Reproduce “SHARE: Shared Memory-Aware Open-Domain Long-Term Dialogue Dataset Constructed from Movie Script” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Eunwon Kim, Chanho Park, Buru Chang
Sogang and Korea Universities researchers constructed SHARE, a dataset for open-domain long-term dialogue derived from movie scripts that explicitly includes shared memories, alongside the EPISODE framework to manage and utilize these memories. The system produced dialogues with improved reflectiveness, consistency, and engagingness by incorporating relational history compared to existing baselines.
Key insights
- Movie scripts offer an authentic and rich source for constructing long-term dialogue datasets, naturally capturing complex relational dynamics and implicit shared memories between characters.
- Explicitly incorporating shared memories into dialogue systems is vital for generating conversations that are more reflective of human relationships, consistent across sessions, and engaging.
- An effective memory management strategy, including accumulation, logical sequential/causal/conflicting updates, and deduplication, is essential for maintaining a coherent and dynamic memory set in extended dialogues.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.