Personalized Recommendation Based On Bidirectional Effect Of Evolutive Trust-rating

The popularity of the Internet has brought explosive growth of online information.Users are drowning in massive amounts of data,and it takes substantial time and operating costs to find the products of interest on the e-commerce platform.Recommended system technology can improve this problem to some extent.With the prosperity of social media,e-commerce platform began to recommend users to buy products they might be interested in through the user's social network in industry,which led to the establishment of trust-based recommendation system in academia field.Existing trust-based recommendation researches mostly apply one-way social network data,which directly apply the data set,to pay attention to the propagation and aggregation rules of trust along the one-way network,and ignore two facts:(1)The trust relationship of users and rating on items will change with time,and the two react upon each other,(2)Recommend the items reference trust object of target user based on the similarity assumptions of the two,then generate the recommended list of trusted users can also learn from trust user's experience rating.In addition,how to improve the recommendation quality of cold start users and long tail items with less ratings is one of the unresolved issues in the recommendation system.The work of this paper is as follows:1)Considering bidirectional effect of evolutive trust-rating over time,this paper proposed a bidirectional effect of evolutive trust-rating model(BETR).Based on this model,the user's rating changes over time due to the influence of their own opinions and trusting friends' opinions,reflecting the influence of trust relationship on the rate.At the same time,the trust degree of user to the trusted users is determined by the trust degree of user himself and the rate similarity between user and trusted friends,reflecting the influence of the rate on the trust relationship.2)Extend one-way trust networks into two-way trust networks by introducing implicit trust metrics,a random walk model based on two-way trust(RWTT)is proposed.The model performs random walks on the two-way trust network to predict the missing rates of the target users and generate a recommendation list.When recommending to a target user,both the rates of this user's trusted users and the other users' rates who trust this user are considered.3)Introduce user ratings within the 3 hop or 6 hop that focus target user in the trust network to improve the recommendation effect of cold start users and long tail items.This paper integrates the BETR model and the random walk model,the BETR model and the RWTT model respectively and applies these two compound models to the Epinions cold start user dataset.The results show that the method effectively alleviates the cold start problem.4)This paper integrates the BETR model and the random walk model,the BETR model and the RWTT model respectively,then apply these two compoundmodels to the Epinions dataset.Experiments show that considering the Bidirectional effect of trust-rating and two-way trust really improve the recommendation performance.