Context Aware Recommender Systems using Deep Neural Network

Abstract

Recommender systems is a class of predictive models that aim at selecting and proposing most relevant items, services or offers to the users based upon their history. These systems aim at enhancing the user’s experience based upon what they enjoyed in past. For example, Amazon, Netflix, YouTube, News, etc. As more and more data are being generated, it is the need of the hour to generate user specific recommendations. For example, application suggesting user to purchase items in user’s checklist by identifying the location (shopping complex) of the user. These applications generating the recommendations based on the context of the user are known as Context Aware Recommender Systems (CARS). Collecting the context of individual users is not a challenging task as many sensor devices are available at affordable costs. However, analysis of the context, is a complicated task as there may be many repeated, irrelevant, and contradicted context. Many data analysis techniques could be used to analyse the context data such as meta-heuristics algorithms, decision making algorithms and others. The problem with these techniques is high convergence and inference rates, leading to latency issues. Recommendations based on the context is a time sensitive task because if the context changes, the recommendations would be ineffective. Another problem is these techniques perform the context engineering and predictions as two separate tasks. So, in this thesis analysis of contextual and non-contextual data and then making recommendations is proposed to be performed using Deep Neural Networks (DNN). DNNs can perform these tasks as a single entity and with experience these systems perform more accurately. Now, contextual, and non-context data analysis using DNN to make recommendations have several issues that have been addressed in this thesis. Firstly, analysing and removing the irrelevant contexts by dimension reduction while retaining the meaningful properties. Secondly, analysing the contextual and non-contextual data collectively. Thirdly, to identify the contextual relationships and sharing them to other DNNs for effective recommendation generations.