Personalized Fashion Recommendation Using Nearest Neighbor PageRank Algorithm

E-Commerce has seen a lot of growth over the past decade. With an increase in commodities, especially fashion accessories and clothing items in the online-market, a need for an efficient recommendation system arises for better information filtering. Several different apparel recommendation systems already exist in the literature. However, as time passes, new challenges are arising, such as computational complexity and an exponential increase in data. Also, due to fast-changing trends, the recommendation model is required to update frequently. This work proposes an improvised collaborative-filtering based recommendation system. A ranking algorithm, Nearest Neighbor PageRank (NNPR), is developed that uses the nearest neighbors of the user along with the PageRank algorithm to generate personalized recommendations. The proposed model, is evaluated in comparison with Alternating Least Square (ALS) algorithm. The experiments are conducted on Amazon Fashion Review Dataset, and the results of this experiment are recorded in Hit-Rate (HR) and Mean-Reciprocal Ranking (MRR). It is observed, that NNPR performs better than ALS in both Active User and Cold Start scenarios. Moreover, the hybrid model ALSNNPR improves the performance of ALS using NNPR as a ranking algorithm.