A Study On Knowledge Representation Algorithm Based On Improved Fully Connected Neural Network

Knowledge graph is a directed graph composed of entities and relations.In the graph,the nodes represent various entities in the real world,and the edges represent the relations between these entities.However,the knowledge graph is usually incomplete,there will be a large number of missing relations,and manual completion of these missing relations will consume a lot of manpower and time.Therefore,knowledge representation learning is proposed to realize the automatic completion of knowledge graph by embedding the entities into a continuous low-dimensional vector space.Trans E model is the first translation-based knowledge representation model proposed by Google.It is the basis for all translation-based knowledge representation models.The core idea of all translation-based models is to represent the relation as a transformation from a head entity to a tail entity.However,existing methods use fixed transformation formulas,which limits their expressive ability,so that they all have certain expression defects.This paper proposes a new translation-based knowledge representation models named Translating Implicitly(Trans I),which uses the traditional fully connected neural network to implicitly learn the transformation process.It can effectively solve two kinds of complex relationship situations that are difficult to deal with other translationbased models.However,due to the large number parameters of traditional fully connected neural networks,the training time is long,and it cannot solve 1 to N and N to N relations.Therefore,the paper further optimizes the traditional fully connected neural network.By transforming the matrices in the network into vectors,the problem of many parameters and long training time is solved,and the Euclidean distance between the vectors in the scoring function is converted into vector dot product to make the model fully expressive.The experiment results prove that the Trans I model is superior to most existing models,and the optimization and improvement of the traditional fully connected neural network also significantly improves the final effect of the TransI model.