COMP5425
Week11 Semester 1, 2025
Conversational Retrieval
Introduction
Components
Benchmark and evaluation
Advance Topics
Applications
Traditional vs Conversational
Traditional
rely on keywords or short phrases
Conversational
natural language dialogue interaction
enhancing the efficiency of information exchange and optimizing the user experience
Complex queries
Longer context
Evolution of Search System
Key word → NLP Based → Conversational
Examples
1966: ELIZA - One of the earliest examples of a natural language processing system.
2010s: Siri & Google Now - Introduction of voice-activated virtual assistants.
2018: BERT - A significant advancement in natural language understanding.
2020: TREC CAsT & WISE Datasets - Development of conversational search datasets.
2022: ChatGPT - A breakthrough in conversational AI
Key Characteristics
Multi-turn
Context
Multi-turn context: Handling multiple turns in a conversation to maintain context.
Context understanding and management: Managing follow-up questions and ambiguous queries.
Knowledge integration and reasoning: Integrating knowledge to answer complex queries.
Challenges
Context Understanding and Management
follow-up questions: “Could you give more details of it?” after a previous query.
Ambiguity and Under-Specified Queries
“Show me the best”
Mixed Task and Chat Intent
“Hi, how are you? Can you check my order?”
Knowledge Integration and Reasoning
“What is the most recent three PMs of Australia?”
Trends
LLM integration
RAG
Multimodal Conversational Search
Domain Specific
Personalized
Etc…