Recommendation engines-neural embedding to graph-based: Techniques and evaluations

The goal of any profit organization is to bolster its revenue by providing useful suggestions to its customer base. In order to achieve this, vast research is being undertaken by companies such as Netflix and Amazon on their Recommendation Systems and providing users with choices, they are most likely to click on. The purpose of this paper is to provide a holistic view of types of Recommendation Engines and how they are implemented, scaled and can provide a basis for revenue generation. The focus would be to implement a Recommendation Engine on PySpark using the ALS (Alternate Least Square) method. Besides, Neo-4j and Cypher query language for implementing recommendations on a graph database and analyzing how heterogeneous information can be levied to tackle the infamous cold start problem in recommender engines would be explored. The dataset used for analysis is the Group-lens 100K Movie-lens dataset and the algorithm is implemented to best fit the dataset. Further, an in-depth comparison of several techniques has been carried out on the basis of different metrics, hyper-parameter selection and the number of epochs used. The claims have been justified by evaluating the performance of the model depending on the different use cases, thus aiding in predictive analytics of the movie, as per the interest of the customer using visualization tools.