the data processing format and the metrics of event detection

[Log4J-Ops]

I used your model but cannot reproduce the same performance. Can you provide the processing data format? The calculation process of the metrics uses the command "from sklearn.metrics import f1_score, precision_score," tha library recognizes events one token for one trigger. However, the definition of event detection is to identify and class the span of a trigger when the trigger consists of multi-tokens . Is it reasonable to use this library in this case?

[Bakser]

Regarding the data format, this repo adopts the same data format as in HMEAE, which shares the preprocessing script and format example. Regarding the evaluation metric, please kindly note that the metrics are calculated over sequences of trigger candidates rather than tokens. We recommend trying our recent OmniEvent toolkit, which implements standardized data preprocessing and evaluation metrics. You can also easily customize your own models within its framework. CLEVE is also included in OmniEvent.