Enabling Natural Language Graph Queries in GraphScope with chatGPT

[acezen]

Is your feature request related to a problem? Please describe. Graph queries are a critical workload in GraphScope, currently submitted through query languages including Gremlin and Cypher. Although these graph query languages greatly reduce the user barrier compared to programming languages like Java or C++, GraphScope users still frequently struggle with query writing. This not only severely limits user graph querying capabilities, but also consumes valuable developer time in answering user questions. With the emergence of chatGPT, people have realized the incredible ability of large language models to handle natural language-related tasks including writing articles, answering questions, and even writing code. We wonder if we can leverage the ability of large language models to further reduce the barrier to using graph queries in GraphScope.

In this task, you will explore this question with us. Specifically, we expect that given the context of graph queries, such as query samples, graph schema information, and reference query implementations (based on Cypher), can we allow users to automatically generate Cypher queries expressed in natural language?

Describe the solution you'd like

• Explore the use of chatGPT to generate additional test samples (including manual labeling) based on existing query samples.
• Use the above test samples to explore the improvement of Cypher query generation accuracy through the prompt method.
• Add natural language querying capabilities to the existing Python interface in GraphScope.

[AugensternF]

wanna join🙋

[Suchun-sv]

wanna join🙋

[longbinlai]

Hi, please sign in as candidate in: https://m.summer-ospp.ac.cn/org/prodetail/2313e0395

[Undertone0809]

I wonder what's the next step for this feature.