Personalized Recommendation System with User Interaction based on LMF and Popularity Model

With the popularization of social media and similar web-based applications, the information generated by the users of these applications is growing exponentially. Thus, more attention is required on the “recommender system”. The recommender system fails to determine the real-time user interest. This paper uses an instant input mechanism to overcome this problem i.e. based on user interaction, a recommender system determines users' real-time interest. This paper uses “Significance analysis” to refine the recommender process performance. The most successful predictions are given to the client during communication to assess the real-time desires of the users and as soon as the model identifies the user it determines the expected recommendations for that user, thus, providing “Personalization of expectations.” There are two problems in the recommender scheme “cold-start” or “potential-false termination”. This problem is dealt using the user interaction which provides interactive recommendation although the system doesn't have any information related to the user. Potential-false dismissal occurs due to “tag sparsity”. To overcome this problem, associations between various tags are taken into account, which refine the recommendations. The paper uses “low rank matrix factorization (LMF)” approach to determine the potential interests and personalize the recommendations during each iteration. This method overcomes the rating sparsity and is enhanced by embedding the similar user and resource information. For new users, real time input is used to recommend similar items.

I. Introduction

This research uses machine learning to build a custom movie scoring and recommendation framework based on previous movie reviews of the consumer. In movies, different people have different preferences, and that's not mirrored in a single score which is seen when a user search for a movie. The film scoring system helps users to quickly discover movies whatever their preferences may be. There are typically two types of existing recommender systems: content-based sorting and cooperative filtering. Among several method one method is tested in available venture, i.e. collaborative filtering. Upon going through some generic research papers it has been observed that collaborative filtering in terms of estimation error and computation time performs better than content-based filtering.