Towards a Supervised Topic Model Based on Searching of Semantic Center of Mass

Abstract

In the age of digital transformation, managing and extracting meaningful topics from large, diverse document collections is increasingly important. This paper introduces the Understanding Map Supervised Topic Model (UM-S-TM), a novel approach that integrates document content with a semantic network–specifically, an Understanding Map–to derive topics. Inspired by the physical concept of the center of mass, we define a “semantic center of mass” (SCOM) to represent a document’s abstract topic. Unlike conventional probabilistic models, UM-S-TM avoids assumptions such as i.i.d. distributions and prior probabilities. We demonstrate the effectiveness of the model through experiments on artificial documents and semantic networks. Results indicate the method’s potential for accurate and interpretable topic extraction. The proposed model is also applicable to document analysis in cyber-physical systems, such as smart manufacturing.