Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Knowledge-Aware Conversation Derailment Forecasting Using Graph Convolutional Networks

Enas Altarawneh, Ameeta Agrawal, Michael Jenkin, Manos Papagelis

TL;DR

This work derives commonsense statements from a knowledge base of dialogue contextual information to enrich a graph neural network classification architecture and fuse the multi-source information on utterance into capsules, which are used by a transformer-based forecaster to predict conversation derailment.

Abstract

Online conversations are particularly susceptible to derailment, which can manifest itself in the form of toxic communication patterns including disrespectful comments and abuse. Forecasting conversation derailment predicts signs of derailment in advance enabling proactive moderation of conversations. State-of-the-art approaches to conversation derailment forecasting sequentially encode conversations and use graph neural networks to model dialogue user dynamics. However, existing graph models are not able to capture complex conversational characteristics such as context propagation and emotional shifts. The use of common sense knowledge enables a model to capture such characteristics, thus improving performance. Following this approach, here we derive commonsense statements from a knowledge base of dialogue contextual information to enrich a graph neural network classification architecture. We fuse the multi-source information on utterance into capsules, which are used by a transformer-based forecaster to predict conversation derailment. Our model captures conversation dynamics and context propagation, outperforming the state-of-the-art models on the CGA and CMV benchmark datasets

Knowledge-Aware Conversation Derailment Forecasting Using Graph Convolutional Networks

TL;DR

This work derives commonsense statements from a knowledge base of dialogue contextual information to enrich a graph neural network classification architecture and fuse the multi-source information on utterance into capsules, which are used by a transformer-based forecaster to predict conversation derailment.

Abstract

Online conversations are particularly susceptible to derailment, which can manifest itself in the form of toxic communication patterns including disrespectful comments and abuse. Forecasting conversation derailment predicts signs of derailment in advance enabling proactive moderation of conversations. State-of-the-art approaches to conversation derailment forecasting sequentially encode conversations and use graph neural networks to model dialogue user dynamics. However, existing graph models are not able to capture complex conversational characteristics such as context propagation and emotional shifts. The use of common sense knowledge enables a model to capture such characteristics, thus improving performance. Following this approach, here we derive commonsense statements from a knowledge base of dialogue contextual information to enrich a graph neural network classification architecture. We fuse the multi-source information on utterance into capsules, which are used by a transformer-based forecaster to predict conversation derailment. Our model captures conversation dynamics and context propagation, outperforming the state-of-the-art models on the CGA and CMV benchmark datasets
Paper Structure (18 sections, 3 equations, 2 figures, 5 tables)

This paper contains 18 sections, 3 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Definition
  4. The Knowledge Aware Conversation Derailment Forecasting Model
  5. Sequential encoding
  6. Graph Construction
  7. Feature Transformation
  8. The Derailment Forecaster
  9. Experimental Setup
  10. Datasets
  11. Evaluation Metrics
  12. Implementation
  13. Baselines
  14. Results and Discussion
  15. Forecasting Conversation Derailment
  16. ...and 3 more sections

Figures (2)

  • Figure 1: A sample conversation from the Reedit change my view (CMV) dataset showing a sequence of text utterances that end with a verbal abuse. Given the conversation context in the previous $N-1$ turns, the task is to predict whether turn $N$ will be a respectful or offensive statement prior to it being presented leading to derailment. The available data for utterances in the CMV dataset contain text, user ID and a public perception negative or positive score through public votes.
  • Figure 2: An overview of the Knowledge Aware Graph convolutional Network forecasting model KA-GCN