Forecasting Communication Derailments Through Conversation Generation

2025-04-11

AI Summary

This study addresses the problem of predicting future communication derailments—such as conflict or misunderstanding—in dialogues, supporting applications including content moderation, conflict intervention, and business negotiation. We propose a generative trajectory consensus prediction paradigm: a fine-tuned large language model conditionally generates multiple future dialogue trajectories guided by sociolinguistic attributes (e.g., power asymmetry, sentiment polarity), and derailment risk is determined via inter-trajectory consistency. Our approach is the first to explicitly incorporate sociolinguistic features into a generative prediction framework, substantially enhancing dynamic interaction modeling. Evaluated on an English communication derailment prediction benchmark, it outperforms state-of-the-art methods. Ablation studies confirm the contribution of each component, demonstrating significant overall accuracy improvement.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Forecasting communication derailment can be useful in real-world settings such as online content moderation, conflict resolution, and business negotiations. However, despite language models' success at identifying offensive speech present in conversations, they struggle to forecast future communication derailments. In contrast to prior work that predicts conversation outcomes solely based on the past conversation history, our approach samples multiple future conversation trajectories conditioned on existing conversation history using a fine-tuned LLM. It predicts the communication outcome based on the consensus of these trajectories. We also experimented with leveraging socio-linguistic attributes, which reflect turn-level conversation dynamics, as guidance when generating future conversations. Our method of future conversation trajectories surpasses state-of-the-art results on English communication derailment prediction benchmarks and demonstrates significant accuracy gains in ablation studies.

Problem

• Forecasting future communication derailments in conversations
• Improving prediction accuracy using generated future trajectories
• Incorporating socio-linguistic attributes for better derailment forecasting

Innovation

• Fine-tuned LLM generates multiple future conversation trajectories
• Consensus of trajectories predicts communication derailment outcomes
• Socio-linguistic attributes guide future conversation generation