Personalized Gift Recommendation Strategy Based On User Intention And Social Relationship

When people choose gifts,they need to choose from a large number of items,gift recommendation help alleviate this problem.Gift recommendation can be divided into personal gift recommendation and group gift recommendation: the existing personal gift recommendation methods are mostly monotonous when solving user preference,and only simply use the social relationship between users,which makes the user preference inaccurate,and recommended gifts are difficult to meet the needs of both the source user and target user;group gift recommendation method is not mature enough,but it can refer to other types of group recommendation methods,however they do not consider impact of user intention on group common intention and challenges faced in solving user intention.Therefore,gift recommendation faces many problems to be solved: How to solve rich and accurate user intention? How to use social relationship more rationally and effectively?How to comprehensively consider the source user and target user in personal gift recommendation? How to design a strategy to solve group common intention in group gift recommendation?To solve these problems,we propose two recommendation methods including a personalized gift recommendation strategy combining user intention and social relationship,and a group gift recommendation method based on users' common intention,which can effectively improve the recommendation effect.Aiming at personal gift recommendation,a personalized gift recommendation model(GRMUSI)is proposed that combines user intention and social relationship.Firstly,a user intention separation model combining user-item interaction history and social relationship is constructed,which uses a graph neural network to collect user-item interaction and social relationship information of each user,and solves the user's intention and item's feature vector;secondly,according to the direct and indirect trust relationship,it employs the social network walking algorithm to calculate the social weight between users;finally,two-way preference is calculated by combining social weight,the feature vector of the source user and the target user(composed of user intentions).And the inner product of the two-way preference and the item's feature vector is calculated to obtain a recommended list of gifts from the source user to the target user.Aiming at group gift recommendation,a group gift recommendation model based on user common intention(GGRMUCI)is proposed.Firstly,a user intention separation model combining user-item interaction history and social relationship is constructed,which uses a graph neural network to collect each user's user-item interaction and social relationship information to solve user intention and item's feature vector;secondly,users are grouped according to the K-Means clustering algorithm,then the group common intention representation is obtained by combining user groups,user feature vector composed of user intentions and group intention aggregation function;finally,calculates the inner product of the group common intent and the item's feature vector to obtain the recommended gift list for the group.Personal gift recommendation strategy utilizes graph neural network to learn user intent and item's feature vector representation from graph-structured data of user-item interactions and social relationships,which can effectively enrich user representations and alleviate data sparsity problems,taking into account direct and indirect trust among users relationship,and designed a network walking algorithm,fully considering the social weight between different users;group gift recommendation method fully considers the impact of the user's personality and the commonality among group members on the group common intention,and combines social relations to alleviate the problem of data sparsity,and designs a group intent aggregation function to obtain accurate group common intention.The experimental results show that compared to the currently related recommendation model with the best recommended effect,on the Gowalla and Yelp-2018 datasets,GRMUSI has better recommendation effects on Recall and NDCG indicators,and GGRMUCI has better recommendation effects on Precision and NDCG indicators.