| With the rapid development of the Internet and social media,people’s information needs are increasing.However,with the continuous output of massive information,information overload problem also arises.In response to this urgent problem,recommendation systems have emerged.The emergence of sequential recommendation system opens up new directions and possibilities for the development of recommendation system.Many existing sequential recommendation algorithms based on self-attention mechanism mainly consider modeling user behavior sequence,while often ignoring the impact of time factor on user behavior.In addition,it is unreasonable to add position embedding information and item embedding information as network input in traditional absolute position encoding method in self-attention mechanism.This will lead to noise correlation and mixed correlation in model training,which interferes with model’s ability to capture complex relationships in sequence.This paper conducts in-depth analysis and research on the above issues.Firstly,A new sequential recommendation model with time information fusion(Sequential Recommendation Fusing Temporal Information,FTIRec)is proposed,which projects user’s historical interaction items into a constant number of latent interests,thus alleviating the problem of over-parameterization and high model complexity.Additionally,by extracting temporal information hidden in the user-item interaction sequence and injecting it into the self-attention network,the context information is enriched,and the model’s ability to capture complex sequence patterns is enhanced.The proposed model’s recommendation performance is verified on multiple real-world datasets by comparison with a set of state-of-the-art sequence-based recommendation models.Secondly,the position embedding is separated from the item embedding,and an independent position coding is designed.Considering the strong correlation between position and time information,the time information is fused with the position information,and the position information weight is calculated in combination with the self-attention mechanism,proposing a model called Self-attention-based Network for Time-aware Sequential Recommendation(SNTSR)and comparing with a set of baseline models.The results show a significant improvement in the model’s recommendation performance.Finally,a simple movie recommendation system is designed and implemented based on the SNTSR model,which can recommend movies that users may like,further verifying the effectiveness and practicality of the recommendation model. |
