Done.
Notice that the list of articles is new (the sample was random, now I fixed to get always the same articles in case we need to fix something).

Please find the samples required in this link.

Updates

• We are exploring cross-lingual article quality. Details on this task: T305390

Thanks @Isaac. I'll check this.

Hi @santhosh, @MunizaA has created a new dump just with pairs with probabilities > 0.9, you can find here. We think that 0.9 might be to high, let us know if you want to try with other values between 0.5 and 0.9.

pyspark2 --master yarn --deploy-mode client --executor-memory 8g --driver-memory 8g --conf spark.dynamicAllocation.maxExecutors=128

but result is same.

In this case fasttext is running on the driver. You can increase the memory driver when calling the spark env, replacing this:

• We have published the documentation about this project here.
• All code and data is available and linked on the documentation page.

• Results and dataset details has been published. You can find them here.
• I'm changing this task as resolves, and updating also in betterworks.

In T305888#7847242, @Aklapper wrote:

Hi @diego. Could you please associate at least one active project with this task (via the Add Action... → Change Project Tags dropdown)? This will allow others to get notified, or see this task when searching via projects. Thanks!

• Last week we have done the onboarding on using PySpark and cluster data.
• As first step @paramita_das will be working on obtain the article quality distribution along the time for enwiki.

• Last week we have done the onboarding on using PySpark and cluster data.
• @Aitolkyn is starting to explore how to match bad references with pageviews.

• We have published the alignments for 205 languages here.
• Each folder contains the alignments from that language to all others. For example 'enwiki' contains the alignments from English to all the other wikis.
• The format is SQLite. @santhosh could you confirm you are able to read the files?
• We are working on the algorithm and output documentation.

• The dataset has been released here.
• The paper will be published in May.

• We finished this project, results can be found on Meta, the code and models could be found in Gitlab.
• I'll discuss future work with @Lydia_Pintscher.

Thanks @Isaac for these inputs. There was a mistake on the title, this work is about article quality and not specifically about citations.

In T305298#7828668, @MoritzMuehlenhoff wrote:

@diego: We also need the estimated end date of the internship (you'll be contacted two weeks before it expires whether to extend access or not).

The internship ends on June 24th.

@diego: We also need the estimated end date of the internship (you'll be contacted two weeks before it expires whether to extend access or not).

@Aitolkyn please update the task description with your SSH key.

@paramita_das please update the task description with your SSH key.

• We have tested our model model on the CX dataset (sections translations done using the CX Tool).
• Results are showing a good performance. @MunizaA please report the precision@5 for the top-100 languages pairs.
• Now, we run the alignments for all the languages, and the results will be ready early next week.

• I'm preparing a short presentation with the main findings of this project. I'm planning to present these results during the next disinformation working group meeting.

Updates

• We decided to go back to the XGBoost based model, because the results were better than using the Spark implementation.
• We noticed a decrease on precision when considering under-resourced languages. Our hypothesis is that the quality of embeddings created by M-Bert is not very high. We decided to create a second model, language-agnostic, and then compare the results. Our intuition is that for some languages the language agnostic model will be better.
• We plan to release all these results at the end of next week.

Updates

• I was comparing the results when adding anonymous edits, until now I haven't find major differences with the previous results. I'll continue working on this during the next week before my next meeting with Lydia.

• I've presented the main results of this work during the Tuesday Research Sessions, slides can be find here.

• We are working in applying the model at scale. @MunizaA has been experimenting with native spark libraries to see if is possible to replace external dependencies. The quality of firsts results are not satisfactory, so we are exploring alternatives.

• No updates.

• Our collaborators got their publication accepted, the dataset should be released in the following week.s

Updates

• No updates

• We are fine tuning the model.

• I'm currently studying the propagation of "climate change" related items (thanks Isaac for the dataset)

• We meet with Lydia and discussed the current results.
• We reviewed the results confirming that most co-edited items corresponds to on going events, even when we change the time window to be considered.
• Now, I'll be studying the relevance/prevalence of anonymous edits on popular content.

Updates

• No update.

