Problem
This project is a replication of a real Tokopedia project presented at the START Summit Extension on February 24, 2022, titled “How We Leverage AI to Match Products.” It aims to improve the customer buying experience, which can sometimes be challenging—for example, searching for “KF 95 Masker” may return results where only 8 out of 10 products are actually related to KF 95, while the rest are KF94.
Result
Before starting the product cataloging process, we cleaned and transformed the data with the following steps:
- Dropped missing values
- Normalized and transformed features
- Combined relevant features into one
Initially, most products were grouped into a single-member catalog, which had a Silhouette Score of 0.1828671.
To improve clustering quality, we applied two re-clustering methods:
Member-to-Member re-clustering → Silhouette Score: 0.10205004
Member-to-Centroid re-clustering → Silhouette Score: 0.12488353