third target node in the all third target nodes, and a branch chain association relationship between any two third target nodes connected to each other, wherein there is one second target node on the main chain in the all third target nodes on the branch chain, and all the first target nodes include the third target nodes; and generating the semantic expression structure according to the attribute information of the second target nodes, the main chain association relationship, the attribute information of the third target nodes and the branch chain association relationship; generating a query language according to the semantic expression structure; querying in a graph database through the query language to obtain the answer.

2. The method of claim 1, wherein, The step of querying a candidate path containing at least one target node in a schema graph according to the entity categories comprises the following steps: performing at least one round of traversal operation in the schema graph according to the entity categories and a preset maximum jump number to query at least one candidate path containing the target node, wherein the maximum jump number is used to indicate a number of nodes required to jump from the target node to a farthest node, the farthest node is a node in the candidate path which is most spaced from the target node, a jump number performed in each round of the traversal operation is less than or equal to the maximum jump number, and there is at least one different node between different candidate paths.

3. The method of claim 2, wherein, The step of performing at least one round of traversal operation in the schema graph according to the entity categories and a preset maximum jump number to query at least one candidate path containing the target node comprises the following steps: taking a target node corresponding to the entity category as a starting node of the traversal operation to perform N times of jump operation in the schema graph, wherein N is a numerical value of the maximum jump number. The following operations are performed in cycles until N times of jump operations have been performed: if the total number M of jump operations that have been performed in the N times of jump operations is less than N times, and a preset relationship is met between a post-jump node to which the M+1th jump operation jumps and a current node located at the time of the Mth jump operation, the post-jump node and the current node are taken as a complex entity, and it is determined that M times of jump operations have been performed, where M is an integer, and 0≤M<N, and the preset relationship includes that one of the post-jump node and the current node is a CVT node, and the CVT node is used to describe a scenario of the post-jump node and another node of the current node except the CVT node; if the total number M of jump operations that have been performed in the N times of jump operations is less than N times, and the preset relationship is not met between the post-jump node to which the M+1th jump operation jumps and the current node located at the time of the Mth jump operation, it is determined that M+1 times of jump operations have been performed; In the case that N times of jump operations have been performed, the starting node and the nodes to which each of the N times of jump operations jumps are connected in the order of the jump operations, to obtain the candidate path.

4. The method of claim 1, wherein, The determining of the candidate entity corresponding to the question information and the entity category of the candidate entity includes: performing entity recognition on the question information to determine a first entity contained in the question information and an entity category of the first entity; determining a second entity that satisfies a preset correlation with the first entity in a graph database according to the entity category, where a plurality of entities are preset in the graph database, and the entity category of the second entity is the entity category of the first entity; determining all the candidate entities according to the first entity and the second entity, and determining an entity category corresponding to each of the candidate entities.

5. The method of claim 1, wherein, The generating of the target query graph according to all the candidate paths and the candidate entities corresponding to the entity categories includes: determining an association relationship between any two candidate nodes according to all the candidate paths, where the candidate nodes are nodes in the candidate paths; determining attribute information corresponding to each candidate node according to the candidate entities corresponding to the entity categories; obtaining a plurality of candidate query graphs according to the association relationship, the candidate nodes, and the attribute information corresponding to each candidate node, where for each candidate query graph, the candidate query graph includes a plurality of candidate nodes, the plurality of candidate nodes are connected according to the association relationship, and each candidate node has corresponding attribute information; determining the target query graph with the highest matching degree to the question information from the plurality of candidate query graphs according to a preset matching degree algorithm.

6. A map search device characterized by comprising: It includes: an acquisition module configured to acquire question information requiring answer search; a determination module configured to determine candidate entities corresponding to the question information and entity categories of the candidate entities; The query module is configured to query, according to the entity category, a candidate path containing at least one target node in the schema graph, wherein the target node is a node used to indicate the entity category. The generation module is configured to generate a target query graph according to all the candidate paths and the candidate entity corresponding to the entity category, wherein the target query graph includes all the target nodes. The answer determination module is configured to parse a query language according to the target query graph and query an answer corresponding to the question information based on the query language, including: determining a semantic expression structure according to each first target node in the target query graph and an association relationship between the first target nodes; determining a main chain and a branch chain in the target query graph; determining all second target nodes on the main chain, attribute information of each second target node in the all second target nodes, and a main chain association relationship between any two mutually connected second target nodes, wherein all the first target nodes include the second target nodes; determining all third target nodes on the branch chain, attribute information of each third target node in the all third target nodes, and a branch chain association relationship between any two mutually connected third target nodes, wherein the all third target nodes on the branch chain include one second target node on the main chain, and all the first target nodes include the third target nodes; generating the semantic expression structure according to the attribute information of the second target nodes, the main chain association relationship, the attribute information of the third target nodes, and the branch chain association relationship; generating the query language according to the semantic expression structure; and querying in a graph database through the query language to obtain the answer.

7. An electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein, The processor, the communication interface, and the memory complete communication with each other through the communication bus, and the method comprises the following steps: The memory is configured to store a computer program. The processor is configured to execute the method steps of any one of claims 1 to 5 by running the computer program stored on the memory.

8. A computer readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is configured to execute the method steps in any one of claims 1 to 5 when running. The storage medium stores a computer program, and the computer program is configured to execute the method steps in any one of claims 1 to 5 when running.

Images