ORES (Object Revision Evaluation Service) is a machine learning project that can predict if an edit on Wikipedia article is damaging or good faith. And there is a model to predict if an edit will be reverted. What we are using on the project is the wp10 model that predicts Wikipedia 1.0  assessment class of an article or draft. The API can also enlist the features used to make the prediction. The API’s use is elevated to also use the features that can give some insight to how the article is structured and the faux pas in the edit.

The features for predicting the article class are listed below. I wasn’t able to find the official documentation for these features even after going through the source code. So I documented the features based on some googling.

  • feature.english.stemmed.revision.stems_length : nil
  • feature.enwiki.main_article_templates : Templates are standard texts the needs to repeatedly included in several pages.When a Wikipedia article is large, it is often written in summary style. This template is used after the heading of the summary, to link to the subtopic article that has been summarized.
  • feature.enwiki.revision.category_links : Category links are links that when inserted into a page that places the page in that particular category. For example [[:Category:Title]] links to category
  • feature.enwiki.revision.cite_templates : Cite templates are templates that come up on a citation link. But it is to be noted that citation can be inserted even without the templates
  • feature.enwiki.revision.cn_templates : The {{citation needed span}} template may be used when there is unsourced text that should be attributed to a reference citation.
  • feature.enwiki.revision.image_links : Links to images 
  • feature.enwiki.revision.infobox_templates : Templates that have text to give a key value based summary on the side of Wikipedia articles. 
  • feature.wikitext.revision.chars : Number of characters in the article
  • feature.wikitext.revision.content_chars : Characters that are only in the content of the article
  • feature.wikitext.revision.external_links : Links to pages that are not in Wikipedia
  • feature.wikitext.revision.headings_by_level(2) : Structure of the article
  • feature.wikitext.revision.headings_by_level(3) : Structure of the article
  • feature.wikitext.revision.ref_tags : The links to citations which can be inserted using cite_templates
  • feature.wikitext.revision.templates : Total number of templates
  • feature.wikitext.revision.wikilinks : Link to pages in Wikipedia

Already some features are used to provide feedback. Another approach was to use the predicted class to provide feedback. The usual Article classes are assigned to an article through talk pages by fellow Wikipedia editors. It is not instantaneous hence using ORES predictions can also get some feedback on Sandbox articles which students mostly use to create fresh content. The article classes already have a pretty good documentation here. It also has documentation on what are the aspects in the article that needs to be improved.

We have already deployed the feature and looking for feedback on the feature. The feedback is crucial in the future course of the feature.