Research On Sequential Recommendation Method Based On User Behavior Distance

In recent years,with the rapid development of the Internet and the explosive growth of information,people need to invest more time and energy to make a reasonable choice.How users can quickly and efficiently find the information they need in the mass of data has become a major problem to be solved in today's era.Personalized recommendation can obtain user preferences by analyzing user's historical data to push information,thus alleviating the problem of information overload to a large extent.Traditional recommendation algorithms only model the static relationship between users and projects,ignoring the fact that users' interests will change over time in real life,resulting in low accuracy of recommendation results.However,some existing recommendation methods based on users' long-term and short-term interest usually regard user behavior sequence as uniform distribution and fail to accurately capture the changes of users' interests in a short period of time,resulting in a single recommendation result.Therefore,in view of the above problems,this paper proposes a sequential product recommendation method based on user behavior distance.The specific work is as follows:(1)In order to accurately extract the user's dynamic interest,meet the demand of the user's shopping,this paper proposed a sequential recommend method based on user's behavior distance recommend method,first of all,the history of the user behavior sequences are divided into different sessions,in each session has a certain consistency between the user's behavior,the behavior within the latest session as the short-term behavior of the user,history session sequence as the long-term behavior,The shortterm and long-term interests of users are modeled respectively,and then the long-term and short-term interests are fused with the adaptive gating unit.And the score of each product is calculated according to the final interest vector of users,and then the recommendation is made after ranking.(2)In view of the short-term behavior of users,the temporal sequence and correlation in user behavior are mainly considered.There is a time distance between user behaviors.In order to better capture the change of user interest,a time gate modeling time interval is added into the gated loop unit to control the change degree of user intention through the time interval between user behaviors.In addition,the historical behavior of users may also imply a variety of different potential intentions.Multiple attention heads of multi-self-attention mechanism are used to model the features of different levels of abstraction to capture the user's interest points in different directions.(3)In view of the long-term behavior of users,the comprehensiveness of user interest characterization is mainly considered.There are multiple sessions in a user's long-term behavior,and the correlation of user behaviors within a single session is greater.Users' intentions are similar,but different sessions are also associated and complement each other.Therefore,this paper uses Deep FM model based on depth weighting to better capture the similarity of user intentions between different sessions and extract stable long-term user interests.In order to verify the effectiveness of the recommended method,a comparative experiment was conducted on taobao data set.The experimental results prove that the fusion of time information in the proposed method can effectively improve the accuracy of recommendation results.Compared with popular algorithms such as GRU4 Rec,SHAN and DIEN,the proposed sequential recommendation method based on user behavior distance also has a certain degree of improvement in hit rate and average reciprocal ranking.In addition,while considering the time factor,the recommendation method proposed in this paper can actively explore the new interests of users and better adapt to the immediate purchase needs of users in different shopping scenarios.