bert-base-spanish-wwm-cased-finetuned-NLP-IE-5

Chat Classification & Fine-Tuning Model

IE University Capstone Project

Developed as part of IE University's capstone project for advanced NLP classification and fine-tuning techniques.

Project Description

This repository contains our work on fine-tuning large language models for chat classification tasks. We explored various approaches to improve model performance on conversational data, including custom training pipelines and optimization techniques.

Technical Implementation

• Fine-tuned transformer models for multi-class chat classification
• Custom data preprocessing pipeline for conversational data
• Implemented various optimization techniques to improve accuracy
• Achieved 94% accuracy on test dataset

Applications

During our research, we found these techniques particularly useful for educational technology applications. in platforms like LectureScribe.io, to help students classify, automatically transcribe and format their lecture notes.

Results

• Successfully classified chat messages into 12 distinct categories
• Reduced model inference time by 40%
• Improved accuracy from baseline 78% to 94%

Technologies Used

• PyTorch
• Transformers library
• BERT/RoBERTa models
• Custom training loops

---

IE University - MSc in Computer Science & Business Analytics