I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
Synopsis:
Create a Pywikibot to find articles in given categories, category trees, and lists. For each such article, add in-line templates to indicate the location of passages with (1) facts and statistics which are likely to have become out of date and have not been updated in a given number of years, and (2) phrases which are likely unclear. Use a customizable set of keywords and the DELPH-IN LOGIN parser [http://erg.delph-in.net/logon] to find such passages for review. Prepare a table of each word in article dumps indicating its age. Convert flagged passages to GIFT questions [http://microformats.org/wiki/gift] for review and present them to one or more subscribed reviewers. Update the source template with the reviewer(s)' answers to the GIFT question, but keep the original text as part of the template. When reviewers disagree, update the template to reflect that fact, and present the question to a third reviewer to break the tie.
Possible stretch goals for Global Learning Xprize Meta-Team systems [http://www.wiki.xprize.org/Meta-team#Goals] integration TBD.
Best regards, James Salsman
On 2/12/15, James Salsman jsalsman@gmail.com wrote:
I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
Synopsis:
Create a Pywikibot to find articles in given categories, category trees, and lists. For each such article, add in-line templates to indicate the location of passages with (1) facts and statistics which are likely to have become out of date and have not been updated in a given number of years, and (2) phrases which are likely unclear. Use a customizable set of keywords and the DELPH-IN LOGIN parser [http://erg.delph-in.net/logon] to find such passages for review. Prepare a table of each word in article dumps indicating its age. Convert flagged passages to GIFT questions [http://microformats.org/wiki/gift] for review and present them to one or more subscribed reviewers. Update the source template with the reviewer(s)' answers to the GIFT question, but keep the original text as part of the template. When reviewers disagree, update the template to reflect that fact, and present the question to a third reviewer to break the tie.
Possible stretch goals for Global Learning Xprize Meta-Team systems [http://www.wiki.xprize.org/Meta-team#Goals] integration TBD.
Best regards, James Salsman
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Have you run this by Wikipedians? (I'm assuming enwikipedia would be your target audience). I would recommend making sure that enwikipedia is politically ok with this first, since it involves adding a bunch of templates to articles, as it would suck for a gsoc student if their work wasn't used due to politics happening at the end.
Prepare a table of each word in article dumps indicating its age.
This in itself is a non-trivial problem (for a gsoc student anyways), assuming you need it for the entire enwikipedia, and you need it up to date as soon as people edit. Even getting the student sufficient storage and CPU resources to actually compute that could potentially be difficult (maybe?)
Convert flagged passages to GIFT questions for review and present them to one or more subscribed reviewers
Wouldn't you want to give the reviewers an actual form where they can fill out the questions, not something in a markup language (Unless you mean you want them to store it in that form internally,which seems like a rather minor implementation detail)
--bawolff
Brian Wolff wrote:
Have you run this by Wikipedians? ... since it involves adding a bunch of templates....
The 2009 strategy proposal linked towards the end, of which the GSoC proposal is a limited subset, did not get any criticism after several high-profile opportunities. Both proposals are intended to be language- and keyword-neutral. It would be best if the initial bot were built and tested on some other wiki than the English Wikipedia, so the Bot Approvals Group will be able to get concrete answers to any questions they might have.
The use of templates should be optional; one way to do that would be to allow the use of a mirroring namespace to hold the templates, instead of the primary namespace. But there is probably a better way. Thank you for something so interesting to think about.
*Prepare a table of each word in article dumps indicating its age. *
*> * *> *This in itself is a non-trivial problem (for a gsoc student anyways),
It would be non-trivial for a large production dump, but for a small subset of articles in a given dump, there are deterministic algorithms which perform with sufficient accuracy to form the specified partial basis of a selection heuristic. Creating such a table is at worst O(N) in revisions, but there are ways to hash words with N-gram contexts so that moved and blanked text is more likely to be treated correctly than what raw diffs would lead people to believe might happen. This is equivalent to the general blame problem, and I look forward to explaining the history of the problem (see e.g., http://wikitrust.soe.ucsc.edu/talks-and-papers ) to show why the N-gram hash solution is best.
*Convert flagged passages to GIFT questions for review and*
*>> present them to one or more subscribed reviewers *
Wouldn't you want to give the reviewers an actual form where they can fill out the questions?
