CF-CRS: A Content based Fuzzy Conformal Recommender System for Recommending Movies with Confidence

Abstract

A recommender system (RS) provides useful information to customers about products that they are interested in to achieve better customer satisfaction, resulting in increased sales in online environments, such as e-commerce systems. Collaborative filtering (CF) is one of the most commonly used methods for generating recommendations for users. CF selects a subset of users, called a user neighbourhood, to filter recommendations for the current user. In this paper, we present an algorithm that uses CF to provide recommendations to a user by selecting an optimal neighbourhood. Experiments were conducted by selecting two different neighbourhoods: k-nearest and threshold-based neighbours. Recommendation results were generated for selecting different numbers of neighbours and different threshold values for neighbours. The results were compared to determine the neighbourhood that gives the least error and best recommendations. The experiments were conducted on the MovieLens 1M dataset. An RS is an intelligent system that assists users in finding items that interest them (e.g., books, movies, music) so they do not have to sort through the vast amounts of data available online. In an effort to overcome the data sparsity issue in RSs and assign a confidence value to each recommended item, this research incorporates a content-based filtering (CBF) technique with a fuzzy logic system and conformal prediction approach, introducing a new framework called a content-based fuzzy conformal recommender system (CF-CRS). The proposed framework is implemented for use in the domain of movies, and it provides quality recommendations to users with confidence levels and improved accuracy. In our proposed framework, a CBF technique is first applied to create user profiles by considering the history of each user. CBF is useful in scenarios where there is a lack of demographic information and data sparsity problems exist. Secondly, a fuzzy-based technique is incorporated to determine the similarities and differences between a user profile and the movies in the dataset using a set of fuzzy rules to obtain a predicted rating for each movie. Thirdly, a conformal prediction algorithm is implemented to measure the non-conformity between the predicted ratings produced by the fuzzy system and actual ratings from the dataset. A p-value (confidence measure) is computed to assign a level of confidence to each recommended item, and a bound is set on the confidence level, which is called a significance level e, where only the movies

2 above this level are recommended to the user. By building a confidence-centric CRS using the CBF approach with fuzzy logic and conformal prediction algorithm, its reliability and accuracy are considerably enhanced. The proposed RS is evaluated on the MovieLens and Movie Tweetings datasets and is compared with other state-of-the-art RSs. The results confirm that the proposed algorithms perform better than the traditional ones.