Learn with Neo4j's New "Get to Know Graph & GenAI" Webinar Series >>

Neo4j logo

Making Graphs More User-Friendly with AI

Session Track: Knowledge Graphs

Session Time:

Session description

In his Nodes2023 session, Guido presented an elaborate NeoDash visualization of a knowledge graph intended to facilitate historical research. Focusing on the printing industry in the Belgian city of Antwerp through the ages, the dataset now consists of 200K+ nodes and 260K+ relationships with data on 5,500+ persons—all of which were entered manually. End-users can easily find information on the NeoDash dashboard but only represented in the way the designer wanted. Tailor-made queries are needed for all other research questions—not very end-user-friendly. ChatGPT to the rescue! Inputwise, they have trained a ChatGPT persona to extract relevant data from OCR-readable PDF-files, compare them with the information in the graph, and present the outcome for manual arbitrage. On average, this reduces data entry time by 80%. Outputwise, they have created a ChatGPT-based UI that enables end-users to obtain natural language answers to any natural language question about the full content of the dataset. This was done by utilizing OpenAI’s API and Neo4j’s API to translate prompts to Cypher queries, run them, and translate the results back to natural language. When a user submits a question, a request is sent to OpenAI’s API using the o3-mini model. The model responds with a Cypher query that represents the user's intent. After running the Cypher query through the Neo4J API, Neo4j responds with the raw query results in JSON—rows of data and column names. The question, query, and results are then fed back into the model, which turns the data into a concise, natural language answer. This whole endeavor is a WordPress plugin that allows for integration anywhere on the website and isn't bound by the constraints of an iframe. By joining this session, you will learn how relatively basic AI-applications can turn a graph database into an even more powerful research tool in the hands of the end-users.

Speaker

photo of Guido Thys

Guido Thys

Vice President, Antwerp Bibliophile Society

Guido Thys, vice president of the Antwerp Bibliophile Society, studied general, psycho- and neurolinguistics and the philosophy of language in Antwerp (B), Leuven (B), Brussels (B) and Salzburg (AT). As a lecturer in strategy, marketing and database management at a business school, he coauthored 20 books and dozens of articles on these subjects and has presented more than 2,600 times to business audiences. Guido wrote the dBase-II code for his first database in 1985 and has been at the forefront of database use ever since. He is an avid collector of books printed in Antwerp.