EmpathAI-DPO-Vi: Vietnamese DPO and SFT Data for Empathetic Customer Service Chatbots

A Vietnamese dataset for Direct Preference Optimization and Supervised Fine-Tuning designed to train empathetic customer service chatbots. The data was generated using a four-stage pipeline with the Gemini 2.5 Flash model on Vertex AI. It was created by author 'thanhhoangnvbg' and last updated on April 26, 2026.

Use Cases

• Fine-tuning language models to politely refuse unreasonable customer requests based on described policy adherence.
• Training chatbots to handle toxic customer interactions without conceding on regulations as mentioned in the description.
• Developing models that ask clarifying questions to better understand customer needs, a key focus of the dataset.

Strengths

• Data generation followed a defined four-stage pipeline using a specific model (Gemini 2.5 Flash) on a named platform (Vertex AI).
• Dataset is specifically designed for a complex, high-value task: training empathetic and policy-aware customer service chatbots.
• Focuses on a less common language (Vietnamese) for preference optimization datasets in the described domain.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.
• Description metadata is limited; the full description is hosted externally, requiring manual inspection for complete details.

Provenance

Source

huggingface

Collection Method

Generated via a four-stage pipeline using the Gemini 2.5 Flash model on Google's Vertex AI platform.

Freshness

Last updated 2026-04-26 18:40:22; freshness should be verified.