Benchmarking for the different algorithms

[assem-ch]

No description provided.

[sneetsher]

@assem-ch , It will be much work for one man. I see many separate reports for different words, it is not practical to follow each manually. Stemmer is never be 100% perfect as manual learned method by human.

• Better to switch to coverage based tests, using manually processed data. At least using word root/derivatives without attached pronouns & conjunctions which may be already available in the web with free use license. (Ex. harfbuzz coverage testing was done on Wikipedia data rendered in multiple browsers. https://github.com/harfbuzz/harfbuzz-testing-wikipedia). Possibly, I can implement them if you could find for me some data-sets on the web, I can use/crap them.
• We merge all reports about specific words in one or multiple reports (each report for one algorithms and one version). I can help in this, just write short, debugging/reporting instructions in readme file. like reporting release number and how can user know which algorithm is used.

[sneetsher]

Btw, not all release packages have mentioned version.

[sneetsher]

@assem-ch If you think the Arabic stemmer is too worthy for Alfanous & too many Arabic project that I put considerable time in it.

I will see If I go with separate project for stop-word list and derivatives list build by crowd-sourced verification to get high quality test data for the stemmer. Because, I see few arabic stop-words list and basically few persons efforts.

All depend on the way how to reach the interested/effective Arab community.

[assem-ch]

@sneetsher there is already a project for testing data https://github.com/ibnmalik/golden-corpus-arabic, it can be exposed with the stemmer to get new suggests from users.

This is my phd project and I should really focus on it... I will work on a demo/review web app for it to welcome feedback and improve the visibility.

For alfanous its not too worthy, but it fix a gap for stemming words that dont exist in quran exactly but in other forms