Research On Sequence Recommendation Method Of Multi-class Behavior Feature Fusion And Collaborative Filtering

A recommendation system is an information filtering technology that aims to provide personalized suggestions or recommendations to users to improve their satisfaction.Recommendation systems typically recommend products or services based on historical user behavior,preferences,and other information.These systems are widely used in e-commerce,social media,music,and video streaming.Common recommendation algorithms include content-based filtering,collaborative filtering-based,deep learning,and hybrid algorithms.The existing recommendation algorithm is mainly based on the user’s historical behavior data,and this static recommendation algorithm cannot adapt to users’ changing interests and behavior,resulting in less accurate and personalized recommendation results.The focus of recommendation system research is how to effectively and rationally use recommendation algorithms to match users,improve recommendation accuracy and efficiency,and provide users with a better experience.To address the above problems,this paper proposes a sequential recommendation method for heterogeneous information networks with a fusion of multi-class behavioral features and a time-aware recommendation method based on collaborative filtering.The main research work and innovation points are as follows:A sequential recommendation method for heterogeneous information networks with a fusion of multi-class behavioral features is proposed,which fully considers the heterogeneity and association relationships of user behaviors in heterogeneous information networks,and embeds and fuses homogeneous relationships to globally fuse user behavioral features from heterogeneous perspectives,and combines internal storage units in memory networks to iteratively update them,and finally predicts ratings to achieve sequential recommendations.In this way,the heterogeneous features of user behavior are effectively utilized to mine the higher-order behavior patterns of users and improve the accuracy of recommendation.A time-aware recommendation method based on collaborative filtering is proposed.Traditional recommendation methods usually only consider the short-term interests of users and ignore the temporal information of interactions.To solve this problem,the method specifically analyzes temporal information from user interaction records and proposes the concepts of relative time and absolute time.At the same time,the method combines time sequences and item information sequences and uses a self-attention mechanism for collaborative representation to obtain candidate items.In this way,the temporal embedding of interaction records can be more fully utilized to improve recommendations’ accuracy and effectiveness and solve the problems of slow computation and poor robustness.Finally,experimental validation is carried out on several real datasets,and comparison experiments are conducted with non-sequential and sequential models and self-attentive recommendation models under different parameter metrics,while the parameters of the models are screened to select the parameters that can better perform the recommendation effect.The experimental results show that thus the recommendation effect and user satisfaction are improved by considering multiple interest directions and behavioral habits of users comprehensively.The method can play an important role in practical applications,such as personalized recommendations in e-commerce platforms,social networks and other areas.