Indigenous Knowledge Systems and Machine Learning: Evaluating the Suitability of Colon Classification

Abstract

This study explores the application of machine learning (ML) techniques to automate subject classification using S.R. Ranganathan’s Colon Classification (CC - 6^th edition), a faceted system rooted in traditional Indian knowledge frameworks. The research integrates the Colon Classification scheme with Annif, an open-source AI/ML-based subject indexing tool developed by the National Library of Finland to predict main class of a text corpus based on Colon Classification 6^th edition. A curated dataset of nearly 100,000 English-language bibliographic records (with CC 6^th notation) from the Indian National Bibliography (INB) was used for model training. The study evaluates the performance of several machine learning backends—fastText, Omikuji (Bonsai), and Support Vector Classification (SVC) - as well as two ensemble models, including a neural network-based ensemble created through hyperparameter optimization. Key retrieval metrics such as F1@5 and NDCG were used to assess models efficacy. Among the tested models, the neural network ensemble achieved the highest scores, with F1@5 = 0.5873 and NDCG = 0.9473, showing strong accuracy in both prediction and ranking. The study demonstrates that machine learning can effectively support traditional classification systems when backed by well-structured data. Finally, through REST/API integration, the framework enables scalable classification, allowing real-time, automated processing of large bibliographic corpora. This work bridges indigenous classification logic with modern AI, contributing to more inclusive knowledge organization systems.