Research On Sequential Recommendation Of Augmented Attention Mechanisms Based On Multi-behavioral Feature

In recent years,with the rapid development of information technology,the amount of information people face has been increasing,so that it is difficult for them to process and understand the information efficiently.Sequential recommendation system is an effective solution to "information overload".It recommends suitable item sequences for users by analyzing users’ historical behavior and preferences,helping users quickly find interesting content.At present,the attention mechanism-based model has become the mainstream modeling method for sequential recommendation.However,most of the existing attention mechanism recommendation models simply apply the Transformer model to the field of sequential recommendation,ignoring the context information and various regular information(multiuser behavior features)in the user behavior sequence.In addition,how to effectively represent multi-user behavior features and incorporate them into the attention mechanism is also a key issue.To this end,this paper proposes two attention-based sequential recommendation models and embeds them into a personalized movie recommendation system.The main research content and innovation points of this paper are as follows:(1)To address the issue of the existing sequential recommendation models tend to ignore the context information and regularity information of user behavior,We propose an attentionbased sequential recommendation model based on user behavior features(AUBRec).This model uses item interaction patterns to represent user behavior features and uses a sliding window mechanism to construct item interaction patterns.At the same time,an adaptive threshold control mechanism is used to fuse item interaction patterns and Transformer to enhance the attention mechanism’s learning ability for information such as relativity,immediacy,continuity,and non-strong sequential dependence in user behavior sequences.In addition,We also propose a recommendation model based on dual-channel adversarial attention mechanism(AUBRec+),which improves the robustness,generalization,and recommendation performance of AUBRec.We use seven recommendation models on four datasets for comparative analysis experiments and conduct sampling evaluations.The results show that compared with other state-of-the-art methods,AUBRec improves recommendation performance by 1.24%~7.28%,and is almost equivalent in training time and model size.(2)To address the problem of the simple representation and difficult integration of user behavior features in the AUBRec model,We propose a multi-user behavior feature fusion and enhancement-based sequential recommendation model(MUFRec).This model further optimizes the representation and fusion of user behavior features,using numerical item interaction patterns to represent user behavior features and combining Embedding technology to encode them,to achieve more complex and diverse behavior feature expressions.Meanwhile,MUFRec also designs a multi-behavior feature fusion network that can align and fuse multi-behavior feature vectors and attention maps to minimize information loss.In this paper,we compare eight recommendation models on three datasets and conduct an unsampled evaluation.The results show that compared with other state-of-the-art methods,MUFRec improves recommendation performance by 10.23%,11.52%,and 7.81% on three datasets,respectively.(3)Finally,We design and implement a simple personalized movie recommendation system.Based on the Movie Lens-1M dataset and crawled movie image data,the system uses the MUFRec model proposed in this research and the existing Deep FM model to recall and reorder candidate movies,thereby generating the final sequence recommendation list.Finally,the recommendation system combines statistics-based classification,hotspots and other recommendation forms to provide personalized push to users.In summary,the attention-based models proposed in this article show significant performance improvement in sequential recommendation tasks,and the implementation of a practical personalized movie recommendation system further proves its practicality.This research has important implications for the theoretical research and application promotion of sequential recommendation models.