STREAM: A Data-Centric Framework for Mining High-Value Task-Oriented Dialogues from Streaming Media

2026-05-24

AI Summary

This work addresses the scarcity of high-quality, complex, and domain-specific data that hinders task-oriented dialogue systems in vertical domains, where existing approaches suffer from high annotation costs, privacy and commercial constraints on real conversations, and poor temporal relevance of static corpora. To overcome these challenges, we propose STREAM, a novel framework that leverages publicly available streaming media—live streams and short videos—as a source of dialogue data. STREAM synthesizes structured multi-turn task-oriented dialogues through persona construction anchored to speaker roles and dialogue blueprint generation, and integrates retrieval-augmented generation to produce knowledge-aware responses. We release StreamDial, a dataset spanning automotive, food, and hotel domains, comprising 87,498 conversations (over 1.49 million turns). Experiments demonstrate that our approach significantly improves intrinsic dialogue quality and dialogue state tracking performance, while also enabling effective multilingual transfer.

Abstract

Large language models for vertical domains are bottlenecked by the scarcity of complex, domain-specific task-oriented dialogues. Existing data acquisition pipelines face a persistent trilemma: expert annotation is expensive, real-world service conversations are constrained by privacy and commercial restrictions, and static corpora quickly become temporally stale. We propose Stream, a data-centric framework that leverages publicly available streaming media (live streams and short videos) to synthesize high-value service dialogues at scale. Stream mines authentic interaction signals from noisy streams and synthesizes conversations by integrating role-grounded persona construction with Conversational Blueprint construction; it further adopts retrieval-augmented generation (RAG) to support knowledge-aware responses. Based on Stream, we release StreamDial, a large-scale multi-domain dataset covering Automotive, Restaurant, and Hotel. StreamDial contains 87,498 dialogue sessions and 1,497,320 turns in total, with an average of 17.11 turns per session and a comparable scale across domains. Each session is organized as a structured quadruplet $\langle P_u, P_a, B, H \rangle$ that pairs dialogue history with explicit user/agent personas and a Conversational Blueprint, capturing realistic service behaviors such as requirement mining, constraint conflicts, negotiation, and recovery. Evaluations with automatic judges and downstream tasks show that StreamDial improves intrinsic dialogue quality over strong baselines, and models trained with StreamDial improve Dialogue State Tracking across backbones; we further report a completed human-evaluation set and encouraging multilingual transfer on Qwen3-8B under a controlled training budget. The data is released in https://github.com/hitxueliang/DialogDataSetBySTREAM.

Problem

• task-oriented dialogue
• data scarcity
• domain-specific
• streaming media
• dialogue dataset

Innovation

• streaming media mining
• conversational blueprint
• role-grounded persona
• retrieval-augmented generation
• task-oriented dialogue synthesis