Ontology Based Recommender System using Fuzzy Clustering Technique

Abstract

Recommender systems (RS) are the agents of information filtering processes. With a myriad of data in the world wide web, reaching the exact entity of interest is the need of the hour. Collaborative filtering is a type of RS concerned with studying past data to predict assumptions for the future. In this work, the collaborative filtering recommender is improved by incorporating ontology and fuzzy clustering algorithms to provide ranked recommendations. Ontology models semantic information from raw natural language data to better represent domain knowledge and user preferences. In this way, data sparsity problem could be sorted. The proposed recommender framework runs on top of a historical data that is not only semantic but also represents users preferences perfectly. Another problem of cold start can also be solved by including maximum expectation strategy by constructing the user ontology and domain ontology separately and mapping it to achieve a unified recommender ontology. Using fuzzy clustering introduce a degree of membership for each user with their interests. It also allows for attributes to be distributed in more than one cluster, thus enabling users to be clustered in the most inclusive manner possible. Which provides for ranked recommendations that represent diversified interests of a single user. The dynamic framework is compared with baseline models. Experimental results have shown significant improvement in the recommendation accuracy.