Graph Databases for public institutions
Making public data available is what builds up modern society. Make it truly happen and see how others can benefit from shared data. Just imagine: one, shared database that is both accessible and easy to use. The perfect tool for enhanced problem-solving and better decision-making? We think so. Such a solution already works for the Swiss government. It’s the graph database. It perfectly maps the relationships in the data structure. It offers better performance with really complex data structures. It’s quite intuitive as well as adaptable. Adding new data and linking it with existing one is incredibly easy.
This solution works well for those managing the data in public institutions. They add the new data by linking it with the already existing one in a graph structure. But there’s one catch that triggered our partner, Zazuko. Although the graph database is implemented, it’s rarely used. Using it requires effort put into understanding links between data by employees not specializing in data or in programming. In their busy schedules, they don’t find time for that.
Talk to your data
From the very beginning, we knew that Natural Language Processing could solve this problem. What can be better than retrieving the information by simply talking to the database? However, the exact idea of how to tackle the issue came out in October 2023. Just when the Large Language Models (LLMs) reached its next level.
Our work started with a simple experiment. Can we make ChatGPT define the data flow? As you could expect, the first attempts were not quite successful. To put it simply, ChatGPT got easily lost in the data, or presented incomplete information. But with the right approach, everything changed. By making ChatGPT aware of data context, data structure and metadata, our chatbot developed a completely new capability.
ChatGPT for Graph Database querying
It has one particular goal in mind. Supporting the user in navigating the graph database. You ask questions. Our model generates a query to get you the right answer. It works that way: the user describes in plain language what wants to get. Now, using the Retrieval Augmented Generation (RAG), the chatbot gets the information about the graph structure. Knowing the context, it generates the query. Now, we can use it to get the information from the real database.
The purpose? Easier and faster access to the information stored in the graph databases. The additional value? It is its educational aspect. Even by simply copying and pasting the query, the user interacts with the code… and learns.
Waiting for the future
That journey is definitely not over yet. At Datali, we are preparing to conduct wider tests of the solution. The aim is simple: gather feedback from the users, the Swiss administration employees, and improve the chatbot as much as possible.
The future is not only about easily getting data via a chatbot. It also lies in state-of-the-art solutions in database management, like graph databases. While the Swiss government sets a strong precedent, the potential for ChatGPT-based database communication extends far beyond public institutions. Medium and large enterprises, research organizations, and even educational institutions face similar challenges with data accessibility and usability.
Are you ready to talk to your data? Because we’re here to help you start the conversation.
