Knowledge graph embedded prediction model-based interpretability method and system

Abstract

The invention discloses an interpretability method and system based on a knowledge graph embedded prediction model. The method includes: acquiring a question (including an entity x and a relation r) from a user, inputting the question into the trained knowledge graph and embedding the knowledge graph into the prediction model to obtain an answer of the question, namely a prediction result; analyzing the entity x in the question input by the user and the obtained prediction result by using a sub-graph extraction tool to obtain a second-order closed sub-graph between the prediction result and the question entity x; and performing perturbation analysis on the obtained second-order closed subgraph by using a perturbation algorithm to obtain important nodes and paths, and taking the important nodes and paths as an interpretation analysis result of a prediction result. Compared with the prior art, the interpretability method and system based on the knowledge graph embedded prediction model provided by the invention can solve the problem that an existing prediction method based on the knowledge graph cannot provide interpretation analysis about the prediction result; the analysis can be used for proving the effectiveness and rationality of the prediction result, and the requirements of high performance and high stability can be met.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and more specifically, relates to an interpretability method and system based on a knowledge map embedded prediction model. Background technique [0002] Knowledge graphs have become an important representation of semantically organized information in different domains. It usually organizes information into collections of facts. fact e 1 , r, e 2 Indicates the head entity e 1 Links to tail entity e under relation r 2 . Knowledge graphs have a wide range of applications in a series of important areas such as search, recommendation, and drug repositioning. [0003] However, knowledge graphs are usually incomplete. To address this problem, researchers have proposed many link prediction methods to predict missing facts: The first class of methods is based on knowledge graph embeddings, which learn embeddings for each entity and relation and use score-based ranking to indicate ...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the above defects or improvement needs of the prior art, the present invention provides an interpretability method and system based on knowledge graph embedding prediction model. The technical problems in the knowledge graph embedding method, and the technical problems that adding learning rules or context paths to the knowledge graph embedding training will increase the training complexity, and the lack of sufficient information for logical reasoning due to only one fact to explain , which in turn causes the technical problem of low interpretation accuracy

Images