Smart Product Recommendation System

Abstract

The world of product recommendation is growing rapidly. With the rapid increase in the number of products available, it becomes increasingly important for people to know what is best for their needs. A recommendation system is a computerized system for recommending products to customers. It uses data and algorithms to determine the best products to recommend, based on the user's preferences. The term is also used to describe the process of determining which products to recommend by analyzing user behavior patterns or other data. We aim to develop a smart recommender system that will be able to recommend and predict the products that the user most likely will purchase based on several recommendation methods including collaborative filtering, the popularity-based method, and content-based method. This work relied on the cosine similarity metric to find the nearest neighbors when applied to character-type data in machine learning due to its dynamic ability to adapt to different data features. Cosine similarity is employed in textual data to find the similarity between the vectorized texts and original text document. The Popularity-based approach provided an accuracy of 99.9% when recommending the most popular items. The Content-based recommendation system reached a 96% accuracy rate when tested on the authors’ website. They can handle circumstances in which various users do not share the same products, but only identical items based on their fundamental characteristics. Finally, the Collaborative-based approach resulted in weak suggestions with only 87% accuracy rate when tested on authors’ website.