Article
FIGS - Driven Contextual Sentiment Analysis for Large Scale Telecom Conversations
The rapid growth of the telecommunications sector has resulted in massive volumes of customer-agent interaction data, making manual sentiment analysis infeasible. This research presents a robust framework for telecom transcript sentiment detection by combining advanced Natural Language Processing (NLP) techniques, transformer-based embeddings, and ensemble Machine Learning (ML). The system begins with data preprocessing, including text cleaning, tokenization, stopword removal, and lemmatization, followed by exploratory analysis using word clouds, document length distributions, POS tagging, and bigram frequency plots to uncover textual patterns. Google Pathways Language Model (PaLM) like embeddings are then extracted using transformer models to capture rich contextual semantics, and class imbalance is addressed using Random Under Sampler for uniform representation across sentiment classes. Multiple ML models, including Logistic Regression Classifier (LRC), Decision Tree Classifier (DTC), Extra Trees Classifier (ETC), Boosted Rules Classifier (BRC), and a custom Fast Interpretable Greedy-Tree Sums (FIGS) ensemble classifier, are trained and evaluated, with the FIGS model aggregating predictions from base learners to enhance accuracy, robustness, and generalization. The framework supports real-time prediction, model persistence, and visualization of performance metrics, providing interpretable insights for telecom operations. Evaluation results, including accuracy, precision, recall, and F1-score, demonstrate the effectiveness of the proposed approach. The system offers a scalable, efficient, and interpretable solution for automated sentiment detection in telecom transcripts, enabling service providers to improve customer experience, monitor agent performance, and make data-driven operational decisions.
Full Text Attachment