• With @MunizaA we have annotated data in Spanish to English and Urdu to English.
  • We found that popularity of sections (amount of articles they appear) has a huge impact on the results' quality.
  • While for popular sections there are multiple possible translations, the most infrequent ones usually has 1 or 2.
  • We are trying to improve the model to address these issues.
• We are also analyzing how to use MT to improve the results.

• We are shaping the paper and checking which new experiments would be required.

• No updates

• No updates.

• @MunizaA has uploaded this sample files containing several languages Each of them contains the top-200 most frequent sections in the source language.
• @Pginer-WMF , please have a look on them. Keep in mind that we are focusing in recall more than precision. By now, we are showing the top-20 most similar target sections, per source section.
• I'll coordinate a meeting in the following days to discuss how to tune these results.

• No updates

• I'm working in identifying collaborative edits on wikidata items not related to current events.

• We are in the process of writing the paper.

• No updates

• We are preparing the data on some of the language pairs suggested by @Pginer-WMF
• @MunizaA has uploaded the code here: https://gitlab.wikimedia.org/mnz/section-alignment/-/tree/muniza-notebooks/notebooks

• No updates

• No updates.

• We have done manual sanity checks on the data extraction pipeline, confirming that is working properly.
• Next steps will be to run the model in 20 new languages.

• No updates

• We are now focusing in understanding collaborations patterns: when/how more than user edits the same item in a given period of time.
  • We found that in Wikidata such collaborations are less frequent than in other Wikimedia projects.
  • We also found that items edited by more than one user are usually related to on going events (awards, deaths, releases)
• I'll present some of these findings:
  • On research meeting (Tuesday) in March
  • And @Lydia_Pintscher will propose a date probably in April to present these results to the Wikidata folks.

• I'm gathering all the last results to organize them to write the report.

• We are analyzing the results showed above before deciding the new steps.

Updates

• No updates

• I'm organizing the new results to be discussed with the stakeholder.

yes.

• @MunizaA has run the first experiments to compare the results with the new language model, with our old FastText-based model, obtaining promising results. (@MunizaA please share the new results here.)
• The next steps are:
  • Test the model for language pairs without training data.
  • Estimate the time required to run the model in the 100+ languages supported by this new approach.

• I'm focusing on modeling the relationship between topics and collaborations/controversies.
  • I'm working on graph representation of these components

• I've updated the task according to the new plan discussed with our collaborators.

• No updates

I've updated the code here https://github.com/digitalTranshumant/templatesAlignment/blob/master/02alignmentsSpark.ipynb

• @MunizaA has developed the full pipeline to efficiently extract all the features used on the original model, such as link similarity and edit distance.
• We are currently preparing the experiment to validate our results using the new Language model (to replace FastText).

• We have received the report from our collaborators with the description of the dataset, and the result of their model.
• We will coordinate for releasing the dataset during this FY.

• We have seen that few items are edited by more than one user.
• We are currently researching about the item and users characteristics related to collaborative work.

• We are testing a new DL model to predict content propagation, using content reliability as one of the features.

Oh got it! This setup has been changed around one year ago. Now we all use the spark environments provided by the JupyterHub.

@KartikMistry this looks like a pyspark configuration issue, which kernel are you using?

• The paper can be found here, and this task is done.

• No updates this week. I'm going to meet with the stakeholder next week.

Updates

• No updates

• The dataset was created, I'm coordinating with our collaborators to discuss the time and format of the data release.

• We obtained the first results with new language models. @MunizaA could you please report the numbers here?

• @MunizaA is testing new languages models that could be more efficient and possible accurate than the FastText embeddings used in the previous experiments.

• I've been working on classifier to predict reverts.
  • The current classifier uses article (item), revision and user information.
  • On a balance test set, the actual model gets results over 70% of accuracy
  • However, there is a set of caveats to be considered:
    • 'auto-reverts': users can revert themselves, this shouldn't be consider as signal of controversy. We need to analyze more this behavior.
    • power-users: we need to take in account that a small set of users produces most of the edits and reverts, this behavior could affect our results. We are working on different sampling method to address this issue.
• The meta page was updated with the results in Q1 and partial results in Q2.

Updates

• Meta page was updated.

Updates

• No updates

• The paper has been accepted in the IEEE BigData 2021 conference.
• I'll upload the paper and write documentation in meta in the following weeks.

• No updates this week.