We expect from historical stats that a significant amount (double digit %) of fulltext search abandonment is due to those queries receiving zero results, in which case there is nothing to click through to. Adjust the data aggregation and dashboards to show the % of fulltext search with zero results, and the % of fulltext search that receivied results but did not click through. In theory these two should add up to the current abandonment rate %.
Description
Description
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | None | T358345 [Epic] Search metrics 2024 | |||
| Resolved | EBernhardson | T370290 Split abandonment and zero-results in the search superset dashboard |
Event Timeline
Comment Actions
ebernhardson opened https://gitlab.wikimedia.org/repos/search-platform/discolytics/-/merge_requests/44
sat metrics: Track sessions that saw no results
Comment Actions
dcausse merged https://gitlab.wikimedia.org/repos/search-platform/discolytics/-/merge_requests/44
sat metrics: Track sessions that saw no results
Comment Actions
Patches have been deployed, since these only take a couple minutes to run (searchsatisfaction is orders of magnitude smaller than webrequests) i cleared all tasks from June 1st forward and it is recalculating those metrics. Next step is to update the superset dashboard to split the numbers
Comment Actions
@dr0ptp4kt I tried to update the dashboard, but for some reason superset is only allowing me to edit the charts. It's not clear why, both of us are listed as owners in the dashboard properties. Regardless, the following are the changes i was going to apply, will try again later or maybe you can see if it lets you perform the edits (also, copy-editing is appreciated):
Abandonment graph title
Add word sessions for clarity that these are per-session metrics and not per-query:
Percentage of fulltext search sessions resulting in SERP abandonment (desktop)
Text below abandonment graph
Update % to match the june 1st start date of new metrics, and add a new line under Keep in mind that explains the separation of the two metrics.
When users on desktop reach a SERP, they may abandon their information seeking. There are several reasons why this can be the case, including: - Their search terms did not have any results, did not have matching results, or the results were insufficient. This may be because of a lack of pages with the search terms, or it could have to do with the words used or their spelling. - The SERP results may have actually satisfied the user's need. We don't attempt to interpret these sorts of interactions. As of 1-June-2024, we see that the abandonment rate hovers around 48%, and sessions that receive zero results hovers around 7%. Although this level of abandonment may be reflective of available content, there is likely room for improvement, and future work is likely to benefit SERP relevancy and sufficiency. Keep in mind: - Whereas desktop fulltext SERP instrumentation is fairly robust and so is included here, there has been some drift with mobile web instrumentation and it is not included here. - Abandonment and Zero results are separate here, as it's reasonably clear why zero result sessions are abandoned. Total abandonment is the sum of both metrics.
Comment Actions
Dashboards have been updated. After some review I found that the problem with saving was due to emoji's in the content. It doesn't seem to be browser specific, tried updating from firefox and chrome. Had to remove all the emoji's to get it to save.
For reference these were the parts that I removed emoji's from:
🔍 Autocomplete 📋 🔍 Aggregate search traffic 📇 Related Articles (mobile web) - All the types of reaching results via user-supplied search are clubbed together. For autocomplete-specific results see the **🔍 Autocomplete** section. - Users on desktop are more likely to type full words that resolve to article titles (or redirect titles). When they do this they can press ENTER or click the Search button, and they will be taken directly to the appropriate page. Such page accesses are not included in autocomplete numbers. Rather, they are incorporated into the **📋🔍 Aggregate search traffic** section, which looks at aggregate behavior for searches. On the other hand, a user may also click or tap on a title match from the autocomplete results, which would be included in autocomplete numbers.