Applying Pairwise Association Rules For Recommendation

Abstract

Recommendation systems are crucial in suggesting items to users which increase user experiences. Commonly, recommender systems are based on content-based and collaborative filtering strategies. However, the two mentioned methods depend on user profiles and item preferences as well as on a broad history of client inclinations. Such strategies confront a number of challenges: counting the cold-start issue [18, 20] in frameworks characterized by security concerns, and settings where the extend of markers speaking to client interface is restricted. Therefore, recommender systems that based on Pairwise Association Rules are portrayed as the recommender calculation that builds a demonstrate of collective inclinations independently of individual client interface which reduce the relying on knowledge of content [12] and does not require a complex framework of evaluations. The execution of the algorithm is analyzed on a huge value-based information set produced by real-world dietary admissions review application. By applied PAR to the assessment, performed on a huge information set of genuine dietary recalls, the result clearly proved that the new algorithm is definitely performed way better compares to other approaches. Nevertheless, user experiences are considered to be better day by day, this thesis aims to develop a more efficient version of Pairwise Association Rules, which focus on re-ranking the order of suggestions by applying rating of the item to the calculation. The expected result of proposed algorithm will prove that the new approach gives approximate value of precision and recall to the original PAR but the quality of each prediction is better for users.