Development of Product Recommendation Engine By Collaborative Filtering and Association Rule Mining Using Machine Learning Algorithms

Recommendation engines are a subclass of information filtering system that seeks to predict the ‘rating’ or ‘preference’ that user would give to an item. It finds information designs in the informational index by learning customers decisions and produces the results that co-identifies with their requirements. Real time examples like Amazon, have been using a recommendation engine for suggesting the goods or products that customers might also like. As the database used in this paper consists of large amount of data, it becomes a difficult and cumbersome process to provide viable choice of products for all the customers. The need of state of art recommendation engine is a necessity in real world e-commerce platforms to solve the issue and fulfil the customers’ needs. There are numerous ways such as collaborative filtering, content-based filtering, hybrid filtering, etc. to build a recommendation system. This paper developed a product recommendation engine that uses collaborative filtering approach, which finds similarity between items bought by the customers with other customers, purchase pattern, and association rule mining framework. The recommendations were generated in order to facilitate ‘cross-sell’ across various items. The collaborative filtering (CF) approach produced the top 10 recommendations for each user. The association rule mining produced the rules based on minimum support (at 0.001) and minimum confidence (at 0.8) values. These values produced around 40,000 viable rules. It can be inferred that selection of metrics and the computation speed is important for quality recommendations.

I. Introduction

The measure of data in this world is expanding more rapidly than our capacity to process it. Users regularly experience challenges in finding the substance they need rapidly, because of the vast measure of data accessible on the web. More often the user endeavors to look for help from other people who had recently had similar requirements for those things or pick up those things that are nearer to what they are searching for and this occasionally brought about serious data overload issue. If one can prescribe two or three things to clients contingent upon their prerequisites and interests, it will make a beneficial outcome on the client experience and lead to more visits [5]. In this way, organizations nowadays are building splendid and vigilant recommendation engines by looking into on the past transaction of their customers. A recommendation engine channels the information with the assistance of various calculations and prescribes the most applicable things to clients. It catches the past conduct of a client and relying upon that, prescribes items which the prone to purchase. Recommendation engines help organizations in executing balanced marketing methodologies, depending on client purchase history to uncover client inclinations and recognize items that clients may purchase. Traditionally, there are two methods to construct a recommender system: Collaborative filtering and content-based recommendation. In the principal technique, the recommender works with information that the client gives, either unequivocally (evaluating) or verifiably (tapping on a connection). In view of that information, a client profile is created, which is then used to make proposals to the client [3]. As the client gives more sources of information or takes activities on the suggestions, the engine turns out to be increasingly precise. In the second approach, recommendation is done based on users’ past behavior. Based on different similarity measures the top similar items are selected and recommended to the desired user. This paper develops a product recommender engine which can generate quality recommendations for the customers. The methods that will be used with respect to the project are distance metric based collaborative filtering, association rule mining and matrix factorization based collaborative filtering.