By Gumma Sri Sougandhika & S Hari Sai Ganesh | Published on January 15, 2024
In the rapidly evolving landscape of e-commerce, the prevalence of smart recommendation systems is undeniable. However, their opaque nature often leaves users in the dark, hindering trust and comprehension. Recognizing this critical gap, our project introduces an innovative XAI (Explainable Artificial Intelligence)-enabled, customer-centric recommendation system designed to demystify the recommendation process. Our core objective was to bolster digital literacy by making these systems more transparent and interpretable. To achieve this, we've structured explanations across four progressive levels: Novice, Intermediate, Advanced, and Expert, each offering an increasing depth of information tailored to the user's understanding.
Further enhancing the system, we addressed key limitations identified in current recommendation paradigms. This involved tackling issues of bias, ensuring user privacy, refining user segmentation, establishing robust data pipelining, and improving response generation. We leveraged a Retrieval Augmented Generation (RAG) pipeline for richer explanations, while user segmentation was meticulously performed using a hybrid approach of decision trees and K-means clustering. Bias mitigation was achieved through the use of a generic dataset, and computational efficiency was optimized through the strategic application of transfer learning and fine-tuning. User privacy remains paramount, with personal information rigorously removed to safeguard sensitive data.
The results of our system achieved an accuracy exceeding 0.98 in its recommendations, delivering personalized and explainable insights aligned precisely with individual customer knowledge levels. This project marks a significant step towards creating more trustworthy, user-friendly, and transparent e-commerce experiences.
12
5