Optimization of Chatbot Intent Classification: a Comparative Analysis of NLP and LLM Embedding Techniques

In the field of natural language processing (NLP) and large language models (LLMS), a comprehensive study has been conducted on a number of machine learning models and methods for extracting, understanding, and generating textual information. The main focus was on using the capabilities of models of different generations to solve a classification problem using machine learning, which was trained, validated and tested on synthetically generated text data from ChatGPT the GPT-4o version.
The experiments began by evaluating basic models such as TF-IDF, Word2Vec, Doc2Vec, FastText, and ended up with Transformer-based models. These models were used to extract embeddings from text data, which laid the foundation for subsequent intent classification. This initial stage of the experiments allowed to obtain valuable information about the effectiveness of various embedding methods and their impact on the subsequent Chatbot classification task. This article describes an optimization through these various stages of experimentation and learning, culminating in a holistic understanding of the various NLP and LLM models and their applications.