Very lightweight annotation visualization for the EDL branch of TAC 2015/2016 with a browser UI. This project was only intended as a quick and small helper for error analysis and is not a full blown entity linking visualization tool.
Select from multiple system outputs to compare system performance.
Shows annotation errors for:
- strong mention match
- strong typed mention match
- strong linked mention match
- select/deselect strong typed mention match and strong linked mention match (i.e. CLASS ERROR and LINK ERROR)
Start by completing the configuration in elvis/config.py (follow the explanations in the config).
Start the server with
$ ./main.py
Which will produce the message (if you didn't change the host and port in config)
* Running on http://127.0.0.1:33507/ (Press CTRL+C to quit)
Point your browser to http://127.0.0.1:33507/ to use elvis.
Only tested on Linux. Requires Python3, pandas and flask.
To resolve the KBIDs to something human readable implement the function get_entity_name_from_id(entity_id) in utils.py.
Known issues that I probably won't fix:
- Does not handle overlapping but not embedding mention spans correctly.
- Clicking on mentions with embedded mentions leads to multiple conflicting popups.