Content-Based Image Retrieval and Recommendation Based on User Reviews

Abstract

Effective search functionality is crucial for enhancing user experience and boosting sales on e-commerce sites. However, relying on specific product names instead of common names can cause potential customers to be lost while searching for products, as many platforms still depend on text-based search engines. While text searches effectively find keywords, Search Engine fall short when customers do not know the exact product identity or only have an image of desired product. With the rise of multiple E-commerce sites, customers often feel confused about where to buy the best product. This paper proposes a novel architecture for recommending the desired product based on image search. The input image provided by the user is processed through YOLO models to extract the brand, category, and color of the shoe, which serve as keywords. Data about shoe with similar keywords is retrieved from multiple e-commerce sites using APIs. The SIFT (Scale-Invariant Feature Transform) algorithm is employed to compare the input image with the extracted images and assign a score. Additionally, reviews, ratings, and price analyses are conducted on the extracted data, and scores are assigned accordingly. The TextBlob library is used for sentiment analysis of reviews. Based on these scores, the best product is recommended to the customer. The architecture is deployed on an EC2 instance on the AWS platform, demonstrating its practical applicability. The result is a model that is able to suggest similar product wanted by the user which has the most positive reviews and ratings from other users and has the most competitive price in the market. The project suggests an easier and comparatively better way to improve the online shopping experience of users by providing what they want which is liked by others as well and are priced competitively.