# How the OpenSource Graph AI platform helps / オープンソースのグラフAIプラットフォームがどのように役立つでしょう？ (Wey Gu)

[slido](https://app.sli.do/event/pNBdBi95c8ujLz9u9fdVeJ/live/questions)

### How much do we have cost to study graph basics? (Cost means not only money.)

wey: Great question!
To me, I started from NebulaGraph(https://docs.nebula-graph.io/) and you could start from GNN as well, i.e. Stanford CS224W https://www.youtube.com/watch?v=JAB_plj2rbA, too, my expirence on this was not to dive into long course, but to start from creating something with it(learning by doing, i.e. redo following the SNS with Graph post https://www.siwei.io/en/nebulagraph-sns/ ), and find the answers you come across and learn with purposes(with the help of google,chatgpt,community,docs).

### How do you collect all your data and manage your database? (about the movie, graphs)

wey: Great question, I collected from two sources in this project: OMDB and MovieLens, with some ETL, I shared how I created the graph with dbt(opensource data transform project) on another blog post:
- https://www.siwei.io/en/nebulagraph-etl-dbt/
- https://www.siwei.io/recommendation-system-with-graphdb/

### Will you show the case that we must not use Graph?

wey: Great question! Intuitively, when the problem is not "relation-crutial", we don't need bother introduce the complecity of graph. But another thing I observed on Graph Open Source community is to over-use Graph Database for Tabular queries, i.e. if our query pattern is more `select * from table_foo` rather than the graph hop travesal, we should not persist data only in Graph Database, instead, we could put single source of truth in a tabular db, and pipline the data into graph database(like nebulagraph of course) :).

### Your graph makes me think of molecular structural images and their relationships. Do you think it is possible for graph AI the solve molecular-level problems?

wey: Excellent idea and exactly, molecular structural and their properties are actually typical Graph probelms, while due to my lack of this domain knowledge, I havent try create opensource toy projects to help adopt graph to solve those problems, but I had seen a bunch of blog post/videos mentioning how they address so, with Graph Queris or GNN! You could explore that :)

### Do you think if Graph ML has bias problem like humanrights violation, It's depends on data or model design?

wey: Wow, that's a deep question 👍🏻, sure it has this issue in Graph ML. There could be challaneges around bias when Graph was introduced during the modeling and data collection, like, if it's appropraite to collect device info for help address fraud issues, we should carefully review what we could leverage, and especially for what could be hidden in the graph relationship(comparing to non-graph data).