Improving the Performance of an Artificial Intelligence Recommendation Engine with Deep Learning Neural Nets

The most valuable artificial intelligence application for e-commerce to social media websites these days is a smart recommendation engine that can filter panoply of information on the internet and recommend personalized products and services to each user. An efficient and reliable recommender engine (RE) increases sells and profit of the e-commerce websites, thus its performance is very crucial. Traditional RE suffers from cold-start, low accuracy, and scalability to Big Data problem. Thus, RE research has started again with great enthusiasm to explore newer techniques with deep learning artificial neural nets, as more computing power in parallel processing framework become available from latter half of this decade. In recent years deep learning (DL) artificial neural nets (ANN) have given breakthrough performance in areas like image processing and natural language processing tasks. So, its usability needs to be researched for recommender engine design also. This paper first explores the traditional ways of making a recommender engine and then evaluates the use of deep learning neural net techniques. The first contribution of this paper is expounding the theoretical foundation of different ways a RE can be built viz. content-based filtering (CBF) and collaborative filtering (CF) that comprises of complex algorithms. The second contribution of this paper is practical experiments with both traditional linear algebra techniques and deep learning auto-encoder architecture on large Movie-Lens dataset. Comparisons of the result shows deep learning methods outperform traditional methods.