Multi-Task Learning For Session-Based Recommendation

In recent years,recommendation technology,as one of the effective means to solve the problem of information overload,has become one of the hot issues in the field of big data research.Session-based recommendation is the most important branch in the recommendation systems.When the user labels are not available,the session-based recommendation system provides individualized services for the users based on the current session.Most studies on session-based recommendation neglect the fact that the incidence relation between the users and the objects varies from user to user,which makes the recommendation less accurate.In addition,most session-based recommendation models overlook the fact that the user preferences hidden in the session history have an impact on the users' current purchase intention,which make the recommendation less individualized.Moreover,little information included in the session-based recommendation scenario is useful.The recommendation could be not comprehensive.To provide individualized recommendation services for the users more efficiently and accurately,this paper,based on existing findings,comes up with a session-based recommendation model based on multi-task learning——GNN-Attention-Mulit Task-Learning Session-based Recommendation(GAM-SR).The research covers the following areas:(1)Propose to obtain the incidence relation with the interaction objects by developing the user behavioral digraph.Get the objects' vector representation by using the graph neural networks to capture the incidence relation with the interaction objects in the user behavior digraph.Make up the object information lost in the session-based recommendation modeling process to make the recommendation more reasonable.(2)Suggest using the attention mechanism to capture the impact of session history on the current session.Learn about the users' preferences from session history and analyze the impact of session history on the current purchasing intentions to provide better individualized recommendations.(3)Propose to abstract the effective hidden data of similar users via multi-task learning.Model and analyze the session data of similar users based on multi-task learning,gain rich and effective shared knowledge,and make up the session information of the target users to get proper user representation and more accurate recommendation.To test the validity and rationality of the GAM-SR model,the user interaction behavioral data sets of Alibaba and Diginetica are evaluated with two indicators,Recall and MRR.The performance of GAM-SR model,base model and self-transformation model was compared in the same operating environment.The results suggest that GAM-SR performs better,proving the efficacy and rationality of this model.