CoMAC: Conversational Agent for Multi-Source Auxiliary Context with Sparse and Symmetric Latent Interactions
Junfeng Liu, Christopher T. Symons, Ranga Raju Vatsavai
TL;DR
CoMAC addresses the challenge of generating grounded, personalized responses by simultaneously leveraging multiple auxiliary data sources (persona and knowledge). It introduces a three-stage pipeline: source-specific input embedding, a post-fusion grounding stage using a novel sparse and symmetric ColBERT-inspired similarity $ abla extsubscript{SSN}$, and a generation stage that conditions the LM on selected persona and knowledge entries hatP and hatK. The method demonstrates significant improvements over state-of-the-art single- and multi-source baselines on the FoCus dataset, including substantial gains in PG and KG grounding accuracies and standard generation metrics (F1, ROUGE-L, BLEU) as well as perplexity reductions. Key insights include the necessity of bi-directional information flow, the value of token-level sparsity to reduce noise, and the complementary roles of TF-IDF sampling and learned token weighting in grounding performance. While promising, the approach acknowledges remaining challenges such as hallucinations and the computational cost of online TF-IDF weighting, with future work aimed at addressing these issues and extending evaluation beyond FoCus.
Abstract
Recent advancements in AI-driven conversational agents have exhibited immense potential of AI applications. Effective response generation is crucial to the success of these agents. While extensive research has focused on leveraging multiple auxiliary data sources (e.g., knowledge bases and personas) to enhance response generation, existing methods often struggle to efficiently extract relevant information from these sources. There are still clear limitations in the ability to combine versatile conversational capabilities with adherence to known facts and adaptation to large variations in user preferences and belief systems, which continues to hinder the wide adoption of conversational AI tools. This paper introduces a novel method, Conversational Agent for Multi-Source Auxiliary Context with Sparse and Symmetric Latent Interactions (CoMAC), for conversation generation, which employs specialized encoding streams and post-fusion grounding networks for multiple data sources to identify relevant persona and knowledge information for the conversation. CoMAC also leverages a novel text similarity metric that allows bi-directional information sharing among multiple sources and focuses on a selective subset of meaningful words. Our experiments show that CoMAC improves the relevant persona and knowledge prediction accuracies and response generation quality significantly over two state-of-the-art methods.