Yes, and I want to store questions in GIFT format to allow follow-on integration with the Global Learning Xprize Meta-Team deliverables. Presentation of a GIFT question means converting it to a form instead of just displaying it in markup. The question pertaining to whether direct integration is a reasonable follow-on goal depends on the extent to which branching scenario interactive fiction role-play content, such as shown in http://www.capuano.biz/papers/EL_2014.pdf and http://talknicer.com/GLMORS_2014.pdf can be automatically created. I think it can be, and look forward to discussing the matter in detail with co-mentor volunteers. http://talknicer.com/GLMORS_2014.pdf
On Thursday, February 12, 2015, James Salsman jsalsman@gmail.com wrote:
I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
Synopsis:
Create a Pywikibot to find articles in given categories, category trees, and lists. For each such article, add in-line templates to indicate the location of passages with (1) facts and statistics which are likely to have become out of date and have not been updated in a given number of years, and (2) phrases which are likely unclear. Use a customizable set of keywords and the DELPH-IN LOGIN parser [http://erg.delph-in.net/logon] to find such passages for review. Prepare a table of each word in article dumps indicating its age. Convert flagged passages to GIFT questions [http://microformats.org/wiki/gift] for review and present them to one or more subscribed reviewers. Update the source template with the reviewer(s)' answers to the GIFT question, but keep the original text as part of the template. When reviewers disagree, update the template to reflect that fact, and present the question to a third reviewer to break the tie.
Possible stretch goals for Global Learning Xprize Meta-Team systems [http://www.wiki.xprize.org/Meta-team#Goals] integration TBD.
Best regards, James Salsman
I'd suggest, James, that relying on suggestions from a six-year-old strategy document when we're about to start a new strategic session, isn't the best course of action.
I'd also query what exactly is the plan for doing something with this information. Collecting lists of things that might no longer be up to date when there is no correlating action plan for updating the same information is probably not good use of anyone's time or effort.
Risker/Anne
On 13 February 2015 at 07:57, James Salsman jsalsman@gmail.com wrote:
Brian Wolff wrote:
Have you run this by Wikipedians? ... since it involves adding a bunch of templates....
The 2009 strategy proposal linked towards the end, of which the GSoC proposal is a limited subset, did not get any criticism after several high-profile opportunities. Both proposals are intended to be language- and keyword-neutral. It would be best if the initial bot were built and tested on some other wiki than the English Wikipedia, so the Bot Approvals Group will be able to get concrete answers to any questions they might have.
The use of templates should be optional; one way to do that would be to allow the use of a mirroring namespace to hold the templates, instead of the primary namespace. But there is probably a better way. Thank you for something so interesting to think about.
*Prepare a table of each word in article dumps indicating its age. *
*> * *> *This in itself is a non-trivial problem (for a gsoc student anyways),
It would be non-trivial for a large production dump, but for a small subset of articles in a given dump, there are deterministic algorithms which perform with sufficient accuracy to form the specified partial basis of a selection heuristic. Creating such a table is at worst O(N) in revisions, but there are ways to hash words with N-gram contexts so that moved and blanked text is more likely to be treated correctly than what raw diffs would lead people to believe might happen. This is equivalent to the general blame problem, and I look forward to explaining the history of the problem (see e.g., http://wikitrust.soe.ucsc.edu/talks-and-papers ) to show why the N-gram hash solution is best.
*Convert flagged passages to GIFT questions for review and*
*>> present them to one or more subscribed reviewers *
Wouldn't you want to give the reviewers an actual form where they can fill out the questions?
