- Title: Re-defining AI Knowledge Bases inside Chatbots
- Supervisor: Dr. Yu Tao, Dr. Lingpeng Kong
-
Students:
- Wu Kunhuan (u3577163@connect.hku.hk)
- Chan Tsz Chun (u3577972@connect.hku.hk)
Introduction
Chatbots are popular nowadays due to their diverse capabilities in different areas such as textual applications, education, marketing, customer service, etc.
However, it is difficult to trace the specific knowledge applied in the logical process, and the knowledge itself is also hard to visualize under the response of popular chatbots including ChatGPT.
In this project, our group aim to discover whether the recent application of knowledge graph-based approaches can provide a more comprehensive solution to AI-based knowledge construction and reasoning.
Features
The name our group defined for the final deliverable is RTSAI, which represents Rational, Teachable and Scalable Artificial Intelligence. We deployed GitHub to manage the codes of RTSAI, and the directory is available via https://github.com/WuKunhuan/RTSAI.
- Rational Reasoning
Our group decided to use text-to-sql-related work to convert users’ natural language input into logical queries. We pay special attention to SPARQL because it has SQL-like functions and can support knowledge graph construction packages such as kgtk to retrieve knowledge graph-like graph structures.
Through fine-tuning the query format, the corresponding knowledge graph is designed to provide the information required for query retrieval. The response obtained by the user will be verified through the knowledge graph.