Research On Sequence Prediction Algorithm Based On Point Process

With the development and progress of Internet technology,Various trades and professions have produced a large number of sequence data through the network,using the sequence data to make predictions,and excavate the inherent law behind its complex dynamics,It has important social significance for dynamic mining of user interest,real-time pushing of electronic advertisements,equipment fault detection,traffic fault prediction and other aspects.However,the current sequence prediction methods still have the shortcoming that the information is not fully utilized.Therefore,this paper studies the point process sequence prediction algorithm that integrates social data,and uses historical sequence data to predict event type and time.The main work and innovation points are as follows:(1)In view of the lack of considering the different dimensions information,this paper proposes a social point process sequence prediction algorithm(SPSP).The point process intensity function is modeled by synthesizing the two dimensional information of time and space.The introduction of social network into the point process sequence prediction algorithm opens up a new idea of modeling intensity function.The SPSP algorithm training process: firstly,in thetime dimension,the background knowledge and historical influence of the intensity function are modeled by using the double LSTMs.Then the double LSTMs outputs are merged through the joint layer to generate the preselected event type representation and the corresponding event time.Finally,in the spatial dimension,the preselected representation of event types is taken as input,and the intensity function is optimized and reconstructed according to the neighbor influence degree algorithm of the social network,and the entire model is trained to predict the final event type and time.(2)In view of the lack of interpretability in modeling point process with recurrent neural networks,this paper proposed a social point process sequence prediction algorithm based on attention mechanism(ASPSP).ASPSP algorithm adds the attention mechanism into the event sequence training,which solves the problem that the prediction ability of the model is limited because all the context input information is compressed into a fixed length vector.After experimental verification,the predictive ability and relationship mining ability of the algorithm are improved and the algorithm modeling process is more interpretable.(3)In practical application,the algorithm proposed in this paper is portable and extensible,and it can be applied to different fields with large span.For example,microblog type prediction in the social network field,elevator fault prediction in the equipment maintenance field and accident location predictionin the traffic field.In this paper,algorithm verification was carried out on sina weibo data set and elevator fault data set respectively,and Logistic,RMTPP,TRPP,ERNN,TRNN and other algorithms were used as the baselines comparison algorithm.Among them,the prediction event type evaluation index were Precision,Recall,F1 value and Accuracy,and the prediction event time evaluation index was MAE.Experimental results show that the ASPSP algorithm in this paper has better prediction performance and can predict the next event type and time efficiently and accurately.