Commonsense-augmented Memory Construction and Management in Long-term Conversations via Context-aware Persona Refinement
Memorizing and utilizing speakers' personas is a common practice for response generation in long-term conversations. Yet, human-authored datasets often provide uninformative persona sentences that hinder response quality. This paper presents a novel framework that leverages commonsense-based persona expansion to address such issues in long-term conversation.
While prior work focuses on not producing personas that contradict others, we focus on transforming contradictory personas into sentences that contain rich speaker information, by refining them based on their contextual backgrounds with designed strategies. As the pioneer of persona expansion in multi-session settings, our framework facilitates better response generation via human-like persona refinement.
Click here to see our paper.
Supplementary video of our framework CAFFEINE ☕️
Visualization of Iterative Graph Refinement
Example of Response Generation #1.
While baselines yield short personas and inconsistent or unconstructive responses (green underlines), CAFFEINE offers informative personas (color red) and leads to a response that provides constructive suggestion for Person A’s Spanish learning.
Example of Response Generation #2.
While baselines yield short personas, CAFFEINE offers informative personas (color red) and leads to a response that reflects Person B’s situation and what B is looking for in a car.
Example of Resolution & Disambiguation #1.
Example of persona resolution and disambiguation done by CAFFEINE. Color magenta highlights the contextual cues from the relevant dialogue contexts.
Example of Resolution & Disambiguation #2.
Example of persona resolution and disambiguation done by CAFFEINE. Color magenta highlights the contextual cues from the relevant dialogue contexts.
Example of Preservation #1.
Examples of "preservation" addressing the sub-optimal performance of NLI models that solely rely on the persona sentences without contextual backgrounds. Color magenta highlights the contextual cues.
Example of Preservation #2.
Examples of "preservation" addressing the sub-optimal performance of NLI models that solely rely on the persona sentences without contextual backgrounds. Color magenta highlights the contextual cues.