Research On Semantic Social Networks Based On Knowlegde Graph

With the development of Internet technology,social networks have become an important source for people to acquire knowledge and exchange information.However,in the face of the explosive growth of network data,it is difficult for people to effectively retrieve and use information,which seriously affects the user experience.As a semantic analysis technology of social network,recommendation system can help users filter invalid data and recommend information they are interested in,which greatly improved the efficiency of users.However,there are some problems in traditional recommendation technology,such as data sparsity,cold start and so on.Knowledge graph can establish a mapping from string description to structured semantic description,and provide users with accurate recommendations by introducing more semantic information.Therefore,the recommendation research of social network based on knowledge graph has important research value.In this paper,we mainly construct a social network knowledge graph,and effectively utilize the semantically related information of knowledge to carry out the research of recommendation system based on the knowledge graph.The main work is as follows:Firstly,aiming at the problems of complex structure,low computational efficiency and difficulty of wide used in BERT model,a knowledge distillation based BERT model for social network entity extraction is proposed in this paper.This method analyzes the problems of BERT,and combines the Sina Weibo to distill the knowledge from the multi-layer model into a lightweight model with fewer layers to lighten the model.Experiments on the Glue reference set prove that the model can greatly compress the structure and improve the computational efficiency by about 5 times while retaining more than 95%accuracy.Secondly,based on the entity extracted by the improved BERT model,similarity,Shannon entropy and relative distance is introduced in this paper to mine the semantic information of entity relationship.Then the knowledge graph of social network is constructed to provide high-quality semantic auxiliary information for recommendation system.Finally,as the traditional recommendation system tends to recommend items similar to the user history over time,which leads to the closed-loop phenomenon and Matthew effect,a novel recommendation method based on Weibo social network is proposed in this paper.In this method,the concept of novelty is introduced,user activity,conformity,innovator index and other indicators are added,and the potential interest of users is mined by the auxiliary semantic information of knowledge graph.Experiments show that the proposed method can achieve a balance between contingency and precision,and the number of unpopular items recommended increases by 10 percentage,which effectively avoids the Matthew Effect in the recommendation system.In conclusion,this paper constructs knowledge graph and mines user interest with its semantic information to solve Matthew effect in recommendation system effectively.The validity of the method is verified through the Weibo data set,which has certain theoretical significance and application value for the semantic research of social networks.