Yes, and I want to store questions in GIFT format to allow follow-on integration with the Global Learning Xprize Meta-Team deliverables. Presentation of a GIFT question means converting it to a form instead of just displaying it in markup. The question pertaining to whether direct integration is a reasonable follow-on goal depends on the extent to which branching scenario interactive fiction role-play content, such as shown in http://www.capuano.biz/papers/EL_2014.pdf and http://talknicer.com/GLMORS_2014.pdf can be automatically created. I think it can be, and look forward to discussing the matter in detail with co-mentor volunteers. http://talknicer.com/GLMORS_2014.pdf
On Thursday, February 12, 2015, James Salsman jsalsman@gmail.com wrote:
I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
Synopsis:
Create a Pywikibot to find articles in given categories, category trees, and lists. For each such article, add in-line templates to indicate the location of passages with (1) facts and statistics which are likely to have become out of date and have not been updated in a given number of years, and (2) phrases which are likely unclear. Use a customizable set of keywords and the DELPH-IN LOGIN parser [http://erg.delph-in.net/logon] to find such passages for review. Prepare a table of each word in article dumps indicating its age. Convert flagged passages to GIFT questions [http://microformats.org/wiki/gift] for review and present them to one or more subscribed reviewers. Update the source template with the reviewer(s)' answers to the GIFT question, but keep the original text as part of the template. When reviewers disagree, update the template to reflect that fact, and present the question to a third reviewer to break the tie.
Possible stretch goals for Global Learning Xprize Meta-Team systems [http://www.wiki.xprize.org/Meta-team#Goals] integration TBD.
Best regards, James Salsman
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Risker wrote:
... relying on suggestions from a six-year-old strategy document when we're about to start a new strategic session, isn't the best course of action.
A strategy proposal which never garnered criticism after so many opportunities would seem to qualify as at least an emergent strategy within the meaning of the slide and narrative at https://www.youtube.com/watch?v=N4Kvj5vCaW0&t=19m30s
Furthermore, the initial limited subtask would be much more difficult to evaluate as a strategy without a working prototype, including by the Bot Approvals Group which demands working code before making a final decision on implementation. Trying to second guess the BAG is presumptuous.
Is it possible that supporting updates to out of date articles would not be part of any successful strategy for the Foundation? I have posted multiple series of statistics to wiki-research-l in the past several months proving that quality issues are transitioning from creating new content to maintaining old content, and will be happy to recapitulate them should anyone suggest that they think it could be.
what exactly is the plan for doing something with this information.
It will be made available to volunteers as a backlog list which community members may or may not choose to work on. The Foundation can't prescribe mandatory content improvement work without putting the safe harbor provisions in jeopardy. Volunteers will be attracted to working on such updates in proportion to the extent they see them as being a worthy use of their editing time.
I have additional detailed plans for testing which I will be happy to discuss with interested co-mentors, because depending on available resources there could be a way to eliminate substantial duplication of effort.
I have updated the synopses at https://www.mediawiki.org/wiki/Accuracy_review and https://phabricator.wikimedia.org/T89416
Best regards, James Salsman
I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
James, it received a single "support" vote. It was not included in any final strategy documents. I think it could more accurately be described as "something not worthy of the attention of the Wikimedia community".
While I have no doubt that the WMF and the Wikimedia community care about the accuracy of articles, there's no basis to believe that this project would have any effect on said accuracy, or that it will actually identify inaccuracies in the text; it looks for "old" edits that haven't been revised using keywords that may or may not have any relevance to the accuracy of the information. It would require tens of thousands of person-hours (if not more) to analyse the data, and not a single article would be improved. Your proposal requires massive time commitment from reviewers of the data obtained in order to assess whether or not an update should be requested; it doesn't even fix out-of-date information. There is no indication at all that there is any interest on the part of Wikipedians to review data identified in the manner you propose.
Risker/Anne
On 13 February 2015 at 12:58, James Salsman jsalsman@gmail.com wrote:
Risker wrote:
... relying on suggestions from a six-year-old strategy document when we're about to start a new strategic session, isn't the best course of action.
A strategy proposal which never garnered criticism after so many opportunities would seem to qualify as at least an emergent strategy within the meaning of the slide and narrative at https://www.youtube.com/watch?v=N4Kvj5vCaW0&t=19m30s
Furthermore, the initial limited subtask would be much more difficult to evaluate as a strategy without a working prototype, including by the Bot Approvals Group which demands working code before making a final decision on implementation. Trying to second guess the BAG is presumptuous.
Is it possible that supporting updates to out of date articles would not be part of any successful strategy for the Foundation? I have posted multiple series of statistics to wiki-research-l in the past several months proving that quality issues are transitioning from creating new content to maintaining old content, and will be happy to recapitulate them should anyone suggest that they think it could be.
what exactly is the plan for doing something with this information.
