Search papers, labs, and topics across Lattice.
The paper introduces Semantic XPath, a tree-structured memory module designed for conversational AI agents to efficiently access and update structured conversational memory. This approach addresses the limitations of in-context memory (scalability) and flat-RAG methods (lack of structure) in long-term, task-oriented interactions. Experiments show a 176.7% performance improvement over flat-RAG baselines while using significantly fewer tokens than in-context memory, demonstrating the effectiveness of structured memory access.
Ditch the flat memory: Semantic XPath unlocks a 176% performance boost in conversational AI by organizing memories into a tree structure and querying them with semantic XPath expressions.
Conversational AI (ConvAI) agents increasingly maintain structured memory to support long-term, task-oriented interactions. In-context memory approaches append the growing history to the model input, which scales poorly under context-window limits. RAG-based methods retrieve request-relevant information, but most assume flat memory collections and ignore structure. We propose Semantic XPath, a tree-structured memory module to access and update structured conversational memory. Semantic XPath improves performance over flat-RAG baselines by 176.7% while using only 9.1% of the tokens required by in-context memory. We also introduce SemanticXPath Chat, an end-to-end ConvAI demo system that visualizes the structured memory and query execution details. Overall, this paper demonstrates a candidate for the next generation of long-term, task-oriented ConvAI systems built on structured memory.
