Hi James,

On Tue, Dec 12, 2017 at 5:57 AM, James Salsman jsalsman@gmail.com wrote:

Leila and Lani,

The Article Expansion Recommendation System is an absolutely spectacular project, which will clearly very substantially improve the encyclopedia in ways that perhaps no other single effort has come near to being able, so I can't wait to learn more about it.

Thanks for the words of encouragement. :)

But I might not be able to make the live-stream time,

A quick note (which you already probably know): you can watch the showcase afterwards, and you know where to find us, so do ping with questions. (I may be slow in responding in the next 2 weeks, but I will get back to you).

so I want to get in this question in advance:

Are you using or do you plan to use ORES quality predictions, the upcoming article importance predictions, and pageview statistics to rank article expansion recommendations?

At the moment, we're not using the systems you mentioned but that's because we're not at the ranking stage, yet. Once we get there, we will start interacting with them more, to figure out where we can use what. In our previous work for recommending what articles to be created, we ranked missing articles in a given language (Section 2.2. in https://arxiv.org/abs/1604.03235). When we're working with already existing articles, our job can become easier, given that, as you mentioned, more information is available about the article. (For example, we had to build a prediction model that would predict article pageviews if they get created in the previous work, while pageviews are now available to us for the already existing articles.).

While we're at it: While ranking is an important step, so far what we have learned is that there is so much missing articles/content across languages that it suffices if the ranking is "good enough". The combination of a good enough ranking to give us, let's say the top 300K thousand high priority items to contribute to, and empowering users to tune recommendations based on their interests will be powerful enough from the users' perspective.

I will add a few slides to talk about the challenges we see if article expansion, inspired by your question. :)

Best, Leila

...

The next Research Showcase will be live-streamed this Wednesday, December 13, 2017 at 11:15 AM (PST) 18:15 UTC.

YouTube stream: https://www.youtube.com/watch?v=OoVwus1Owtk

As usual, you can join the conversation on IRC at #wikimedia-research. And, you can watch our past research showcases here.

This month's presentation: "The State of the Article Expansion Recommendation System"

By Leila Zia

Only 1% of English Wikipedia articles are labeled with quality class Good or better, and 37% of the articles are stubs. We are building an article expansion recommendation system to change this in Wikipedia, across many languages. In this presentation, I will talk with you about our current thinking of the vision and direction of the research that can help us build such a recommendation system, and share more about one specific area of research we have heavily focused on in the past months: building a recommendation system that can help editors identify what sections to add to an already existing article. I present some of the challenges we faced, the methods we devised or used to overcome them, and the result of the first line of experiments on the quality of such recommendations (teaser: the results are really promising. The precision and recall at 10 is 80%.)