A Multilabel Text Classification and Association Rule Mining Framework for Data-Driven Skincare Product Name Generation in E-Commerce
Abstrak
The rapid expansion of e-commerce has produced large volumes of unstructured product-name data, particularly in the skincare sector, where a single name often conveys multiple functions. This study proposes an integrated framework combining multilabel text classification and association rule mining to extract structured knowledge from skincare product names and generate data-driven naming recommendations. Using product names scraped from Tokopedia, the dataset was preprocessed through text cleaning, TF-IDF vectorization, and multilabel binarization. Three multilabel classification strategies, Binary Relevance, Classifier Chain, and Label Powerset, were evaluated with Support Vector Machine (SVM), Logistic Regression, Random Forest, and K-Nearest Neighbor algorithms under 60:40, 75:25, and 80:20 train–test splits. According to the experimental results, Binary Relevance achieved the best performance across all strategies, with an F1-test score of 0.925 for the SVM model at an 80:20 split, followed by Random Forest (0.912) and Logistic Regression (0.883). Classifier Chain had consistent results across all, still slightly lower, but robust findings, while Label Powerset obtained the lowest results due to label sparsity. The predicted labels were further explored using the Apriori algorithm, which revealed strong co-occurrence patterns, in particular, moisturizers were a dominant component of product bundles. High-confidence rules (0.99) and strong lift values (2.73) provide evidence of a significant association in the dataset. These insights formed the basis for proposing new naming themes such as Daily Radiance, Glow Hydration, UV Protection, and Clean & Fresh. Overall, the study demonstrates that combining multilabel classification with association rule mining is effective for uncovering naming structures in skincare e-commerce data.
Identifier
- OpenAlex: https://openalex.org/W7169846803
- DOI: https://doi.org/10.67561/jaids.v1i2.4