• New features had slightly improved the accuracy (now is 75%), I'm still working on improving the model.

• @Trokhymovych had finished the version of this model, covering 47 languages. @MunizaA reviewed and adapted the code, and now we are coordinating with @achou to update the model on Lift Wing.

• We are discussing the schema for this (and other) ML-generated events (T331401)

The problem that I see with 1) is that we are already filtering (and rightfully so) a lot of events, meanwhile researchers may want the whole stream scored.

@elukey, do we know which is the data that is being filtered out?

And for the revert-risk model:

score:
    model_name: revertrisk
    model_version: 1.0.1
    prediction:
    - true
    probability:
     true: 0.9
     false: 0.1

This works for me.

• We are working in increasing the number of languages covered by this model. Currently, the model hosted on Lift Wing has 6. The next version will cover 47. (Keep in mind that the language agnostic model T314385 cover all wikis)

• We are discussing how to integrate this model on the Recent Changes page on MediaWiki (T329071)

• Currently I'm working on featuring engineering. The current model has around 72% accuracy on balanced data.

We should all sync up and work on some big standardized modeling design decisions and ideas. It would be great if we could share intensions and strategies for the future so we can prioritize work between ML and Event Platform especially.

I would add research on this ;)

@achou the main requestor will be the aforementioned API, for evaluating the the model. I don't expect high traffic. Let's say a couple of thousands per week.

• The paper was officially accepted in WWW'23. We made some final updates to the text.

• Discussing the integration of Revert Risk on MediaWiki: T329071

• Still working on the data evaluation. Currently I'm studying the use of tags and user groups and their relation with reverts.

Apparently this is currently done by: https://www.mediawiki.org/wiki/Extension:ORES

• We are working on manually evaluating reverts to identify the right data to train the model.

• We have submitted the Camera Ready version of this paper.
• We have started working on evaluating references in other languages.

• We are coordinating with the ML-team to create a public stream with this model's score.

• We are working on paper to share results of this model.

• Our paper was conditionally accepted at TheWebConf'23 (a.k.a WWW'23)

@Miriam please find the data in csv format here and the code used to generate it here.

@Miriam what should we do with sections that has multiple images? do you want

<page id>,<page title>,<section title>,<img_title>,<n_recommendations>

?

The model is already available, check here how to use it: T314385#8496547

• We have presented the results of this project at the WMF's "Monthly Tech All Meeting"

*The final decision should be out this week.

Hi all, we have a use case here T326179. These models are already hosted on LiftWing. The suggested end-point could be mediawiki-revision-score-revert-risk-la

For the records here a snippet (by @achou) to try the models from the WMF's cluster

For the records here a snippet (by @achou) to try this model:

For the records here a snippet (by @achou) to try this model:

Update

• The Airflow DAG is ready to be deployed.

• Multilingual and language-agnostic models has been deployed to production. Check the details in the related tasks.
• We are now onboarding @Sheilakaruku to work on developing an user-interface to work with these models (T318634)

• We have received the reviews from the WWW, and submitted the rebuttal. Now, we need to wait for the final decision.

I'm trying to implement a link-prediction task on Wikidata, to be used as proxy for claims coverage. I'm building on top of Goyal & Ferrara's work. The existing libraries might require some tweaks to work on the full Wikidata Graph, but before addressing the scalability issues I want to test this approach on a small sample to see the suitability of this approach.

Regarding article quality, you can find the scores for all revisions in all languages from 2020-01-01 until 2022-09-31 here: /user/dsaez/paramita_article_quality/scores_all_v3_from_2020-01-01.parquet (HDFS)

In T318348#8457369, @mfossati wrote:

In T318348#8454997, @CBogen wrote:

@mfossati do you have an update on the status of the evaluation for proposal 2? thanks!

I sent the data check request as per our formal cross-team process, solicited Research during the last meeting, and personally pinged @diego and @MunizaA . I haven't seen or received any update so far, I'll follow up again at the next meeting.

