Research talk:Anonymous mobile editing in Italian Wikipedia/Work log/2014-12-31

Wednesday, December 31, 2014[edit]

I have a couple hours today while I am waiting for something else to finish, so I figured I'd spend the time updating my previous work. I said that I'd look into patroller workloads (e.g. reverts, warnings, bannings, etc) but I think I'll just look at the trends first.

First things first, I'm updating my query to run until the end of 2014.

Big query

SET @start = "20131101";
SET @end = "20141231";

SELECT
    date,
    IFNULL(mobile_edits.count, 0) AS mobile_edits,
    IFNULL(anon_edits.count, 0) AS anon_edits,
    IFNULL(mobile_anon_edits.count, 0) AS mobile_anon_edits,
    IFNULL(mobile_registrations.registrations, 0) AS mobile_registrations
FROM (
    SELECT
        LEFT(rev_timestamp, 8) AS date,
        COUNT(*) AS count
    FROM revision
    INNER JOIN tag_summary ON
        ts_rev_id = rev_id AND
        ts_tags LIKE "%mobile edit%" AND
        ts_tags NOT LIKE "%mobile app edit%"
    WHERE
        rev_timestamp BETWEEN @start AND @end
    GROUP BY date
) AS mobile_edits
LEFT JOIN (
    SELECT
        LEFT(rev_timestamp, 8) AS date,
        COUNT(*) AS count
    FROM revision
    INNER JOIN tag_summary ON
        ts_rev_id = rev_id AND
        ts_tags LIKE "%mobile edit%" AND
        ts_tags NOT LIKE "%mobile app edit%"
    WHERE
        rev_timestamp BETWEEN @start AND @end AND
        rev_user = 0
    GROUP BY date
) AS mobile_anon_edits USING (date)
LEFT JOIN (
    SELECT
        LEFT(rev_timestamp, 8) AS date,
        COUNT(*) AS count
    FROM revision
    WHERE
        rev_timestamp BETWEEN @start AND @end AND
        rev_user = 0
    GROUP BY date
) AS anon_edits USING (date)
LEFT JOIN (
    SELECT
        LEFT(timestamp, 8) AS date,
        COUNT(*) AS registrations
    FROM log.ServerSideAccountCreation_5487345
    WHERE
        wiki = "itwiki" AND
        event_isSelfMade = 1 AND
        event_displayMobile = 1 AND
        timestamp BETWEEN @start AND @end
    GROUP BY date
) AS mobile_registrations USING (date)

Now, to re-run the component extraction. --Halfak (WMF) (talk) 21:40, 31 December 2014 (UTC)[reply]

Anonymous edits. The number of anonymous edits per day is plotted with a weekly seasonal component modeled and the extracted trend overlayed. A vertical line is plotted on Nov. 1st.

Mobile edits. The number of mobile edits per day is plotted with a weekly seasonal component modeled and the extracted trend overlayed. A vertical line is plotted on Nov. 1st.

Mobile anonymous edits. The number of mobile anonymous edits per day is plotted with a weekly seasonal component modeled and the extracted trend overlayed. A vertical line is plotted on Nov. 1st.

Mobile registrations. The number of mobile registration per day is plotted with a weekly seasonal component modeled and the extracted trend overlayed. A vertical line is plotted on Nov. 1st.

So, there's one part of the trend that I didn't expect, but makes sense -- the number of clicks between reading and edit pane was reduced from three to two on Dec 17th. It looks like we see a relatively dramatic increase in mobile anonymous edits at that time. In fact, it doesn't look like we see the characteristic holiday dip there like we do in the rest of the plots!

So, next step is to do a quality analysis of the edits that anons mobile users made before and after the deployment. I'd like to use content persistence strategies to see whether the content that anons add survives scrutiny or not. In order to do that, I'll need to gather a random sample of mobile web edits before and after the stitch was flipped. I'll want to be careful of the Dec. 17th date.

Bah! Gotta run. back to this next time. --Halfak (WMF) (talk) 22:07, 31 December 2014 (UTC)[reply]