Research On Relational Reasoning Of Knowledge Graph Based On Formal Concept Analysis

Formal concept analysis is an effective tool for data analysis and rule extraction.The research of formal concept analysis on knowledge acquisition is the research of implication.Decision implication is a concept proposed to reduce the scale of implication,which is logical and explanatory.At present,the above research has been widely used in many fields such as text mining,recommendation system,attribute reduction and concept-based cognitive learning.With the deepening of research,some scholars have found that it also has certain application value in relation completion reasoning based on knowledge graph.Relation completion,as one of the tasks of knowledge completion,is often solved by embedded models such as translation models and convolutional neural network based models.These knowledge embedding models all inherit the powerful ability of representation learning and perform well in relationship prediction tasks.However,because these knowledge graph-based embedding models deal with triples independently,lack attention to the inherent or potential relationships near a given entity in knowledge graph,and ignore the network structure of knowledge graph and the logical relationships between triples,the models have weak interpretability.Therefore,in order to solve the drawbacks of the reasoning methods,this paper proposes a new method for knowledge reasoning,solving the problem of relational reasoning prediction from the perspective of formal concept analysis.The research results obtained not only theoretically prove that the method can infer some missing information,but also prove that the method has advantages over the above knowledge graph reasoning methods through experiments.In addition,in order to further improve the performance of knowledge reasoning based on translation model,this paper also studies whether the new knowledge deduced by decision implication reasoning method can be added to the training set of translation model to improve the embedded representation performance of entities and relationships.The specific research work includes :(1)It was theoretically proved that the inference rules based on knowledge graph can be represented by implication and decision implication based on formal concept analysis.Then,in order to quickly mine decision implication for knowledge reasoning,complex contexts were reduced many times,and it was proved that the decision implications mined from the reduced contexts are equivalent to the implications mined in the original context.In addition,the object reduction method is also compared with the existing attribute reduction methods.Finally,the feasibility of the proposed method is verified by specific examples and experiments.(2)A knowledge reasoning method integrating decision implication and translation model was proposed.The new knowledge deduced by the decision implication reasoning method was added to the training set of the translation model for training to obtain a new entity relationship feature vector.The ablation experiment was designed to verify the influence of the richness,knowledge quantity and accuracy of new knowledge on the accuracy of model reasoning.The results showed that the accuracy and richness of new knowledge affected the embedding representation performance of entity relationship to a certain extent,thus affecting the reasoning accuracy.There was a positive correlation between the amount of knowledge and the accuracy of model reasoning.The larger the amount of knowledge in the range,the better the model reasoning effect.At the same time,the model reasoning of the fusion decision implication reasoning method was better than the single model reasoning.The above research further verifies the feasibility of decision implication reasoning method,and provides new ideas for the application of decision implication reasoning method and fusion reasoning.