Strong adaptive knowledge base replenishment method

Abstract

The present invention relates to a strong adaptive knowledge base replenishment method. The method comprises the following steps: retrieving a data source from a knowledge base and performing partial subgraph traversal; setting path feature extractors, wherein the path feature extractors comprise a PRA feature classifier, a path binary feature extractor, a modified one-sided feature extractor, a bilateral contrast feature extractor and a generalized feature extractor, the extraction processes of all the path feature extractors are the same and comprise path feature extraction and path feature selection, the input is a partial subgraph and the output is a path feature; constructing a feature matrix according to the feature extractor; and selecting a classification model, transmitting the feature matrix to the classification model, training the classification model, outputting an established entity and a relationship type corresponding to the entity by using the classification model, and transmitting an output result to the knowledge base, so that the knowledge base replenishment is realized. The method provided by the present invention has relatively stable performance, namely, a relatively good knowledge base replenishment effect can be obtained on different data sets.

Description

technical field [0001] The invention relates to a knowledge base completion method, in particular to a highly adaptable knowledge base completion method applied in the computer field. Background technique [0002] At present, large knowledge bases such as YOGO, NELL, Freebase, and DBPedia emerge in an endless stream. Based on these knowledge bases, scholars have carried out a lot of work such as relationship extraction, relationship inference, natural language question answering and knowledge discovery, which has largely promoted the development of corresponding fields. However, even large contemporary knowledge bases suffer from serious information imperfections. Occupational information belongs to the basic information of people. However, among the 2 million human subjects included in Freebase, only 300,000 have this information, and most of them are politicians or celebrities. The lack of these basic information will greatly hinder people from further utilizing the know...

AI Technical Summary

Problems solved by technology

[0005] Although SFE has made great improvements on the basis of PRA, there are still the following shortcomings: (1) when a single feature extractor is proposed, the factors considered are limited, and overfitting may occur; (2) before the final SFE algorithm is determined There is no evaluation on the pros and cons of a single feature extractor, resulting in the final SFE algorithm is not optimal; (3) The high dependence of the knowledge base completion model on the data set is ignored

This dependence will not only affect the performance of the same model on different data sets, but also affect the relative pros and cons of different models on different data sets.

Images