With the increasing awareness of environmental protection, the demand for sustainable product design has been an important issue for the past few decades. This research aims to propose an idea development model by leveraging advanced NLP, machine learning, and deep learning techniques. It collects 3,016 posts from sustainability-related social communities to compare the identification performance of seven classifiers. Through model verification using 75–25% criteria, the experimental results reveal that Word2Vec accompanied with SVM performs the best (F1 = 0.72, AUC = 0.75). Based on 1704 creative posts, LDA is adopted to categorize four topics containing 243, 285, 412 and 764 articles, respectively. By employing Apriori and association rules, the discovered knowledge containing 30, 5110, 42 and 103 rules for each topic are extracted as a basis for idea generation conducive to sustainable products development. Discussion and implication are addressed.