At Agoda, we handle approximately 50,000 emails from suppliers and customers daily, and this number is continuously increasing. Efficiently classifying response emails from suppliers, hotels, or…

Agoda Engineering Blog offers insights, technical articles, and updates on building scalable and high-performance travel technology platforms. Covering topics such as backend development, distributed systems, and machine learning applications, Agoda Engineering Blog provides resources for software engineers and data scientists working in the travel and hospitality industry, helping them deliver exceptional user experiences and drive business growth through technology innovation.

agoda

This post discusses how cosine similarity is used for fine-tuning dataset estimation in GPT models for email classification. It explains the challenges in dataset preparation, the concept of fine-tuning, embeddings, and cosine similarity. The solution developed involves utilizing cosine similarity to calculate the minimum dataset size required for fine-tuning. An experiment is conducted to classify responses to cancellation fee waiver requests, and the results show a significant reduction in dataset requirements without sacrificing accuracy.

How We Leverage Cosine Similarity for Fine-Tuning Dataset Estimation

Key Challenges in Dataset Preparation for GPT Models