Novel Approach for Neighbourhood-based Collaborative Filtering

Abstract

Recommender systems hold an integral part in online marketing. It plays an important role for the websites that provide the users an environment to rate and review the products. Several methods can be used to make recommender systems, like content-based filtering, collaborative filtering [1], hybrid approach, which combines content-based as well as collaborative filtering. Collaborative filtering is the most widely used technique to deal with recommender systems. Matrix factorization and neighbourhood approach are the techniques that can be used while dealing with collaborative filtering. Both the methods depends on the ratings that the user has provided in the past. Here we concentrate on neighbourhood approach. Neighbourhood approach depends on the similarity between items [4] or similarity between users [5], depending on which prediction for an unrated item can be made. The similarity between users or similarity between items can be computed to provide recommendations. Some of the widely used techniques are the Pearson correlation, cosine-based similarity, adjusted cosine, etc. In this thesis a new approach to find similarity between items is used, here the similarity between items is calculated using a modified singularity measure. In this approach, the singularity of ratings provided by each user is taken into consideration [2]. By, using this method recommendation can be found with greater efficiency compared to other existing algorithms as this technique uses the contextual information present in the data.