Change Details

Right now, all the tasks we're giving newcomers come from maintenance templates, which say that an article needs some work in general.  We want to try to see if it would be possible to give newcomers something very specific to do -- like add a link to a specific word or phrase. @DED from the Research team is trying out an algorithm that attempts to predict that a specific word or phrase should be an internal link. We want our ambassadors to give us some initial feedback on whether this algorithm is going in the right direction. Here's what was done: # Took five articles from each of Czech, Korean, Arabic, and Vietnamese Wikipedias that have an "add links" maintenance template. # Put the wikitext from each article into a Google Doc. # Highlighted in blue the text that the algorithm thinks should be a link.  This includes a score between 0 and 1 that says how confident the algorithm is. Here are the tasks for the ambassadors: # In the documents, leave a comment on each suggested link saying whether you think it should or should not be a link.  In other words, was the algorithm correct?  You can do this by highlighting the blue text and leaving a comment. # Leave comments on a few words or phrases (maybe 5) that you think the algorithm //should have// suggested as links, but did not.  For instance, in the first Czech article on Milton Friedman, the algorithm probably should have suggested that "ekonom" be a link. A couple things to note: * @DED mentioned that the algorithm is currently guessing conservatively, and is therefore likely missing many links that it should suggest.  He plans to improve that in the future. * The algorithm doesn't account for only recommending the //first// occurrence of a word or phrase.  That can be fixed in the future. Here are the documents for each wiki: [x] [[ https://docs.google.com/document/d/1wVTIYruhvmSjX_ylPRdta2NSJAfYhz9CQ5qH5kC0pOE/edit# | Arabic ]] [] [[ https://docs.google.com/document/d/19GkmMeXsJaz10ArbgEuW7s3vMBTMGN4y_rNn-F_ZXlg/edit# | Czech ]] [x] [[ https://docs.google.com/document/d/1dw_uz5v-x3rtuA5EHu0HvFpdIH7red9ugGmJVRJ3GQ8/edit# | Korean ]] [x] [[ https://docs.google.com/document/d/1d_hEmAFhP6NNRrfMhflW4ysCmfaC3akuW9wYJTxnqkE/edit# | Vietnamese ]]

Right now, all the tasks we're giving newcomers come from maintenance templates, which say that an article needs some work in general.  We want to try to see if it would be possible to give newcomers something very specific to do -- like add a link to a specific word or phrase. @DED from the Research team is trying out an algorithm that attempts to predict that a specific word or phrase should be an internal link. We want our ambassadors to give us some initial feedback on whether this algorithm is going in the right direction. Here's what was done: # Took five articles from each of Czech, Korean, Arabic, and Vietnamese Wikipedias that have an "add links" maintenance template. # Put the wikitext from each article into a Google Doc. # Highlighted in blue the text that the algorithm thinks should be a link.  This includes a score between 0 and 1 that says how confident the algorithm is. Here are the tasks for the ambassadors: # In the documents, leave a comment on each suggested link saying whether you think it should or should not be a link.  In other words, was the algorithm correct?  You can do this by highlighting the blue text and leaving a comment. # Leave comments on a few words or phrases (maybe 5) that you think the algorithm //should have// suggested as links, but did not.  For instance, in the first Czech article on Milton Friedman, the algorithm probably should have suggested that "ekonom" be a link. A couple things to note: * @DED mentioned that the algorithm is currently guessing conservatively, and is therefore likely missing many links that it should suggest.  He plans to improve that in the future. * The algorithm doesn't account for only recommending the //first// occurrence of a word or phrase.  That can be fixed in the future. Here are the documents for each wiki: [x] [[ https://docs.google.com/document/d/1wVTIYruhvmSjX_ylPRdta2NSJAfYhz9CQ5qH5kC0pOE/edit# | Arabic ]] [x] [[ https://docs.google.com/document/d/19GkmMeXsJaz10ArbgEuW7s3vMBTMGN4y_rNn-F_ZXlg/edit# | Czech ]] [x] [[ https://docs.google.com/document/d/1dw_uz5v-x3rtuA5EHu0HvFpdIH7red9ugGmJVRJ3GQ8/edit# | Korean ]] [x] [[ https://docs.google.com/document/d/1d_hEmAFhP6NNRrfMhflW4ysCmfaC3akuW9wYJTxnqkE/edit# | Vietnamese ]]

Right now, all the tasks we're giving newcomers come from maintenance templates, which say that an article needs some work in general.  We want to try to see if it would be possible to give newcomers something very specific to do -- like add a link to a specific word or phrase. @DED from the Research team is trying out an algorithm that attempts to predict that a specific word or phrase should be an internal link. We want our ambassadors to give us some initial feedback on whether this algorithm is going in the right direction. Here's what was done: # Took five articles from each of Czech, Korean, Arabic, and Vietnamese Wikipedias that have an "add links" maintenance template. # Put the wikitext from each article into a Google Doc. # Highlighted in blue the text that the algorithm thinks should be a link.  This includes a score between 0 and 1 that says how confident the algorithm is. Here are the tasks for the ambassadors: # In the documents, leave a comment on each suggested link saying whether you think it should or should not be a link.  In other words, was the algorithm correct?  You can do this by highlighting the blue text and leaving a comment. # Leave comments on a few words or phrases (maybe 5) that you think the algorithm //should have// suggested as links, but did not.  For instance, in the first Czech article on Milton Friedman, the algorithm probably should have suggested that "ekonom" be a link. A couple things to note: * @DED mentioned that the algorithm is currently guessing conservatively, and is therefore likely missing many links that it should suggest.  He plans to improve that in the future. * The algorithm doesn't account for only recommending the //first// occurrence of a word or phrase.  That can be fixed in the future. Here are the documents for each wiki: [x] [[ https://docs.google.com/document/d/1wVTIYruhvmSjX_ylPRdta2NSJAfYhz9CQ5qH5kC0pOE/edit# | Arabic ]] [x] [[ https://docs.google.com/document/d/19GkmMeXsJaz10ArbgEuW7s3vMBTMGN4y_rNn-F_ZXlg/edit# | Czech ]] [x] [[ https://docs.google.com/document/d/1dw_uz5v-x3rtuA5EHu0HvFpdIH7red9ugGmJVRJ3GQ8/edit# | Korean ]] [x] [[ https://docs.google.com/document/d/1d_hEmAFhP6NNRrfMhflW4ysCmfaC3akuW9wYJTxnqkE/edit# | Vietnamese ]]