Entity alignment method based on weighted neighbor information coding

Abstract

The invention discloses an entity alignment method based on weighted neighbor information coding, and the method specifically comprises the steps: 1), carrying out the preprocessing of data in two knowledge bases needing to be aligned, and extracting two knowledge base triples, entities and neighbor information thereof, and entities and type information thereof; 2) based on all currently discovered matching entity pairs, obtaining a vector representation corresponding to each entity through triple-based knowledge representation learning, weighted neighbor information coding and cross-knowledge-base entity-type graph embedding; 3) reasoning matching entity pairs in combination with three different vector representations of the entity; and 4) forming new training data by the discovered matching entity pairs and the priori aligned seed entity pairs, repeating the steps 1)-4) until the specified iteration times are reached, and outputting the discovered matching entity pairs. According tothe method, fewer entities appearing in the triad can be more accurately matched, and the method has a wide application prospect in the fields of knowledge fusion, knowledge questions and answers andthe like.

Description

technical field [0001] The invention relates to the field of knowledge base entity alignment, in particular to an entity alignment method based on weighted neighbor information coding. Background technique [0002] A knowledge base organizes human knowledge in a structured form, aiming to describe various entities and their relationships existing in the real world. With the development of Web 3.0, many knowledge bases have appeared one after another. In addition to comprehensive knowledge bases such as DBpedia and Freebase, it also includes many domain-specific knowledge bases such as movie knowledge bases (such as IMDb, LinkedMDB), music knowledge bases (such as MusicBrainz, Discogs), and these knowledge bases are used as advanced systems for question answering systems and recommendation systems. Experienced knowledge is playing an increasingly important role. However, different organizations or institutions only consider their own needs when constructing knowledge bases,...

AI Technical Summary

Problems solved by technology

However, existing entity alignment methods based on representation learning usually assume that there are enough training triples for each entity in the knowledge base, and cannot perform long-tailed entities in the knowledge base (i. Entities) for full learning may lead to wrong matching of long-tailed entities and affect the accuracy of entity alignment results

Introducing the entity and its neighbor information in knowledge representation learning, and then using the vector representation of the entity based on the neighbor information for matching entity pair reasoning, helps to solve the long-tail entity mismatch problem, but the role of different neighbors of entities in entity alignment exists. difference, which the existing neighbor information encoding methods fail to take into account

Images