In Depth Analysis of Semantic Similarity for Context Attributes in Recommender Systems

摘要

Internet has perhaps been the most outstanding innovation and technological marvel in thefield of ICT in last couple of decades; huge amount of information and content is available in almost alldomains and is ever expanding gigantically across dimensions. On the other hand, as a disadvantageousafter effect, this uncontrolled proliferation has resulted in data overloading problem. Recommendersystems have been designed in-order to overcome the data overloading problem that exists today inWorld Wide Web, by aiding the users towards seamlessly narrowing down to the required informationand discard the unwanted ones. Research output demonstrate that context aware recommender systemsare useful and enhances the prediction accuracy when context parameters are induced appropriately, butif contexts are not properly assimilated, the objectives of context aware recommender system are not met,rather it gives rise to unwanted complications and reduces the quality of output. In this paper, we discussand analyze the semantic similarity of context attributes of recommender system towards increasing theprediction accuracy and overcoming data sparseness. The context attributes, in many cases aremeaningfully similar or semantically closer within a given knowledge domain; in such situations, thesesemantically closer attributes can be consciously be considered and exploited for further processing andthereby enhancing the veracity of Recommender system. A hybrid method consisting of both structurebased approach and weighted feature based approach is proposed and analyzed here for determining thesemantic similarity and its effect on the quality of Recommender system is also analysed.