-
Notifications
You must be signed in to change notification settings - Fork 0
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Analysis task - genereate safety themes from safety issuess #144
Comments
After a dicussion aroudn the robustness and unsuccessfulness of using the llm a more transpraent method will be investigiated. This will involve using clustering of the embeded safety issues. From there we could see what the clusters are. |
I have developed some potential models that do a reasonable job of clustering the safety issues. There are individual models for each mode, merged models and topic models fit to all of the safety issues. Becuase I am working with such a low amount of documents I expect it to be much more sophisticated and powerful when more safety issues are added to the dataset from other organisations. |
Safety themes models seemed to be good enough. I need to try and create some simple 2d graph that can show all of the safety issues and hteh associated safety themes. |
I am preforming an exhatstive hyper parameter search to see what the best topic model will be. Then I will find the one that has the lowest number of outliers as well as the most resaonable looking safety themes. |
Meeting with TAIC has come up with these points:
|
The feedback on these demo models have some actionable items.
|
After meeting with TAIC this goal has been put on hold so that focus can be put on the searching problem #146. |
Currently the model has generated general themes based off of the "ThemeSummary" generated for each report.
These theme summaries do not conform to the TAIC investigation model. Therefore it would be good to get some overall safety themes after reading all of the safety issues.
This will be completed as a oneoff analysis task. The answer g iven and it could be intergrated into the engine.
The text was updated successfully, but these errors were encountered: