Prediction Research And Application Of Weighted Bipartite Networks Based On Semantic Features

In the era of big data,in order to meet the individual needs of users,an intelligence analysis framework IAF based on complex network has been built by our research team for sensemaking,understanding and prediction of human,objects,organizations,true and false events,and relationships.It is a key algorithm of the prediction module in our framework of intelligent intelligence analysis to study prediction of “human-thing”weighted bipartite network from large-scale social networking based on human electronic footprints in social/physical/cyberspace.Firstly,a personalized weighted bipartite network prediction model based on hierarchical attention and latent factor model(HALF)is proposed.On the one hand,in order to represent the characteristics of network nodes from text reviews,our model designs a hierarchical attention mechanism,and combines this mechanism with deep neural networks which can mine the latent semantic features in texts.One the other hand,the latent factor model is used to generate personalized node features to guide the network's prediction of unknown link weight values between heterogeneous nodes.Experimental results on open source dataset show that our proposed algorithm outperforms other comparison algorithms and achieves very good results.Secondly,a deep weighted bipartite network prediction model based on hierarchical multi-head attention(DWMA)is proposed.To obtain the high-quality representation of network node features,our model combines multi-head attention mechanism with convolution neural network,and utilizes them to extract various aspects of latent semantics combined in text reviews from various aspects and multiple angles.Experimental results dictates DWMA improves the performance of weighted bipartite network prediction.Finally,a case study is carried out on the author-paper bipartite network dataset.We designs a paper recommendation service based on our proposed weighted bipartite network prediction model for forecasting author's individual demands.This service enables authors to quickly select the papers that suit them,and is an importance part of application of intelligence prediction research.In summary,this paper proposes two weighted bipartite network prediction algorithms based on semantic features for the problem of user personalized demand forecasting in intelligence analysis.It has higher accuracy stability and can provide users with better individualized requirement service.