Research And Application Of Recommendation Algorithm Based On User Behavior

Traditional collaborative filtering and content-based recommendation algorithms are the mainstream recommendation algorithms at present,which have the characteristics of fast operation and strong explanation.It is difficult to obtain satisfactory recommendations due to the limitations of poor generalization ability and low degree of personalization.With the rapid development of artificial intelligence,it has become a feasible way to use deep learning technology to solve the problem of recommendation system.The research work is based on user behavior data such as clicks,browses,and ratings on the Internet and neural network algorithms,with the following highlights:For the explicit scoring behavior,a recommendation algorithm based on spatial dimensional distance measurement and user explicit behavior(SDDM-UEB)is proposed.First of all,the Item2 Vec model is improved by using a window that changes dynamically with the rating level to improve the accuracy of item feature training;secondly,the proposed spatial dimension distance measurement method is used in the feature fusion phase of user and item,and the interactive characteristics of user and item are obtained by measuring the distance between user and item in different dimensions in the potential space,and the predicted rating is output through the full connection layer.Finally,a Top-K list reordering method based on cosine similarity is adopted to alleviate the problem of user interest fatigue caused by long-term recommendation of similar items.For the implicit behavior,a combination of labels and implicit behavior is used to mine user preferences,and click prediction is realized through neural network.In particular,in order to achieve the purpose of improving the click prediction performance,a time decay function is introduced in the user feature representation to capture the user's preference changes.The RMSE of SDDM-UEB algorithm on Movie Lens 1M and Movie Lens 100 K data sets is 0.820 and 0.867 respectively,and the MAE is 0.641 and 0.677 respectively.The Recall of the recommendation algorithm which combines label information and user implicit behavior on Movie Lens 1M dataset is 0.37(recommendation length is 100).The two proposed algorithms are better than the traditional algorithms.Finally,the system requirements of the application of the film recommendation system are analyzed,and the client and server of the system are designed and implemented based on PyQt framework and MySql database.Figure 28;Table 12;Reference 64...