Search

CN-121998067-A - Heterogeneous network reasoning method, device, equipment, storage medium and product

CN121998067ACN 121998067 ACN121998067 ACN 121998067ACN-121998067-A

Abstract

The application discloses a heterogeneous network reasoning method, a device, equipment, a storage medium and a product, which relate to the technical field of artificial intelligence, wherein the heterogeneous network reasoning method comprises the steps of determining superside information of an entity to be inferred according to a heterogeneous supergraph of the entity to be inferred when detecting that the entity to be inferred has association, wherein the heterogeneous supergraph is a graph structure comprising multiple types of entities and multiple types of supersides; searching in a preset search library based on the superside information to obtain candidate relation information between the entities to be inferred, wherein the preset search library comprises the relation between the entities in the heterogeneous supergraphs and the superside information corresponding to the entities, and inputting the candidate relation information into a preset text large model to obtain an entity relation reasoning result. Compared with the existing prediction of the correlation between the entities based on the triple single-hop information in the knowledge graph, the method provided by the application can improve the reasoning efficiency of the entity relationship by utilizing the retrieval enhancement and the prediction capability of the large model.

Inventors

  • LIU HUANYONG

Assignees

  • 北京奇虎科技有限公司

Dates

Publication Date
20260508
Application Date
20241101

Claims (10)

  1. 1. A heterogeneous network reasoning method, characterized in that the method comprises the steps of: when detecting that the relationship exists between the entities to be inferred, determining the superside information of the entities to be inferred according to the heterogeneous supergraph of the entities to be inferred, wherein the heterogeneous supergraph is a graph structure comprising multiple types of entities and multiple types of supersides; searching in a preset search library based on the superside information to obtain candidate relation information between the entities to be inferred, wherein the preset search library comprises the relation between the entities in the heterogeneous supergraphs and the superside information corresponding to the entities; and inputting the candidate relation information into a preset text big model to obtain an entity relation reasoning result.
  2. 2. The heterogeneous network reasoning method of claim 1, wherein before the step of retrieving in a preset retrieval library based on the superside information to obtain candidate relationship information between the entities to be reasoning, the method further comprises: counting relation information among entities in the heterogeneous hypergraph; traversing the relation information, determining a target entity corresponding to the traversed target relation, and determining entity superside information corresponding to the target entity according to the heterogeneous supergraph; and constructing a preset search library according to the entity superside information and the target relation, and returning to the step of traversing the relation information.
  3. 3. The heterogeneous network reasoning method of claim 2, wherein the step of constructing a preset search library according to the entity superside information and the target relationship comprises: Vectorizing the entity superside information to obtain an entity superside vector; and constructing a preset search library according to the entity superside vector and the target relation.
  4. 4. The heterogeneous network reasoning method of claim 1, wherein the step of retrieving in a preset retrieval library based on the superside information to obtain candidate relationship information between the entities to be inferred comprises: Searching in a preset search library based on the superside information to obtain similar entity superside information similar to the superside information; Ordering the similar entity superside information according to the similarity with the superside information to obtain target entity superside information; And determining candidate relation information between the entities to be inferred according to the target entity superside information and the preset search library.
  5. 5. The heterogeneous network reasoning method of claim 1, wherein when detecting that there is a correlation between entities to be reasoned, before the step of determining superside information of the entities to be reasoned according to a heterogeneous hypergraph in which the entities to be reasoned are located, the method further comprises: Inquiring path information among entities to be inferred in a heterogeneous hypergraph where the entities to be inferred are located; if the path information is not queried, querying first superside information of the entity to be inferred in the heterogeneous hypergraph; And determining whether the entity to be inferred is associated or not according to the first superside information and a preset text big model.
  6. 6. The heterogeneous network reasoning method of claim 5, wherein the step of determining whether there is an association between the entities to be reasoned according to the first superside information and a pre-set text big model comprises: Acquiring a preset prompting template; constructing an input template according to the preset prompting template and the first superside information; And inputting the input template into a preset text large model to obtain a judging result output by the preset text large model, wherein the judging result comprises whether the entity to be inferred has a correlation or not.
  7. 7. A heterogeneous network reasoning apparatus, characterized in that the heterogeneous network reasoning apparatus comprises: The determining module is used for determining the superside information of the entity to be inferred according to the heterogeneous supergraph of the entity to be inferred when the association between the entities to be inferred is detected, wherein the heterogeneous supergraph is a graph structure comprising multiple types of entities and multiple types of supersides; The search module is used for searching in a preset search library based on the superside information to obtain candidate relation information between the entities to be inferred, wherein the preset search library comprises the relation between the entities in the heterogeneous supergraph and the superside information corresponding to the entities; And the prediction module is used for inputting the candidate relation information into a preset text big model to obtain an entity relation reasoning result.
  8. 8. A heterogeneous network reasoning device, characterized in that the device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the heterogeneous network reasoning method as claimed in any of claims 1 to 6.
  9. 9. A storage medium, characterized in that the storage medium is a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the steps of the heterogeneous network reasoning method as claimed in any of the claims 1 to 6.
  10. 10. A computer program product, characterized in that the computer program product comprises a computer program which, when being executed by a processor, implements the steps of the heterogeneous network reasoning method of any of claims 1 to 6.

