Recommender Systems Using Semantic Web Technologies

Recommender Systems Using Semantic Web Technologies

Recommender Systems (RS) have risen in popularity over the years, and their ability to ease decision-making for the user in various domains has made them ubiquitous. However, the sparsity of data continues to be one of the biggest shortcomings of the suggestions offered. Recommendation algorithms typically model user preferences in the form of a profile, which is then used to match user preferences to items of their interest. Consequently, the quality of recommendations is directly related to the level of detail contained in these profiles. Through the review of related literature, it is evident that the genre of a movie is a major factor influencing user decisions about movies. However, the degree of membership of a movie to a genre is typically unavailable. Sometimes, certain genres memberships to a movie might not be assigned at all. Such genre membership information, if available, would provide a better description of items and consequently lead to quality recommendations. To capture complete information on content pertaining to different genre in movies, we have used two approaches ?? one that utilizes the available binary genre information and augments it by inferring the genre degree using the information available in folksonomies and another that does not rely on previous movie categorization but captures genres that manifest automatically when forming keyword clusters. Folksonomies or tags are user-defined metadata for items and embed abundant information about various facets of user likes and their opinions on the quality and the type of object tagged. The degree of genre presence in a movie is inferred by examining the various tags conferred on them by various users. Leveraging on tags to guide the genre degree determination exploits crowd sourcing to enrich item content description. Fuzzy logic naturally models human logic, allowing for the nuanced representation of features of objects and thus is utilized to derive such gradual representation as well as for modelling user profiles. Fuzzy user and object representations are leveraged for the design of both content-based as well as collaborative recommender systems. Experimental evaluations establish the effectiveness of the proposed approaches as compared to other baselines. We call this the Fuzzy User-Based Recommendation Approach (FUBRA). Keywords related to a movie indirectly contain information related to the various narrative styles. User profiles are also constructed based on user preferences for such keyword clusters. We call this the Keyword Clustering-Based Recommendation Approach (KCBRA).These profiles are then utilized to perform both Content-Based (CB) filtering as well as Collaborative Filtering (CF). This approach scores over the direct keyword-matching, genre-based user profiling method and the traditional CF methods under sparse data scenarios as established by various experiments.

Sebastian Mampilli Bonson

Computer Science