The Product Care Association (PCA) plays a crucial role in diverting post-consumer lighting products from landfills through recycling programs across multiple Canadian provinces. However, the current product inquiry process—determining whether a product qualifies for recycling and identifying its product category—is largely manual, leading to inefficiencies, inconsistent decision-making, and operational delays. This research aims to develop an AI-powered Product Inquiry Response System to enhance efficiency by leveraging machine learning (ML).
The primary research question is:
· How can I design and implement an end-to-end AI-powered system that effectively integrates text and image data to enhance matching accuracy and minimize response times in the current product inquiry process?
The study hypothesizes that a combined Natural Language Processing (NLP) and Computer Vision (CV) approach will significantly improve matching accuracy and reduce response times by at least 50%. Data is extracted from PCA’s Product Guide (PDF). The system utilizes Sentence-BERT (SBERT) models via SentenceTransformer Library in Hugging Face for text similarity and Convolutional Neural Networks (CNNs) for image comparison. An ensemble model combines weighted text and image similarity scores to match new products accurately.
Initial findings indicate that automating inquiries can enhance operational efficiency, ensuring faster response times and consistency in decision-making. The results have significant implications for scalability, cost savings, and AI-driven process optimization in recycling programs. This project represents PCA’s first ML-powered system, laying the foundation for future AI applications in waste management.
Poster submission was sponsored by Dr. Padmapriya Kandhadai, (Commerce and Business Administration Department) and was presented at the New Westminster campus on April 10, 2025, for Student Research Days 2025.