Description

Heterogeneous network reasoning method, device, equipment, storage medium and product Technical Field The application relates to the technical field of artificial intelligence, in particular to a heterogeneous network reasoning method, a heterogeneous network reasoning device, heterogeneous network reasoning equipment, a storage medium and a heterogeneous network reasoning product. Background The graph inference is mainly used for solving the inference task of the graph, for example, given two entities, whether there is an association (yes or no) between the two entities, what kind of association exists (given a certain association name) is needed to be determined, and these tasks are supported by relying on knowledge graph technology. However, the traditional knowledge graph form based on the triplet expression is insufficient in characterizing the entity, for example, for the correlation prediction between two entities, the correlation prediction result of the entity is not high in reliability due to fitting objective functions, such as TransE translation models, and whether the real information or the triplet information is used for single-hop. Disclosure of Invention The application mainly aims to provide a heterogeneous network reasoning method, a device, equipment, a storage medium and a product, which aim to solve the technical problem of low credibility of the existing entity association prediction result. In order to achieve the above purpose, the application provides a heterogeneous network reasoning method, which comprises the steps of determining superside information of an entity to be reasoning according to a heterogeneous supergraph of the entity to be reasoning when detecting that the entity to be reasoning has correlation, wherein the heterogeneous supergraph is a graph structure comprising multiple types of entities and multiple types of supersides; searching in a preset search library based on the superside information to obtain candidate relation information between the entities to be inferred, wherein the preset search library comprises the relation between the entities in the heterogeneous supergraphs and the superside information corresponding to the entities; and inputting the candidate relation information into a preset text big model to obtain an entity relation reasoning result. Optionally, before the step of retrieving in a preset retrieval library based on the superside information to obtain the candidate relationship information between the entities to be inferred, the method further includes: counting relation information among entities in the heterogeneous hypergraph; traversing the relation information, determining a target entity corresponding to the traversed target relation, and determining entity superside information corresponding to the target entity according to the heterogeneous supergraph; and constructing a preset search library according to the entity superside information and the target relation, and returning to the step of traversing the relation information. Optionally, the step of constructing a preset search library according to the entity superside information and the target relationship includes: Vectorizing the entity superside information to obtain an entity superside vector; and constructing a preset search library according to the entity superside vector and the target relation. Optionally, the step of retrieving in a preset retrieval library based on the superside information to obtain candidate relationship information between the entities to be inferred includes: Searching in a preset search library based on the superside information to obtain similar entity superside information similar to the superside information; Ordering the similar entity superside information according to the similarity with the superside information to obtain target entity superside information; And determining candidate relation information between the entities to be inferred according to the target entity superside information and the preset search library. Optionally, before the step of determining the superside information of the entity to be inferred according to the heterogeneous supergraph where the entity to be inferred is located when the association between the entities to be inferred is detected, the method further includes: Inquiring path information among entities to be inferred in a heterogeneous hypergraph where the entities to be inferred are located; if the path information is not queried, querying first superside information of the entity to be inferred in the heterogeneous hypergraph; And determining whether the entity to be inferred is associated or not according to the first superside information and a preset text big model. Optionally, the step of determining whether there is an association between the entities to be inferred according to the first superside information and a preset text big model includes: Acquiring a preset prompting template; constru