A Knowledge Representation Method Combining Logical Rules and Confidence

Abstract

The invention relates to a knowledge representation method combining logical rules and confidence levels. For dynamic knowledge fragments that flow in in real time, if they already exist in the existing knowledge base, use the model trained by the knowledge base to perform knowledge reasoning; Stored in the cache area; when the data in the cache area meets the preset conditions, the knowledge fragments in the cache area are added to the knowledge base, and the knowledge base is retrained for model training. Its training includes using the rule mining algorithm to mine the Horn logic rules of the knowledge base on the knowledge base, and calculating the PCA confidence degree of the rule for each rule; calculating the confidence degree of the newly launched fact triple according to the confidence degree of the rule combined with the probability soft logic ; Embed the joint confidence of the triplets that fuse the semantic information of logical rules together to improve the performance of knowledge representation. The invention realizes the knowledge reasoning of uncertain reasoning under the dynamic knowledge map.

Description

technical field [0001] The invention relates to the fields of knowledge representation and knowledge reasoning, in particular to a knowledge representation method combining logic rules and confidence. Background technique [0002] At present, the knowledge representation methods based on logical rules mainly include AMIE, AMIE+, HornConcerto and so on. These methods mainly take the means of applying inference rules to the knowledge base, and derive new facts by triggering the antecedents of the rules. However, these logical rule-based knowledge representation methods only consider the direct facts in the knowledge graph, ignoring the important influence of the semantic information hidden by the logical rules in the knowledge graph on the embedding of relations and entities. In addition, the knowledge representation method for uncertain knowledge is AR-TransE. Although this method considers the confidence information of triples in knowledge representation, it also ignores th...

AI Technical Summary

Problems solved by technology

However, these knowledge representation methods based on logical rules only consider the direct facts in the knowledge graph, ignoring the important impact of the hidden semantic information of the logical rules in the knowledge graph on the embedding of relations and entities.

In addition, AR-TransE is a knowledge representation method for uncertain knowledge. Although this method considers the confidence information of triples in knowledge representation, it also ignores the semantic information contained in the rules.

And this method selects the maximum value of rule confidence as the confidence of new inference facts, which cannot represent the confidence of inference new facts well

[0003] The explosive growth of information carries a lot of uncertainty, and the existing knowledge representation method based on logical rules cannot solve the reasoning problem under uncertain knowledge.

Moreover, the existing knowledge representation methods mainly consider the direct facts in the knowledge graph, ignoring the semantic information contained in the logical rules.

At the same time, with the real-time inflow of dynamic knowledge fragments, existing models are not well adapted to knowledge reasoning problems under dynamic knowledge graphs.

Images