Research On LSTM Recommendation Algorithm Based On Fusing Features

With the rapid development of Internet technology and the popularity of mobile devices,people's online communication is becoming more and more frequent,which leads to the rapid growth of network information.Finding valuable information in such a large amount of data becomes very difficult.Recommender system is an effective way to solve this problem.It not only helps users to search quickly,but also has a wide range of applications in the business field,bringing profits to many companies,so it has been paid more attention by many research institutions.Efficient recommendation methods are not only practical but also of high commercial value.F-LFM-LSTM rating prediction model is proposed by combining deep learning with traditional recommendation methods.On Movie Lens100 k datasets,this paper studies the influence of different network parameters on the accuracy of the recommendation algorithm.In addition,the influence of user's label information and item's label information on recommendation accuracy was also studied.The specific studies are as follows:(1)This paper studies a LSTM rating prediction model based on effective features.Firstly,the effective characteristics of users and items are extracted through latent factor model(LFM);then the LSTM network is used to combine and optimize it,and F-LFM-LSTM rating prediction model is proposed.Secondly,this paper tests the influence of different network parameters on the prediction effect of the F-LFM-LSTM rating prediction model,and determines the optimal value of the relevant parameters.Finally,through experiments on Movie Lens100 k data sets,the results show that the proposed F-LFM-LSTM rating prediction model has better prediction accuracy than the LFM model before improvement.Moreover,this paper compares the F-LFM-LSTM rating prediction model with two other excellent recommendation prediction algorithms,Mean Square Difference(MSD)and Weight Slope One Algorithm(WSOA).From the evaluation index MAE and RMSE,we can find that the proposed F-LFM-LSTM rating prediction model has better effect and can improve the prediction accuracy.(2)This paper studies F-LFM-LSTM rating prediction model based on fusing features.In the extended area of the F-LFM-LSTM rating prediction model,the influence of user label information and item label information on the prediction results of F-LFM-LSTM rating prediction model is analyzed.In this paper,label information is divided into single label information and multi-label information for research.Firstly,the effect of single label information on the prediction effect of F-LFM-LSTM rating prediction model is studied.The study on Movie Lens100 k data sets suggests that item category information is an active and productive label information for improving prediction accuracy.In addition,based on the research results of single label,taking the item category information as the main one,other label information is fused to form multi-label information,and then the influence of multi-label information on the prediction effect of F-LFM-LSTM rating prediction model is analyzed.The experimental results show that the fusion label information is not the more the better,the appropriate amount of fusion label information can effectively improve the prediction effect.In summary,this paper mainly studies the prediction effect of F-LFM-LSTM rating prediction model.Meanwhile,the impact on recommendation effect of user's label information and item's label information is analyzed from two aspects: single label and multi-label.