Hi Ludovic, This work sounds interesting, I'm looking forward to learning more about it as your papers come out! I read through the post on LinkedIn and from how I interpret it you are only looking at two quality classes (Features Articles vs other articles). This seems somewhat odd to me and I'd like to know more about why? The current trend when it comes to predicting article quality in the English Wikipedia does not limit the prediction problem to just FAs vs the rest, instead it's using the whole quality scale[1]. See the list below for some papers along this line of research. I'm also really curious about what "standardize the cognitive accessibility of Wikipedia" means? That might mean more than just "article quality", hence why I'm asking. All that being said, I think the approach sounds interesting and probably adds some signal, so I'm curious to learn more how it works and performs. References: - Warncke-Wang, M., Cosley, D., & Riedl, J. Tell me more: an actionable quality model for Wikipedia. OpenSym/WikiSym 2013. [We argue that metadata isn't useful because contributors can't change it] - Warncke-Wang, M., Ayukaev, V. R., Hecht, B., & Terveen, L. G. The success and failure of quality improvement projects in peer production communities. CSCW 2015. [See the Appendix for details of the improved model and how to get good training data] - https://www.mediawiki.org/wiki/ORES builds upon the 2015 paper and is a readily accessible API, reference datasets are available on figshare <https://figshare.com/articles/English_Wikipedia_Quality_Asssessment_Dataset/1375406> and also in the GitHub repository <https://github.com/wikimedia/articlequality>. Now the benchmark to compare against, as in the three other papers listed below. - Dang, Q. V., & Ignat, C. L. Measuring quality of collaboratively edited documents: the case of Wikipedia. CIC 2016. [Shows that adding readability features can improve predictions] - Dang, Q. V., & Ignat, C. L. An end-to-end learning solution for assessing the quality of Wikipedia articles. OpenSym 2017. [Shows the performance of RNNs, also contains an important discussion of performance, interpretability, etc] I also came across this recent paper by Schmidt and Zangerle that reports significant improvements, but haven't yet had the time to read the paper closely: - Schmidt, M., & Zangerle, E. Article quality classification on Wikipedia: introducing document embeddings and content features. OpenSym 2019. Footnotes: 1. Typically without A-class articles due to how few of them they are. Cheers, Morten On Mon, 23 Sep 2019 at 13:09, Ludovic Bocken &lt;lbocken(a)gmail.com&gt; wrote: