Design & Implementation of Product Recommendation Solution using Sentiment Analysis

E-Commerce portals and online selling websites are becoming popular day by day. This paper presents a product recommender model using the natural language processing of the reviews and feedbacks of the customers to enhance the quality of recommendation module. The descriptive data mining can be used to find the most accurate recommendations based on their preferences and behaviors. Various studies on product recommendation for e-commerce portals are being conducted to improve the selection in the quickest time frame, and it has been found that the majority of recommender system only works on selection frequency and user rating. The user's past buying history, as well as the opinions of other users on a product, can aid the development of trust in an online shopping website. In this dimension, an approach is being used and implemented on a variety of relevant projects to investigate the gap area in the traditional system and potential solutions to close it. This research work considers unstructured dataset as data input and perform data cleaning followed by stop word removal and lemmatization. Afterwards Sentiwordnet and ambiguity word net library has been used to estimate two different sentiment score for same sentence to prepare a hybrid sentimental score based on natural meaning and probability of ambiguous word arrangement. This work also integrates FP Intersect clustering algorithm to improvise searching queries after product recommendation. Proposed solution has been implemented using java technology and hadoop ecosystem to provide a big data infrastructure and consider Amazon dataset for experimental analysis. The complete solution was estimated on basis of computation time and also performed for two different dataset to evaluate the consistency of proposed solution. A significant improvement has been observed in multi node cluster solution in compare to single node cluster setup irrespective of enhancement in data size.