Topological Mapping of Organised Fruit and Vegetable Retail Discourse in West Bengal: A Computational Corpus Analytics Approach

Abstract

This study, conducted in February, 2026, investigated the semantic structure of organised fruit and vegetable retail chains in West Bengal using Natural Language Processing (NLP) and Topological Data Analysis (TDA). The rapid expansion of organised retail chains in India significantly transformed food marketing systems, particularly in regions where traditional and modern retail formats coexisted. Understanding the interaction among product quality, consumer perception, branding strategies, and supply chain dynamics within retail discourse, therefore, became increasingly important. A structured textual corpus describing store characteristics, product range, consumer perceptions, and improvement strategies was processed through a systematic preprocessing framework. Contextual embeddings were generated using transformer-based language models, producing 768-dimensional semantic vectors that captured contextual relationships within the corpus. These embeddings were reduced using Uniform Manifold Approximation and Projection (UMAP), followed by the application of the Mapper algorithm to construct a topological network of semantic clusters. The resulting Mapper graph revealed a thematically segmented yet structurally interpretable discourse network with distinct conceptual clusters and limited transitional overlap. Network refinement identified a central connected component representing the intersection of branding strategies, product positioning, consumer evaluation, supply relationships, and market dynamics. Topic modelling, coherence analysis, hyperparameter stability assessment, and spectral analysis confirmed the robustness and interpretability of the semantic structure. The findings indicated that organised retail discourse was predominantly consumer-centric, emphasising product quality, pricing perceptions, and brand-related considerations. The study demonstrated the effectiveness of integrating NLP and topological modelling to uncover hidden semantic structures in retail-related text and provided valuable insights for retail managers, researchers, and policymakers.