Modeling and Visualization of E-commerce Sellers' Ratings Using Transaction-level and Feedback Data

The paper investigates the sellers' ratings of e-commerce platforms. E-commerce platforms require a robust rating system for the sellers to build their credibility and establish trust in the procurement process. We combine the transaction-level data and feedback scores to develop a model for the computation of ratings. Using a real dataset from the Government e-Marketplace and employing a parametric dimension reduction technique, principal component analysis, we obtain weights of three transaction-level and two feedback variables. The canonical correlation and the Wilks' test reveal that the rating derived from transaction-level da-ta and users' feedback is uncorrelated to a great extent. Hence, we conclude that considering only transactional-level data or only feedback score is unlikely to reveal the sellers’ intrinsic rating. The t-distributed stochastic neighbor embedding (t-SNE) visualizes the data using labels obtained from the k-means clustering. The t-SNE visualization with k-means labels demonstrates distinct groups akin to the five-star rating system. The findings can help e-commerce professionals better understand the sellers’ ratings for making decisions