It will be made available to volunteers as a backlog list which community members may or may not choose to work on. The Foundation can't prescribe mandatory content improvement work without putting the safe harbor provisions in jeopardy. Volunteers will be attracted to working on such updates in proportion to the extent they see them as being a worthy use of their editing time.
I have additional detailed plans for testing which I will be happy to discuss with interested co-mentors, because depending on available resources there could be a way to eliminate substantial duplication of effort.
I have updated the synopses at https://www.mediawiki.org/wiki/Accuracy_review and https://phabricator.wikimedia.org/T89416
Best regards, James Salsman
I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
James Salman's proposal for Google Summer of Code is being discussed at https://phabricator.wikimedia.org/T89416 and I think we need more feedback from more people. I have summarized in https://phabricator.wikimedia.org/T89416#1087673 my opinion against including this project until receiving community backing. James' arguments also must be considered though, and I think we are in a situation where more voices are welcome.
On Fri, Feb 13, 2015 at 7:31 PM, Risker risker.wp@gmail.com wrote:
While I have no doubt that the WMF and the Wikimedia community care about the accuracy of articles, there's no basis to believe that this project would have any effect on said accuracy, or that it will actually identify inaccuracies in the text; it looks for "old" edits that haven't been revised using keywords that may or may not have any relevance to the accuracy of the information. It would require tens of thousands of person-hours (if not more) to analyse the data, and not a single article would be improved. Your proposal requires massive time commitment from reviewers of the data obtained in order to assess whether or not an update should be requested; it doesn't even fix out-of-date information. There is no indication at all that there is any interest on the part of Wikipedians to review data identified in the manner you propose.
Risker/Anne
On 13 February 2015 at 12:58, James Salsman jsalsman@gmail.com wrote:
Risker wrote:
... relying on suggestions from a six-year-old strategy document when we're about to start a new strategic session, isn't the best course of action.
A strategy proposal which never garnered criticism after so many opportunities would seem to qualify as at least an emergent strategy within the meaning of the slide and narrative at https://www.youtube.com/watch?v=N4Kvj5vCaW0&t=19m30s
Furthermore, the initial limited subtask would be much more difficult to evaluate as a strategy without a working prototype, including by the Bot Approvals Group which demands working code before making a final decision on implementation. Trying to second guess the BAG is presumptuous.
Is it possible that supporting updates to out of date articles would not be part of any successful strategy for the Foundation? I have posted multiple series of statistics to wiki-research-l in the past several months proving that quality issues are transitioning from creating new content to maintaining old content, and will be happy to recapitulate them should anyone suggest that they think it could be.
what exactly is the plan for doing something with this information.
It will be made available to volunteers as a backlog list which community members may or may not choose to work on. The Foundation can't prescribe mandatory content improvement work without putting the safe harbor provisions in jeopardy. Volunteers will be attracted to working on such updates in proportion to the extent they see them as being a worthy use of their editing time.
I have additional detailed plans for testing which I will be happy to discuss with interested co-mentors, because depending on available resources there could be a way to eliminate substantial duplication of effort.
I have updated the synopses at https://www.mediawiki.org/wiki/Accuracy_review and https://phabricator.wikimedia.org/T89416
Best regards, James Salsman
I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Risker wrote:
... it received a single "support" vote
There are two supporters including myself who indicated they are willing to work on it, and it also recieved support at https://strategy.wikimedia.org/wiki/Favorites/Lodewijk
Many of the implemented proposals received less formal process support, for example: https://strategy.wikimedia.org/wiki/Proposal:Foundation-Announce-l https://strategy.wikimedia.org/wiki/Proposal:Create_Wikisource_for_Yiddish https://strategy.wikimedia.org/wiki/Proposal:Allow_IPs_to_edit_sections_on_E...) https://strategy.wikimedia.org/wiki/Proposal_talk:Implement_secret_ballots_(...) https://strategy.wikimedia.org/wiki/Proposal:IPhone/iPod_Touch_Offical_Wikip...) https://strategy.wikimedia.org/wiki/Proposal:Mobiltelefonversion_von_Wikiped...) ... and at least four more just that I have looked through so far.
Moreover, according to the vote scoring system, I believe it ranked in the top 8% out of several hundred proposals, although that information is apparently no longer available.
