Research And Implementation Of Interpretable Personalized Recommendation System

With the maturity of Internet technology and the arrival of the era of big data,personalized recommendations have become an integral part of artificial intelligence.Recommendation systems obtain a significant improvement in accuracy because of massive data and massive feature combinations,but people ignore the full mining of data information.With the improvement of accuracy,people are more and more concerned about the interpretability of recommendation results.But the current user-level or item-level explanations given by recommendation systems are too simple to convincing.In addition,the current recommendation systems are designed based on specific areas,such as video recommendations,e-commerce recommendations,etc.Such recommendation systems can design strategies based on specific business scenarios.However,these systems are tightly integrated with the business,which cannot flexibly adapt to different data and lack generality.The main contents of this thesis are as follows:(1)An interpretable personalized recommendation algorithm is proposed and implemented.Based on automatic feature selection and multi-task learning,important feature combinations are given as the recommendation explanation.Two-stage multi-task recommendation model boosted feature selection(TMRM)is introduced.Experiments on two public data sets are conducted on TMRM and other state-of-the-art models to verify the high accuracy of the algorithm,and case analysis and user research are designed to verify the good interpretability of the algorithm.(2)Aiming at the problem that TMRM takes too long to train,a Fast TMRM algorithm is proposed and implemented.And experiments were performed based on the same data sets,which verified that compared with TMRM,Fast TMRM has improved training efficiency by more than 60%,and maintains interpretability.(3)A universal and interpretable personalized recommendation system is designed by conducting a needs analysis and summarizing the recommendation process.Then,the system is verified in specific project in the laboratory.