Transcript Document
Guided Conversational Agents and Knowledge Trees for Natural Language Interfaces to Relational Databases Mr. Majdi Owda, Dr. Zuhair Bandar, Dr. Keeley Crockett The Intelligent Systems Group, Department of Computing and Mathematics, Manchester Metropolitan University. Background to Research • Databases – Hierarchal Databases – Relational Databases * – Object Oriented Databases • Artificial Intelligence – Knowledge Representation • Knowledge Trees * – Expert Systems – Natural Language Processing • Conversational Agents * – Machine Learning • Human-Computer Interaction – Natural Language Interfaces * • Introduction – Natural Language Interfaces to Databases – Guided Conversational Agents – Knowledge Trees • • • • Proposed Framework Developed Prototype Conclusions and Future Work Q/A Contents • Introduction – Natural Language Interfaces to Databases – Guided Conversational Agents – Knowledge Trees • • • • Proposed Framework Developed Prototype Conclusions and Future Work Q/A Natural Language Interfaces to Databases • Where the Complexity comes from !! • Past Approaches – – – – Pattern-Matching Intermediate Language Syntax-Based Family Semantic-Grammar The Problem: Creating Reliable Natural Language Interfaces to Relational Databases. Contents • Introduction – Natural Language Interfaces to Databases – Guided Conversational Agents – Knowledge Trees • • • • Proposed Framework Developed Prototype Conclusions and Future Work Q/A Guided Conversation Agents • • • • • Alan Turing (Turing Test) 1950 Joseph Weizenbaum (Eliza) 1960s Colboy (Parry) late 1960s Wallace (Alice) 2000 MMU (InfoChat-Adam) 2001 Idea: use a guided conversational agent for NLIDBs. Algorithm: having a guided conversational agent component trained to converse within a database domain knowledge. Guided Conversation Agents – Why InfoChat • • • • • • Autonomous general purpose CA Deals set of contexts Direct the users towards a goal Flexible and robust Converse freely within a specific domain Extract, manipulate, and store information Contents • Introduction – Natural Language Interfaces to Databases – Guided Conversational Agents – Knowledge Trees • • • • Proposed Framework Developed Prototype Conclusions and Future Work Q/A Knowledge Trees Direction Node Goal Node Idea: using knowledge trees for NLIDBs. Algorithm: having knowledge trees component within the new framework. Knowledge Trees Benefits • Easy way to revise and maintain the knowledge base • Overcome the lacking of connectivity between CA and the Relational Database • Road map for the conversational agent dialogue flow • Direct the conversational agent towards the goal. Contents • Introduction – Natural Language Interfaces to Databases – Guided Conversational Agents – Knowledge Trees • • • • Proposed Framework Developed Prototype Conclusions and Future Work Q/A Conversation-Based NLI-RDB Framework User Query Agent Response • Main components Conversation Manager Response Generation Knowledge Tree Context Switching & Manage Conversational Agent SQL statements Rule Matching Context Script files Information Extraction Relational Database – – – – Conversational Agents Knowledge Trees Conversation Manager Relational Database Contents • Introduction – Natural Language Interfaces to Databases – Guided Conversational Agents – Knowledge Trees • • • • Proposed Framework Developed Prototype Conclusions and Future Work Q/A Conversation-Based NLI-RDB Prototype Tools Conversation-Based NLI-RDB Interface Conversation-Based NLI-RDB Interface Contents • Introduction – Natural Language Interfaces to Databases – Guided Conversational Agents – Knowledge Trees • • • • Proposed Framework Developed Prototype Conclusions and Future Work Q/A Conclusions • Easy and flexible way in order to develop a Conversation-Based NLI-RDB • General purpose framework which can be applied to a wide range of domains • Utilizing dialogue interaction • Knowledge trees are easy to create, structure, update, revise, and maintain • Capability of handling simple and complex queries Current & Future Work • An adaptive conversation-based NLIDB • Dynamic knowledge trees Idea: There is still big room to do further research. Special thanks “MMU Research Team” Dr. Keeley Crockett Mr James O’Shea Dr. Zuhair Bandar Dr. David Mclean Questions [email protected]