End-to-End Natural Language Understanding Pipeline for Bangla Conversational Agents

TL;DR

This paper presents a novel end-to-end pipeline for building a Bangla conversational agent capable of understanding both native Bangla and transliterated English, addressing the low-resource language challenge with custom data and machine learning models.

Contribution

The study develops a comprehensive pipeline for Bangla chatbot development, including data preparation, modeling, and deployment, specifically tailored for low-resource language and transliteration support.

Findings

Achieved 83.02% accuracy in intent classification

Demonstrated effective handling of Bangla and transliterated English

Evaluated different components to optimize performance

Abstract

Chatbots are intelligent software built to be used as a replacement for human interaction. Existing studies typically do not provide enough support for low-resource languages like Bangla. Due to the increasing popularity of social media, we can also see the rise of interactions in Bangla transliteration (mostly in English) among the native Bangla speakers. In this paper, we propose a novel approach to build a Bangla chatbot aimed to be used as a business assistant which can communicate in low-resource languages like Bangla and Bangla Transliteration in English with high confidence consistently. Since annotated data was not available for this purpose, we had to work on the whole machine learning life cycle (data preparation, machine learning modeling, and model deployment) using Rasa Open Source Framework, fastText embeddings, Polyglot embeddings, Flask, and other systems as building blocks. While working with the skewed annotated dataset, we try out different components and pipelines to evaluate which works best and provide possible reasoning behind the observed results. Finally, we present a pipeline for intent classification and entity extraction which achieves reasonable performance (accuracy: 83.02%, precision: 80.82%, recall: 83.02%, F1-score: 80%).