Research On Deep Collaborative Filtering Recommendation Algorithm Based On Multi-dimensional Feature Intersection And Multi-objective Optimization

While the era of big data brings convenience to people,a large amount of data also troubles people-it is difficult to obtain the required information from the complex data,and sometimes it even misleads oneself.Recommender systems can extract information that people may need from big data,which greatly alleviates the problem of information overload.In recent years,the rapid development of deep learning has also promoted the development of recommendation systems.The performance of recommendation algorithms based on deep learning has been continuously improved,but there are also disadvantages such as sparse data,insufficient use of implicit feedback,and insufficient feature crossover.In view of the shortcomings of the above deep recommendation algorithms,this thesis proposes two solutions by using the deep learning framework by studying the implicit feedback data,multi-dimensional feature intersection,and combining with the multi-objective optimization method.The main research contents of this thesis are as follows:1)Analyze the advantages and disadvantages of the existing models of the current deep recommendation algorithm,and propose a deep collaborative filtering algorithm DCF based on multi-dimensional feature intersection in view of the problems of insufficient use of implicit feedback and insufficient diversity of features.Using implicit feedback data and ID data,the neural network is used to reduce the dimension of high-dimensional sparse data,and the features of the data after dimension reduction are intersected to enrich the feature information.The low-dimensional embedding vectors of the implicit feedback and ID data are then concatenated,allowing the neural network to extract more information.The method also considers the effectiveness of implicit feedback and ID data for feature extraction,and uses feature intersection to enrich feature information,which provides convenience for deep neural networks to extract high-dimensional feature interaction.2)Aiming at the problems of sparse data and the mutual independence between implicit feedback data and ID data,a multi-objective deep collaborative filtering algorithm MDCF is proposed,which combines implicit feedback and feature information.This method can effectively establish the connection between the implicit feedback data and the ID data,further increase the information contained in the features,and at the same time introduce the feature information,which can further alleviate the data sparse problem.Through multi-objective optimization,the influence of the previous prediction score and the latter prediction score on the final score prediction task is considered at the same time,and the implicit feedback and feature information are used as the input of the score prediction task,and the two prediction scores are controlled by hyperparameters for the final recommendation.The effect of precision.3)The recommendation performance of DCF and MDCF is evaluated on Movie Lens and Film Trust datasets.The impact of important hyperparameters in the model on performance is studied.Compared with the same type of model,DCF’s indexes MAE and RMSE on Movie Lens datasets decrease by 44.3% and 37.2% on average,and MAPE and s MAPE decrease by 43.9% and40.0% on average.,on the Film Trust dataset,MAE and RMSE decreased by 31.2% and 31.6% on average,MAPE and s MAPE decreased by 35.3% and 27.8% on average,and MDCF on the Movie Lens dataset compared with the same type of models.The average decrease in MAE and RMSE 14.5% and 13.8%,with an average decrease of 20% and 12.5% on MAPE and s MAPE.