Joint friend and item recommendation based on multidimensional feature reciprocal interaction in social e-commerce

Social e-commerce platforms need to undertake the two core tasks of recommending potential social friends and preferred consumption items for users. However, the use of traditional one-dimensional information is no longer able to accurately make personalized recommendations. Early scholars have confirmed that users' social and consumption behaviors do not exist independently: users with the same interests are more likely to become friends, and there is a high probability of similar consumption behaviors among friends. In this paper, we propose a joint friend and item recommendation model based on multidimensional feature reciprocal interaction (MFRI). Which is based on the user's social friends and item preference information, extracts the shallow and deep features of the user's social and consumption behaviors, and utilizes the reciprocity between unusual behaviors to achieve mutual enhancement. The reciprocity between shallow and deep features in similar behaviors is also explored based on the attention mechanism, and the model is trained by a joint loss function. We conducted experiments on real datasets, and the results confirm the effectiveness and robustness of MFRI for potential friend and preference item recommendations.