A Multiparty Chat-Based Dialogue System with Concurrent Conversation Tracking and Memory

Most of the recent developments in conversational agents only consider interactions with one user at a time. To interact with multiple users at the same time, extensions to the two-party dialogue system framework have been explored. However, this approach either assumes that multiparty interactions can be approximated by a sequence of two-party independent conversations, or compromises the agent's capacity for individual engagement (i.e., recalling user preferences and details from previous interactions). In this work, these limitations are addressed through two novel contributions. First, we develop a module that keeps a concurrent record of conversations, where each conversation is explicitly marked as a group or individual discussion. Second, we extend the intent framework to include agent memory across multiple timescales. This allows conversational agents to reason about past interactions by recalling previously recorded intents. The practical application of the system is verified using a case study on a real-world chat-based conversational agent implementation.