Abstract- Recent research onrecommendation has shifted from explicit ratings to implicit feedback, such aspurchases, clicks, and watches. For optimizing recommendation models BayesianPersonalized Ranking method is used. It is widely known that the performance ofBPR depends largely on the quality of the negative sampler. In implicitfeedback-based recommender systems, user exposure data, which record whether ornot a recommended item has been interacted by a user, provide an important clueon selecting negative training samples. The overall performance of BPR dependsmostly on the quality of negative sampler. This project make two contributionswith respect to Bayesian personalized ranking. In first contribution, we findthat sampling negative items from the whole space is unnecessary and may evendegrade the performance. Second contribution focusing on the purchase feedback ofE-commerce. In this project propose an effective sampler for BPR. In our proposedsampler, users view are considered as an intermediate feedback between thosepurchased and unobserved interactions. Also in proposed system we implement anApriori algorithm to find frequent items and we have considered all reviews ofnegative sampler items.
Keywords: BPR,Sampler, View Data, Recommender Systems, Implicit Feedback.