| In the recommendation system,how to efficiently carry out personalized recommendation for thousands of people is a challenging task,and it is also a hot research topic in academia.In order to carry out personalized and accurate recommendation,it is necessary to deeply mine and utilize the user’s interest preferences,which are not only contained in the user’s short-term behavior,but also in the user’s long-term historical behavior;therefore,good mining needs to comprehensively consider the user’s preferences.long-term and short-term behavior.Recently,many deep learning models have achieved good results in user behavior modeling.In this paper,the research and application of personalized recommendation technology are carried out in the field of online hotel reservation.By analyzing the data of large-scale users when they book hotels online,the author finds that there are two important challenging problems:First,there are differences in the data and roles of users’ longterm and short-term behaviors.The short-term behavior of users within a few hours before placing an order is very important and can largely reflect the needs and preferences of users.However,most users have less shortterm behaviors,and it is extremely difficult to accurately predict users’interest preferences through less user behaviors.On the other hand,there is a lot of long-term behavior data of users,but the noise is large.Therefore,in order to fully understand users,it is necessary to take effective methods to jointly model long-term and short-term behaviors.The second is that user decision-making is extremely complex,considering many factors,such as price and distance,and these factors interact with each other,and different users have different preferences for different factors.Therefore,it is necessary to model users’ preferences for different factors to improve the effect of personalized recommendation.Aiming at the above two problems,this paper proposes a personalized recommendation model that combines long-term and short-term behavior sequence modeling of users.Among them,for the first problem,this paper innovatively proposes the idea of using short-term behavior to guide longterm behavior modeling;for the second problem,this paper proposes the idea of fusion of explicit points of interest and auxiliary information to express different interests.To realize these two considerations,an explicit interest modeling module and an implicit interest modeling module based on attention mechanism are designed in the model.The explicit interest modeling module models different interests in long-term behavior and short-term behavior respectively,and then combines long-term interests and short-term interests through a gated structure;this module can dynamically expand to model different specific interest preferences.The implicit interest modeling module is mainly composed of a short-term sequence modeling module and a long-term sequence modeling module.The short-term sequence modeling module extracts real-time preferences from short-term behaviors,and then guides long-term sequences to perform behavior extraction through the attention mechanism to filter noise.effect.Experimental results on large-scale real data demonstrate the effectiveness of the model.Based on the in-depth analysis of the personalized recommendation requirements in the field of online hotel reservations,a personalized recommendation system for online hotel reservations is designed and implemented.The system mainly includes user information management module,hotel information management module and personalized recommendation module.The recall and sorting function of user behavior can give different hotel recommendation results based on user browsing behavior.Tests show that the system can fully meet the expected requirements. |
