The literature on the theory of public procurement points out two well-known informational problems arising out of information asymmetry: (i) adverse selection and (ii) moral hazard. To reduce these issues and foster credibility and trust in the procurement process while maintaining quality and efficiency in public procurement, e-procurement platforms have turned to reputation or rating systems. Therefore, the research and design of such rating systems are crucial. In this study, we discuss the theoretical underpinnings of procurement and employ the information-theoretic, regression analysis, artificial neural network and principal component analysis (PCA) approaches to estimate the weights of the variables entering the rating system. Using real data from Government e-Marketplace, a business-to-business public e-commerce portal, we empirically determine the weights of the rating variables derived from the transaction-level and user feedback data for sellers. The weights obtained from the PCA are the most applicable compared with the other three methods. We compare the old rating system with the newly proposed design using the Wilcoxon signed-rank test. This results in a statistically significant difference between the two ratings. The canonical correlation and Wilks' trial reveal that the ratings derived from transaction-level data and user feedback are uncorrelated to a great extent. Hence, considering only transaction-level data or user feedback is unlikely to divulge sellers' intrinsic worth. E-commerce platforms can use this approach to quickly implement methods to obtain rating scores on a real-time basis for sellers on online platforms.