WIPE: A Novel Web-Based Intelligent Packaging Evaluation via Machine Learning and Association Mining

Aim:

            Our study aims to introduce the Web-Based Intelligent Packaging Evaluation (WIPE) platform, which uses machine learning and association rule mining to assess packaging performance in e-commerce. By analyzing customer reviews, WIPE identifies packaging defects, their causes, and effects, offering a dynamic, real-world alternative to traditional laboratory methods.

 Abstract:

        The rapid growth of e-commerce has introduced unique challenges to packaging systems, such as increased handling points and unpredictable hazards, which traditional laboratory-based evaluation methods fail to fully address. This paper presents the Web-Based Intelligent Packaging Evaluation (WIPE) platform, a novel solution designed to assess the performance of product and packaging systems in the e-commerce sector. WIPE employs advanced machine learning algorithms and association rule mining to analyze customer reviews on e-commerce platforms, identifying patterns that link packaging defects to their causes and potential impacts.

      Through two case studies focusing on laundry detergent liquid bottles and pods sold on Amazon, WIPE demonstrated its ability to extract actionable insights, uncover specific packaging flaws, and predict their causes. By integrating sentiment analysis and data-driven evaluation, WIPE represents a transformative approach to packaging assessment, enabling real-world, customer-centric insights that can enhance product design and improve customer satisfaction in the dynamic e-commerce landscape.

Existing System:

        The existing system for packaging evaluation combines traditional methods with advanced machine learning techniques to assess packaging performance in the e-commerce sector. Traditional approaches, including field testing, laboratory testing, and computational modeling, provide valuable insights but face challenges such as high costs, lack of repeatability, and limited real-world applicability.

      Recently, artificial intelligence-based methods, including sentiment analysis using algorithms like LSTM (Long Short-Term Memory), Naïve Bayes, and AFINN, as well as association rule mining with FP-growth, have been introduced to automate defect detection, optimize packaging, and uncover patterns in customer feedback. These AI methods improve analysis accuracy and scalability, but they still face limitations in data quality and contextual understanding. The combination of these approaches offers a more dynamic, though still evolving, and solution for packaging evaluation.

Proposing System:

          In our proposed system, we utilized real-time Amazon review data, leveraging a pre-trained BERT model for sentiment classification with an impressive accuracy of 95.5%. The system scrapes Amazon reviews in real-time, classifies them as positive or negative using BERT, and generates dynamic word clouds based on customer feedback, providing insights into common packaging or product issues. It supports both individual and batch predictions, offering scalable sentiment analysis for large datasets, and uses a majority voting mechanism for batch predictions.

       In addition, the system allows live scraping of any Amazon review, providing immediate sentiment classification (positive or negative) using BERT, and generating dynamic word clouds based on the review content. A set of Amazon dataset reviews can be processed at once, with the results displayed accordingly, and individual reviews from the dataset can also be passed through the system to generate sentiment predictions and visualizations. With a REST API for easy integration, this system provides actionable insights for businesses to improve product designs and packaging by analyzing up-to-date customer sentiment and trends.

 Advantages:

• Unlike costly and time-consuming field testing, laboratory testing, and computational modeling, the system provides real-time, scalable sentiment analysis that can process large volumes of Amazon reviews efficiently.
• By using a pre-trained BERT, it ensures precise predictions even with varying data quality. Additionally, the system captures the full context of customer feedback by generating dynamic word clouds, which visually represent common issues and sentiments, offering deeper insights into customer concerns.
• This approach improves data quality and contextual understanding, enhancing the real-world applicability of the results and providing businesses with actionable insights to optimize product and packaging designs.