3 ms·
Duckling, our open-sourced probabilistic parser to detect entities like dates and times, numbers, and durations Are there any benchmark for how well this compa
by nl 10y ago
Duckling, our open-sourced probabilistic parser to detect entities like dates and times, numbers, and durations
Are there any benchmark for how well this compares with something like HeidelTime or SUTime[2]?
[1] https://github.com/HeidelTime https://github.com/HeidelTime
[2] https://nlp.stanford.edu/software/sutime.html https://nlp.stanford.edu/software/sutime.html
- blandinw 10y ago(disclaimer: I work on Duckling) I don't know of any benchmarks. HeidelTime and SUTime are two solid projects... although the rule files [0][1] are a bit scary if you ask me :-). Quick thoughts: - SUTime relies on TokensRegex [2] which is similar to how Duckling parses sentences at a high-level - SUTime seems to only provide English rules - I don't know of any production use cases of either [0] https://github.com/stanfordnlp/CoreNLP/tree/master/src/edu/stanford/nlp/time https://github.com/stanfordnlp/CoreNLP/tree/master/src/edu/s... [1] https://github.com/HeidelTime/heideltime/blob/master/resources/english/rules/resources_rules_daterules.txt https://github.com/HeidelTime/heideltime/blob/master/resourc... [2] https://nlp.stanford.edu/pubs/tokensregex-tr-2014.pdf https://nlp.stanford.edu/pubs/tokensregex-tr-2014.pdf
- nl 10y agoYes, the rules are horrible. Yes, SUTime only has English rules I don't know of any production use cases of either Hi... Using SUTime for English, HeidelTime for non-English. Not at FB scale, but running against millions of messages per day.