Page MenuHomePhabricator

Property searches in wikidatacompletionsearchclicks have mostly null values
Closed, ResolvedPublic

Description

Most of the autocomplete data is coming back looking pretty good, but property search has a significant number of events with mostly null values. Click Index varies, which suggests users are choosing an item from some list, but whatever that list and wherever it came from is not recorded in the schema.

Figure out where these are coming from and fix up the data collection so that we don't have all these null values, and so we know where the data is coming from.

Click logs originally requested in T196186.

Example bad records:

{"clickIndex":5,"clickPage":null,"context":"property","language":"gl","searchResults":"","searchTerm":""}
{"clickIndex":2,"clickPage":null,"context":"property","language":"ar","searchResults":"","searchTerm":""}
{"clickIndex":7,"clickPage":null,"context":"property","language":"en","searchResults":"","searchTerm":""}

Event Timeline

The empty items seem to be from property suggestions mechanism. The property suggestions API offers some properties even before the user typed anything. We should ignore such clicks.

Change 463181 had a related patch set uploaded (by Smalyshev; owner: Smalyshev):
[mediawiki/extensions/WikimediaEvents@master] Ignore clicks with empty search string

https://gerrit.wikimedia.org/r/463181

Smalyshev triaged this task as Medium priority.Sep 27 2018, 3:01 AM
Smalyshev moved this task from Incoming to Needs review on the Discovery-Search (Current work) board.
Smalyshev moved this task from Next to In review on the User-Smalyshev board.

Change 463181 merged by jenkins-bot:
[mediawiki/extensions/WikimediaEvents@master] Ignore clicks with empty search string

https://gerrit.wikimedia.org/r/463181

We should ignore such clicks.

Also, is it providing value to users? Or does it seem like a bug, that users might click out of confusion? Might be worth a follow up task to potentially remove in general.

Also, is it providing value to users?

I assume yes. It basically suggests properties that have been statistically frequently used on entities of this kind. It's useful but it's not search so we should not collect it as part of the search data.

Change 463296 had a related patch set uploaded (by Smalyshev; owner: Smalyshev):
[mediawiki/extensions/WikimediaEvents@wmf/1.32.0-wmf.23] Ignore clicks with empty search string

https://gerrit.wikimedia.org/r/463296

Change 463296 merged by jenkins-bot:
[mediawiki/extensions/WikimediaEvents@wmf/1.32.0-wmf.23] Ignore clicks with empty search string

https://gerrit.wikimedia.org/r/463296

Mentioned in SAL (#wikimedia-operations) [2018-09-27T18:49:21Z] <thcipriani@deploy1001> Synchronized php-1.32.0-wmf.23/extensions/WikimediaEvents/modules/wikibase/ext.wikimediaEvents.completionClicks.js: SWAT: [[gerrit:463296|Ignore clicks with empty search string]] T205301 (duration: 00m 56s)

Also, is it providing value to users?

I assume yes. It basically suggests properties that have been statistically frequently used on entities of this kind. It's useful but it's not search so we should not collect it as part of the search data.

Ah, I didn't know it was based on statistics (thought maybe it was an empty string prefix index). Awesome :)

Smalyshev claimed this task.