As mentioned before in our meetings, the main problem we have is the confusing usage of "Blue Links" as synonym of "topics". In NLP topics are either categories or clusters of documents. The second important problem we have is the lack of a evaluation task or guidelines. If we are using links as tags, and we want to evaluate the importance/relevance of such tags, we need a task, because relevance depends on the context.

Based on your work at T314384, we would love to incorporate new fields like:

• vandalism_count
• vandalism_ratio
• vandalism_reverts_ratio
• seconds_to_revert_vandalism_avg

Your feedback would be highly appreciated, so thanks in advance for your interest and happy to brainstorm together on this :)

Just to clarify we have a "revert probability", we can't claim this is "vandalism". Different from previous model we have just one single score.
Maybe you might be interested on collecting abuse filter information. I have some code to do that, and from there you might be able to compute something like "abuse filters hits".

Please remember to record your contributions on the Outreachy website! The deadline is (today) Nov 4th!

In T314384#8353933, @Pablo wrote:

It has been decided to focus on knowledge integrity risks from two categories of our taxonomy:

In T318634#8358851, @Sannan2252 wrote:

Hi @diego, should we mention a single contribution on the Outreachy website or mention our contribution in multiple parts?

Yeah! Thanks @achou ! Please, can you write here an example of how to hit the endpoint ?

Hello everybody,

I understand your concerns, but I'll start considering the "Instance Of" as the main "category" for the item. We could later try to cluster instances based on statements similarities, but I would keep that for later.

• The code is being refactored by @MunizaA and reviewed by @achou. They are trying to find the optimal architecture in order to make the code easier to maintain and update.
• We have found a poor performance from the model for anonymous edits. I'm working on updating the model to improve this.

• The code is fully functional, and can be found here.
• @MunizaA is experimenting with AirFlow to be able to run this script periodically.

In T318634#8327615, @Claire3z wrote:

In T318634#8326752, @Caseyy0000 wrote:

@diego I experienced some issues when comparing the revisions for all the 5000 recentchanges, their individual wikitext contents are quite large and running them through mwedittypes.SimpleEditTypes() return an empty output.

Hi @Caseyy0000. Please check the documentation for medittypes here. It could be that in some specific cases the library fails, but that should be very exceptional.

Looks good to me!

In T318634#8310130, @Drew21-mch wrote:

Hi everyone, I am Andy, outreachy applicant, nice to meet you all,
Please i have a question, any recomendation on any video tutorial one can watch to getting started with mediawiki/mwapi api ?
I will be grateful.

Sorry, I've read the documentation here: https://meta.wikimedia.org/wiki/User-Agent_policy and everything is clear. I'm going to close this ticket.

I see that the error says:

Scripted requests from your IP have been blocked.

However, the error persists from different IPs.

• @MunizaA had written the model to be hosted in Liftwing and shared with @achou.
• @achou is testing the model locally before uploading to Liftwing
• @diego is working on creating a new model adding new features.

• We have developed a language agnostic ML model to predict reverts.
• The model has an accuracy of 80% on a balanced dataset, compared to the 66% given by ORES.
• Research code is available here.
• @MunizaA is working in implementing the code as a service, and then @AikoChou will deploy to LiftWing. T

• We are currently working on writing the results

Thanks @Ottomata , the contract finish at December 15th.

In T315262#8160758, @cmooney wrote:

Hi @Trokhymovych thanks for sharing the key, I assume you need shell/ssh access is that correct? Could you advise of exactly what systems you'll need to ssh into?

@cmooney we need to give access to @Trokhymovych to the stat machines and Spark Cluster.

Hi @BCornwall, just to say this is a high priority for us. We are already lost 4 days of work with @MunizaA been locked-out from the servers.

@Muehlenhoff I just re-confirmed via call with @MunizaA that ssh-key is correct.

@Dzahn, there is a -ctr email: maslam-ctr@wikimedia.org , would that solve the problem?

Thanks @RhinosF1 !