Emotion to Purchase Conversion Model for E-commerce System

Abstract

The overwhelming development of e-commerce sites has brought with it greater use of smart personalization to boost customer satisfaction and purchase conversion rates. Historical user behaviour and collaborative filtering are the key methods looked upon to make traditional recommendation system, whenever real-time search intent, emotional signals in product reviews are disregarded. The proposed paper presents a proposal of an Emotion-to-Purchase Conversion Model which will combine intent detection based on Bidirectional Encoder Representations with Transformers (BERT) and sentiment analysis based on a Multinomial Naive Bayes classification with a Random Forest-based fake review filtering mechanism in a hybrid recommendation system. The model lexically derives emotional polarity on the text of product reviews and contextual search intent on the query entered by users which are combined with the help of weighted ensemble ranking technique. Based on Python, Scikit-learn, Transformers, and Streamlit, the system can recreate a real-time e-commerce setting and has explainable AI traits such as voice-based recommendation explanations. The results of the experiment prove an increase in the potential rate of conversion, emotional consistency, and more adequate recommendations in reference to the traditional ones, which can be utilized in order to create more efficient and user-oriented online shopping experiences. The framework will be very flexible to the changing preferences of the users which makes it suitable to intelligent recommendation systems in future