Semantic Clustering Driven Approaches to Recommender Systems

Abstract

Recommender Systems (RS) have increasingly evolved from novelties used by few E-commerce sites to an essential component of business tools handling the world of E-commerce. Recommender Systems have been widely used for product recommendations such as books and movies as well as, it is also gaining ground in service recommendations such as hotels, restaurants and travel attractions. Collaborative filtering based on reviews and ratings is usually applied that uses Clustering technique. The primary step of converting textual reviews into a Feature Matrix (FM) can be greatly refined by using semantic similarity between terms. In this paper Wordnet based Synset grouping approach is presented that not only reduces dimensions in FM but also generates Feature vectors (FV) for each cluster with significantly improved cluster quality. The paper presents a three step approach of validating the reviews, grouping of reviews and review based recommendations using Feature vector. Real datasets extracted from travel sites are used for experiments.