Persuasive Dialogue Corpus: Graph-Based Approach Combining Persuader and Persuadee Perspectives

Persuasion is omnipresent in our daily communication. As a mechanism for changing or forming one’s opinion or behavior, persuasive dialogues and their strategies have gained interest for developing intelligent conversational systems. Given the complexity of this task, persuasion systems, especially dealing in conversations that require ‘no action’ by the user but rather a change in opinion or belief, require specialized annotated corpora and the understanding of logical structure, natural language, and persuasive strategies. The sparsity of available annotated data and a wide range of proposed models make it challenging for developing strategic chatbots specific to user needs. To address these issues, this study introduces a novel framework combining a replicable data collection tool and a topic-independent annotation schema for designing an argument-graph corpus and incorporating both persuader and persuadee perspectives, essential for building smart conversational agents.