Conversation-based interfaces to relational databases (C-BIRDS)

The development of reliable natural language interfaces to relational databases can accelerate the progress of many areas of applications, such as the need for interactive relational database interfaces for non-technical users. This thesis presents a novel development of C-BIRDs framework. The C-BIRD framework was developed in two phases. The first phase involved the development of a Static Approach-Based C-BIRD framework, which is based on a combination of a Goal-Oriented Conversational Agent (GOCA) and a knowledge tree (KT). GO CAs have proven their capability to disambiguate the user's needs through natural language conversations. KT is used to overcome the lack of connectivity between the GOCA and the relational database, through organizing the domain knowledge in a knowledge tree. In addition the Static Approach-Based C-BIRD framework, a number of strategies were employed based on scripting structures in order to enhance the reasoning capabilities towards answering user queries. The second phase involved the development of a Dynamic Approach- Based C-BIRD framework which is based on information extraction (lE) in order to dynamically create an SQL statement that answers user queries. lE component utilised a number of SQL query templates, which are made of relational database semantically understandable patterns such as table and column names. In addition, the dynamic approach used the conversational agent to disambiguate the dynamically generated SQL queries by confirming these queries with the user by means of SQL template specific strategies scripts. The Static Approach-Based C-BIRD prototype showed excellent results in terms of successfully mapping natural language conversations into SQL statements (i.e. task success, in which 5 tasks performed by 20 participants with an overall result of 91 %). The dynamic approach also showed very good results in terms of task success; 5 tasks performed by 20 participants with an overall result came to 74%. In summary, the proposed static approach C-BIRD framework offered a novel methodology to develop reliable conversational interfaces to relational databases in which engineered queries can be answered. In addition, the dynamic approach introduced a novel way to map natural language utterances into SQL statements and confirming the results with the user, before providing the final answer. Ultimately, the user experiences a real-time and friendly conversational interface with the relational.