... there's no basis to believe that this ... will actually identify inaccuracies in the text
Do you believe that if you find an article about a geographic region with the words "population 1,234,567" or "gross national product" within the same grammatical clause as a number, and you know that text was inserted 10 years ago, that you have not found a likely out-of-date inaccuracy? What reason could there possibly be to believe otherwise?
... It would require tens of thousands of person-hours (if not more) to analyse the data, and not a single article would be improved.
On the contrary, we can try it on 100 randomly selected vital articles, and if we don't have enough data to make an extrapolation with useful confidence intervals, we can try it on a slightly larger sample of them. This is something the GSoC students can do themselves, without and volunteer support. But what reason is there to believe that such support won't be forthcoming if requested from the copyeditor's guild or similar wikiproject, for example?
... Your proposal requires massive time commitment from reviewers
Why would it require any more time commitment than the existing 17,200 articles in [[Category:Wikipedia articles needing factual verification]]? Where is the requirement? Volunteer editors are free to spend their time in the manner which they believe will best serve improvements.
... it doesn't even fix out-of-date information.
Do you think actual fact checking should be done by people or bots?
... There is no indication at all that there is any interest on the part of Wikipedians to review data identified in the manner you propose.
Most of the WP:BACKLOG categories have articles entering and exiting them every day. What reason is to believe that articles selected by an automated accuracy review process would be any different?
... there's no basis to believe that this project would have any effect on accuracy
Even if you had airtight evidence that was incontrovertibly true (and for the reasons above, there can obviously be no such evidence) wouldn't it still be the case that there would only be one way to find out?
Best regards, James Salsman
On Fri, Feb 13, 2015 at 10:58 AM, James Salsman jsalsman@gmail.com wrote:
Risker wrote:
... relying on suggestions from a six-year-old strategy document when we're about to start a new strategic session, isn't the best course of action.
A strategy proposal which never garnered criticism after so many opportunities would seem to qualify as at least an emergent strategy within the meaning of the slide and narrative at https://www.youtube.com/watch?v=N4Kvj5vCaW0&t=19m30s
Furthermore, the initial limited subtask would be much more difficult to evaluate as a strategy without a working prototype, including by the Bot Approvals Group which demands working code before making a final decision on implementation. Trying to second guess the BAG is presumptuous.
Is it possible that supporting updates to out of date articles would not be part of any successful strategy for the Foundation? I have posted multiple series of statistics to wiki-research-l in the past several months proving that quality issues are transitioning from creating new content to maintaining old content, and will be happy to recapitulate them should anyone suggest that they think it could be.
what exactly is the plan for doing something with this information.
It will be made available to volunteers as a backlog list which community members may or may not choose to work on. The Foundation can't prescribe mandatory content improvement work without putting the safe harbor provisions in jeopardy. Volunteers will be attracted to working on such updates in proportion to the extent they see them as being a worthy use of their editing time.
I have additional detailed plans for testing which I will be happy to discuss with interested co-mentors, because depending on available resources there could be a way to eliminate substantial duplication of effort.
I have updated the synopses at https://www.mediawiki.org/wiki/Accuracy_review and https://phabricator.wikimedia.org/T89416
Best regards, James Salsman
I invite review of this preliminary proposal for a Google Summer of Code project: http://www.mediawiki.org/wiki/Accuracy_review
If you would like to co-mentor this project, please sign up. I've been a GSoC mentor every year since 2010, and successfully mentored two students in 2012 resulting in work which has become academically relevant, including in languages which I can not read, i.e., http://talknicer.com/turkish-tablet.pdf .) I am most interested in co-mentors at the WMF or Wiki Education Foundation involved with engineering, design, or education.
Furthermore, the initial limited subtask would be much more difficult to evaluate as a strategy without a working prototype, including by the Bot Approvals Group which demands working code before making a final decision on implementation. Trying to second guess the BAG is presumptuous.
es
Im not saying you need formal approval from BAG approval before you begin. Im saying you should have an informal discussion on VP or somewhere to make sure that relavent stakeholders think the idea would potentially be useful in principle.
Six years is a long time, people change, things change, especially for a proposal that while didnt garner a lot of opposisition, didnt garner people jumping up and down in support either.
--bawolff
wikitech-l@lists.wikimedia.org