Semantic Approach for Multifarious Ranked Novelty and Diversity in News Recommendations

Abstract

Information overload on web creates lots of inconvenience for end users to unearth requisite information which instigate the demand of personalized recommendations. Recommender systems strive to attain just accuracy in recommendations based on the history of user preferences, resulting in over specialized recommendations. This leads to gradual loss of user’s interest in the system. These approaches also fail to recommend other products in long list which could be of user interest but user is not aware of. This in-turn leads to sparse user profiles.In dynamic domains, novelty and diversity in coverage can handle over specialization problem. To bring useful novelty and diversity, it is proposed to semantically expand the user preferences. For this purpose our designed multi level ontology is used with linkages among concepts, entities and properties, carrying different weights. Ontology is also enriched with different relations from online semantic lexicon and annotated with additional information about entities from open and linked external knowledge source.For relevant accurate recommendations in highly dynamic domains, we propose semantic approach to implicitly capture individual user preferences in separate profiles. One profile is to capture short term dynamic interests with temporal effects and another is to capture long term static interests. Semantic profiling helps to handle preferences in manageable number of news categories which are arranged hierarchically. It also helps to handle synonymy and polysemy problems, reducing ambiguity in profiling. An additional important step of outlier analysis and rectification is proposed to handle the effects of sudden unexpected temporary drifts in dynamic user preferences. This focused analysis helps to maintain correct ranking of user preferences which is generally overlooked. The rectified ranking of user interests is fetched for semantic expansion of user preferences, to bring correct ranking in diversity and novelty.Proposed approach semantically brings reasonable multifarious novelty and diversity by analyzing different features of item. Testing is based on live data gathered from RSS feeds of popular news providers for trustworthiness. Transparency is achieved by presenting new recommendations in separate sections along with explicit trace path options, based on difference in approaches. Transparency assists end user in making decision. User’s interests in diverse and novel recommendations are updated in their individual profiles, resulting in reduction of sparsity.