Research On Personalized Recommendation Based On Long-term And Short-term Preference Depth Joint Modeling

In the era of information explosion,recommendation systems play an indispensable role and provide many conveniences for our lives.As an important topic in the field of artificial intelligence,the research on recommender systems has drawn more and more attention from academia and industry.In recent years,with the continuous development of technology,deep learning has made breakthroughs in many research fields such as computer vision and natural language processing,and it has also brought new opportunities for the research of recommender systems.At present,the recommender systems based on deep learning has become a research hotspot in this field,and many deep learning-based recommendation algorithms have been developed and proved to have made great progress.However,the existing work is mainly based on users' long-term preference or short-term preference,and there is insufficient research on the joint modeling of users' long-term and short-term preference.Therefore,the recommender systems based on users' long-term and short-term preferences are an interesting research direction.Research shows that unexpected behaviors in the historical interactions cannot reflect the inherent preference of the users,and not identifying the unexpected behaviors in the recommendation will damage the performance of the recommender systems.In response to the above problems,the main innovations of this paper are as follows:(1)In order to realize the deep joint modeling of long-term preference and short-term preference and provide higher quality recommendation results for users,a Long and Short-term Preference Model(LSPM)based on the deep neural network was proposed.LSPM incorporates LSTM network and self-attention mechanism to learn the short-term preference from the user's recent historical interactions and jointly model the long-term preference by a neural latent factor model.Thus,the deep joint modeling of long-term and short-term preferences is realized.By fusing the long-term and short-term preferences,the system can get a better performance.The experiments on public datasets have demonstrated the effectiveness of the LSPM model.Compared with the state-of-the-art methods,LSPM got a significant improvement in HR@10 and NDCG@10,which relatively increased by 3.875%and 6.363%.(2)For eliminating the influence of unexpected behaviors on recommendation results and enhance the robustness of recommender systems,an Unexpected Behaviors' Effects Elimination Model(UBEEM)was proposed.A guidance mechanism is designed to help the UBEEM model eliminate the influence of unexpected behaviors by using the informationfrom target objects.At the same time,UBEEM introduces a Time Convolution Network(TCN)to model the user's recent historical interactions.Using TCN for high-level feature extraction increases the receptive field and can capture a better context information,which effectively improves the accuracy of the model.The proposed recommender systems can be effective in learning both short-term and long-term user preferences and generate better recommendations as a result.The paper also designs a guidance mechanism to help the recommender systems eliminate the effect of unexpected behavior in historical interactions.The experimental results show that the proposed(LSPM and UBEEM)model is superior to the existing algorithms and has certain reference value to the designing of recommender systems.