Ticket #1736 (closed New feature: fixed)
Segmentation
| Reported by: | vipsoft | Owned by: | matt |
|---|---|---|---|
| Priority: | critical | Milestone: | Piwik 1.2 |
| Component: | Core | Keywords: | |
| Cc: | Sensitive: | no |
Description (last modified by matt) (diff)
The goal is to implement segmentation in Piwik. First a simpler version allowing pre-chosen segments and reports, then evolve into a more open segmentation model similar to what GA offers.
Segmentation: API
All get* functions returning analytics reports will get a new parameter $segment. This parameter, by default empty, means we return data for all visitors (current behavior).
When set eg. $segment = "country==fr" the report returned is segmented to visits having this custom value.
We must define and document a list of available dimensions to query, eg.:
- country (match to location_country)
- keyword (match to referer_keyword)
- customVarName1 (match to custom_name_1 see #1984 )
- etc.
Later we can imagine extending the syntax to support AND and OR eg. $segment = "customName1==loggedIn,customValue1=yes" or even support 'contain', 'is not' operators, etc.
Segmentation: Archiving
Archiving reports for segments mean: filter these segments, then aggregate.
Queries doing the Piwik archiving have to be updated WHERE ....
AND segment1 = $value
GROUP BY label, segment1
For example segment= country==fr, the query for best browsers would become select count ... where ... and location_country=fr ..... group by visitor_browser, location_country
Archives stored must also somehow store the segment query as well as site,date1,date2: do we need a new key in the archive tables? or probably this can be encoded in the archive.name attribute somehow (one more reason we need to keep CONCAT(custom names + values) small in length)
Segmentation: Known segments control list
To keep things fast for 'known segments', we can allow user to create a list of segments he is interested in.
User would also provide a list of reports to pre-process, rather than pre-processing all reports.
For example: 'keyword==Piwik' // known valid segment
=> array('Referer.getKeywords','Page.getPageUrls', .. ) // reports forced to be pre-processed
In this case Piwik will, during archiving, only process these reports and not more. If later the user would want to access more reports, and if logs are still available, he could change the list which would affect reports going forward.
In V1, if a requested segment/report pair wasn't pre-processed, archiving will return no data.
New setting
- Enable segmentation from the API (and UI) since this could impact performance for anonymous allowed piwik (eg. piwik demo). Disabled by default.
- Enable segmentation preprocessed reports control list (see above).
- Update the code to handle multiple segmentation constraints AND and OR, which would allow any kind of segmentation based on visit attributes
Performance Impact
Possibly huge depending on data set. Real time performance likely to be slow with mysql, so pre-processing reports highly recommended (defining a list of known reports, see above)
Feature requests See other ticket #2092
Please submit any feedback/question.
