Cross Domain Item Based Book Recommendation System

Abstract

Recommendation systems (RCS) are particularly extremely resourceful systems which can anticipate the type for content the user would like to stream, rate, upload, download or subscribe to. It is based on feedback process, where it tracks the type of content a user is rating, streaming, downloading, uploading and subscribing to and based on that, Recommendation system further recommends the user the type of content they would further like to stream, rate, download, upload or subscribe to. Basically Recommendation system keeps up with the appetite of the user by satisfying their needs online. In the modern world all the giant companies for example like Facebook, YouTube, Netflix, Spotify, Instagram, Amazon, Ebay, Ali Baba (companies from Entertainment Industry, Telecommunication industry and E-commerce) are heavily relying on Recommendation systems for their businesses. The Aforementioned Recommendation systems are single domain Recommendation system, meaning the Source domain and the Target domain is the same. For example if a user is listening to a song on YouTube of a particular singer and likes it, the user will be further recommended more videos from that singer. So here our source and target domains are the same, which is YouTube videos. When we talk about Cross domain, here Our Source and Target are two completely different domains. We takes users from Source domain and recommend them things from target domain. For Cross Domain to work, there has to be some sort of link between them. In this thesis, a Cross domain item based book recommendation system is proposed. The proposed technique is based on initially building a user item rating matrix for movies, then using KNN algorithm to make movie recommendations, then taking recommended movie genres and computing their semantic similarity with the books genres using wpath method. Then those books genres are shortlisted which have semantic similarity score of more than 0.5 with the movie genres. Lastly Multi label Binarizer approach is used to break a book into its genres and make sequences of the books. For final recommendations three things are taken into account, 1) No of times book occurred in a sequence 2) Total rating count of the book 3) Average ratings of the book. This Cross Domain Item based Book Recommendation system is capable of providing better recommendations but also establishes strong relation between source and target domains. The output results and their comparison with other different techniques is provided which shows the overall improved results of this approach.