8th International Conference on Electrical and Electronics Engineering (ICEEE), Antalya, Turkey, 9 - 11 April 2021, pp.254-258
According to the expansion of users and the variety of products in the World Wide Web, users have been surrounded by a huge amount of data and information, so without proper guidance and navigation, they may make wrong or non-optimal choices. Recommender systems (RS) are useful in guiding the user to reach his/her favorite option among a huge volume of possible choices, so this process is specific to that user. Collaborative Filtering (CF) recommender system is one of the most popular approaches that exploits information about the past behavior or the opinions of an existing user community for predicting which items the current user of the system will most probably like or dislike. This paper improves Item-based collaborative filtering recommender system by finding similar items based on their content. Our main objectives are to solve the item cold start problem and to improve the quality of user's recommended list, or in other words, to improve the items ranking. To achieve these goals, the latent semantic features of the items have been first extracted using a Convolutional Neural Network (CNN) and, then the semantic similarity between the items has been calculated and finally used in the numerical prediction, indicating to what degree the current user will like or be interested in a certain item. Our experiments on the Jester Dataset3 and Jester Dataset4 show that the proposed method has not only been effective in solving the above two problems but has also improved the ratings prediction accuracy and the recommendation quality.