Research And Implementation Of A User Behavior Prediction Method Based On Deep Learning

In recent years,the popularity and rapid development of the Internet have greatly changed people's life.Most behaviors are performed through the Internet,the behavior data generated by users contain great value.At the same time,we are in the age of big data,information overload not only greatly increases the operating cost of service providers,but also makes it difficult for service consumers to efficiently find the information they need.Users need to fr-equently filter information on the Internet,the cost of making decisions is getting higher and higher if users want to get a good experience.Making full use of user behavior data can alleviate such problems well,it can help users make decisions by using user behavior prediction method,and it can improve user experience.The user behavior prediction method proposed in this paper provides an auxiliary reference for user's next behavior decision by predicting user's next behavior or object list of next behavior.There are many applications of the method,such as e-commerce,entertainment industry,security industry,online education,etc,and many modeling methods can be used.In user behavior prediction,recurrent neural networks(RNNs)are better than traditional machine learning methods.After literature investigation,it is found that the previous work only used recurrent neural network to model the user's sequential behavior,but did not fully consider user preferences and behavioral intentions in the behavior sequence.In addition,the previous work did not deal with the timeliness of data.However,the method proposed in this paper can solve those problems.Specifically,we explore a bidirectional gated recurrent unit(BiGRU)encoder with attention mechanism to model user's sequential behavior,it can not only solve long-term dependency problem,but also capture user's intention.In the prediction stage,embedding vector matching method proposed in this paper can greatly reduce the amount of network parameters.In addition,in order to deal with the problem of data timeliness,a time-transfer learning method is proposed in this paper,it not only significantly improves the prediction effect but also greatly reduces the training time.In terms of experiment,Recall and MRR evaluation indicators are used to evaluate all methods.The results of experiments on two datasets from different fields show that the proposed method has obvious effect on user behavior prediction,and it performs better than other widely used methods.The prediction effect is significantly improved espacially when the time-transfer learning method is used.