Improving recommendation quality by identifying more similar neighbours in a collaborative filtering mechanism

Recommender systems (RS) act as an information filtering technology to ease the decision-making process of online consumers. Of all the known recommendation techniques, collaborative filtering (CF) remains the most popular. CF mechanism is based on the principle of word-of-mouth communication between like-minded users who share similar historical rating preferences for a common set of items. Traditionally, only those like-minded or similar users of the given user are selected as neighbours who have rated the item under consideration. Resultantly, the similarity strength of neighbours deteriorates as the most similar users may not have rated that item. This paper proposes a new approach for neighbourhood formation by selecting more similar neighbours who have not necessarily rated the item under consideration. Owing to data sparsity, most of the selected neighbours have missing ratings which are predicted using a unique algorithm adopting item based regression. The efficacy of the proposed approach remains superior over existing methods.