RECOMMENDATION SYSTEMS USING DEEP LEARNING METHODS

Abstract

Recommendation systems (RS) are quite useful these days as they offer the content according to the users’ tastes and interests. They are used in every online platform like social media, e-commerce, streaming services, news and content websites, traveling sites, job portals, online advertising, food delivery and restaurant apps, online learning platforms, dating and marriage related apps and many more. Some examples of recommendations include friend suggestions and news feed content on Facebook, job recommendation and content sharing on LinkedIn, movie recommendation on Netflix, products recommendation on e-commerce sites, hotels and flights recommendation on traveling sites, courses and learning material recommendation on E-learning platforms like Coursera, Udemy. Recommendation systems are mainly of three types: Content based RS, Collaborative-Filtering based RS (CFRS) and Hybrid Systems. Content-based RS are related to only one individual user and recommend items related to items’ description and the user’s personal choice/interests. Whereas, Collaborative RS creates a matrix of user-item pairs, which has each users’ ratings for liked items. Now, a user gets the recommendation of items based on his interests as well as the based on the others users’ interests with the similar profiles. Both Content based RS and CFRS have their own disadvantages. So, to overcome these disadvantages, hybrid systems take individual output of both content based RS and CFRS and then combine these outputs to make recommendations. With the recent advancements in technology, Deep Learning (DL) models handle RS effectively. They are capable of handling intricate structures of data like image, text, audio, video and learn complex patterns from this data. This ability of DL to handle high dimensional data and learn hierarchical representation of features makes it suitable to be applied in RS. DL time-series models like RNN, LSTM have the capability to capture users’ dynamic interests and their evolving temporal preferences. Such models help in capturing the changing needs of the users and make the recommendations accordingly.