Product activity prediction using natural language processing techniques and artificial neural networks

This research introduces a novel, data-driven framework designed to enhance the e-commerce experience for consumers. The framework assists users in making more informed and efficient purchasing decisions by providing concise information on product usability and functionality, thereby eliminating the need for consumers to sift through numerous, often confusing, product reviews. While many studies have delved into product review sentiment analysis, none have examined reviews specifically from a usability and functionality perspective, thus leaving this gap unexplored. The proposed prototype framework seeks to bridge this gap by utilising state-of-the-art Machine Learning and Natural Language Processing (NLP) models to predict product usability, thus supporting consumers in making informed purchasing decisions. NLP methods, including stop-word removal, POS tagging, and other feature engineering techniques, have been integrated and applied within the prototype framework to identify the activities performed by consumers in product reviews. A sequence of machine learning processes, encompassing feature engineering, ensemble methods, and artificial neural networks, culminated in the creation of the LSTM-CNN-Glove hybrid neural network model, which features the highest accuracy. The efficacy of the proposed prototype framework was empirically tested using recognised evaluation metrics such as accuracy, precision, and recall. Additionally, the performance of the prototype framework was qualitatively assessed through a questionnaire in a field survey study. An intuitive dashboard that enables users to engage with the proposed prototype framework and provide feedback has been developed. It allows users to contribute to the enhancement of the model’s performance, thus further refining the framework’s capabilities. The findings of this study indicate that the LSTM-CNN-GloVe hybrid model attains an accuracy rate of 83% in predicting product usability. In contrast to current studies, this research offers a novel methodology, synthesizing reviews to determine consumer usage and behaviours, thereby greatly enhancing the online shopping experience. It also contributes to a reduction in online return rates and the carbon footprint which has a positive impact on the environment. Overall, this research contributes significant insights and innovative solutions for the e-commerce domain.