Machine learning model to design more sustainable products

Context/Motivation

A massive number of product reviews is posted online every day. This is both a convenient way for customers to make their voices heard, and an opportunity for designers to improve the features of their products on this basis. In the meantime, the recent advances in IT tools (e.g., web scrapping) and machine learning techniques (e.g., natural language processing) enable researchers and industrialists to extract, proceed, and analyze large datasets of online reviews. 

Yet, as of now, few studies have tried to link product sustainability features and online reviews, and present several shortcomings, notably in terms of sustainable design implications. This is a lack of automatic procedures or algorithms to identify complex topics in reviews like product sustainability (Saidani et al., 2021a). A recent study, manually extracting and interpreting a set of 100+ reviews on three different products, concludes that around 15-20% of these reviews contain valuable information that could be used to improve the sustainability of products (Saidani et al., 2021b).

For instance, here are some illustrations of sustainable design learning that could be elicited from product reviews:

• Positive/negative perceptions of sustainable features (e.g., use of plastic, unsafe to a user, like “glass cracking”);
• Non-sustainable use (cases/patterns) of a product supposed to be sustainable (wasted potential);
• Life duration, wear and tear (e.g., an early failure) of specific components/parts.

Objectives

The main objective of this project is to automate this process, i.e., to build the machine learning pipeline to scrap 10000+ reviews in order to generate sustainable design insights from big data (here, online product reviews). 

While there are existing tools or packages for online review scrapping (data collection) and preprocessing (data preparation), the focus of this project is on: 

• (i) the automatic identification/classification of product reviews that contain sustainability-related information;
• and (ii) the linkage between valuable sustainability-related data and key/sensitive product features;
• to (iii) extract/generate sustainable product design leads or insights from these reviews. 

References

• Saidani, M., Kim, H., & Yannou, B. (2021a, August). Can Machine Learning Tools Support the Identification of Sustainable Design Leads From Product Reviews? Opportunities and Challenges. In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (Vol. 85383, p. V03AT03A005). American Society of Mechanical Engineers.
• Saidani, M., Kim, H., Ayadhi, N., & Yannou, B. (2021b, August). Can Online Customer Reviews Help Design More Sustainable Products? A Preliminary Study on Amazon Climate Pledge Friendly Products. In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (Vol. 85420, p. V006T06A002). American Society of Mechanical Engineers.