Research And Application Of Long-term Effective Recommendation Algorithm For Interactive Sequence

While the Internet has brought convenience to people’s lives,it has also created the problem of information overload.In this situation,recommendation systems are born.Based on the history of user-item interactions,recommendation systems model user preferences and item characteristics to predict the items of interest to users,allowing them to quickly access the content of interest from a large amount of information.The research on user-item interaction sequence is an emerging topic in recommender systems and is receiving more and more attention from researchers.Despite the numerous research results in this area,there are still the following problems: 1.Only focus on user rating data of items,which has the problem of data sparsity;2.Only consider the interaction sequence,discarding the temporal information of interaction and ignoring the dynamic nature of user and item characteristics;3.The time interval between user-item interaction elements is non-uniform,ignoring this temporal irregularity;4.The historical interaction elements of users are usually multiple and non-identical in topic,ignoring this semantic irregularity.To address the above problems,thesis proposes two recommendation algorithms for user-item interaction sequences,and based on them,we design and implement a highly available online music recommendation system using various enterprise-level application development techniques.The main contributions of thesis include.1.Proposing a deep time-aware recommendation algorithm incorporating user reviews.1)Incorporating user and item reviews to better implement the modeling of user and item features from rich semantic information,effectively solving the problem of data sparsity.2)Introducing temporal information to consider the dynamics of user and item features over time.3)In the implementation part of the algorithm model,by introducing convolutional neural network for multi-dimensional extraction of user and project features to achieve more accurate modeling.The MSE and MAE are used as evaluation metrics to conduct comparison experiments with the baseline model on five publicly available datasets,and the results show that the algorithm has better performance performance.2.Proposing a personalized recommendation algorithm with sequence information enhancement.1)User comments are pre-processed and text summaries are extracted to improve the execution efficiency of the algorithm while maximizing the retention of core user sentiment information.2)Historical user interactions are treated as sequences with different time intervals,and this time interval information is incorporated into the modeling of interaction relationships.3)CNNs are introduced to uses multiple convolutional kernels of different sizes to learn user and item features for the purpose of capturing preference information at different stages in the entire historical interaction sequence.Using HR@N and NDCG@N as evaluation metrics,we conduct comparison experiments with the baseline model on five publicly available datasets,and the results show that the algorithm has better performance.3.Based on the recommendation algorithm proposed in the paper,a highly available online music recommendation system is designed and implemented by combining the application scenario of music recommendation,introducing distributed development ideas,and using various enterprise-level application development techniques.