Wikimedia monthly activities meetings/Quarterly reviews/Analytics/September 2014
The following are notes from the Quarterly Review meeting with the Wikimedia Foundation's Analytics team, September 25, 2014, 9:30AM - 11:00AM PDT.
- Attending
Present (in the office): Erik Möller, Anasuya Sengupta, Rachel diCerbo, Ellery Wulcyzn, Leila Zia, Toby Negrin, Dario Taraborelli, Lila Tretikov, Kevin Leduc, Tilman Bayer (taking minutes), Carolynne Schloeder, Howie Fung, Jeff Gage, Abbey Ripstra, Daisy Chen; participating remotely: Aaron Halfaker, Jonathan Morgan, Oliver Keyes, Dan Andreescu, Andrew Otto, Christian Aistleitner, Nuria Ruiz, Arthur Richards, Erik Zachte
Please keep in mind that these minutes are mostly a rough transcript of what was said at the meeting, rather than a source of authoritative information. Consider referring to the presentation slides, blog posts, press releases and other official material
[slide 2]
Agenda
1. Welcome/Agenda
2. Performance against goals
3. Research and Data (includes Q&A)
4. Development (includes Q&A)
5. Summary/Asks
Toby: Welcome
This has been a very collaboratively written slide deck
[slide 3]
priorities:
standardize baseline metrics
focus on editors this q, then readers
hit most of those goals
on dev side, focused on dashboard and vital signs
[slide 4]
[slide 5]
Research and Data
editDario:
Q1 goals
[slide 10]
Team started in Q1 on new editor metrics
Q2, Q3...
[slide 11]
goal: ship definitions (human readable and machine readable version)
involves identifying bots,....
[slide 12]
alignment with editor model (for Growth and Mobile teams)
began work with Multimedia team, but on hold
[slide 13]
progress towards traffic definitions (pageviews)
both project and article level metrics (community cares a lot about the latter)
Oliver: including mobile views...
(Dario:)
unique client definition (on hold)
Lila: do we have a path to get unblocked?
Dario, Toby: yes
[slide 14]
(Dario:)
Growth research - led by Aaron
anon acq, task suggestions
[slide 15]
did extensive analysis of mobile app traffic
mobile traffic and participation trends, effect of tablets switchover
Lila: in this chart, editor activated on mobile means what?
Dario: registered on mobile device[?], and makes 5 edits within 24h
[slide 16]
Prediction models for survival periods (worked on this earlier this year, still needs final completion)
Lila: (question about time limitation)
Leila: this is from 2001 onwards
Lila: wonder about 5 edit threshold, what about someone who does a 100 edits in a short burst? What was the rationale to focus on this metric (active editors)?
in the end, what matters is content - whether it is being created and kept up to date
Howie: should have a session examining what this metric is good for (and what not) - there has been a lot of research about this in recent years
Lila: also, e.g. Commons will have different usage patterns
Dario: yes, we mostly do these analyses on a per-project basis
Lila: ...
[slide 17]
Dario: also, supported a number of other teams with smaller requests, in consultation mode
This slide is actually not comprehensive, need to get better at capturing everything
Lila: this kind of "response train" is normal, it's not bad that requests are coming
but needs to be managed
Toby: sometimes Oliver and Leila were hit by 3 things at once...
[slide 18]
Dario: Team process:
joint standups with other teams
Suffer from lack of scrum master, project management support
Lila: where are the inefficienceies - when interacting with other teams, or within team?
Dario: both
Leila: btw, there are also community requests, not just other WMF teams
Lila: for capacity planning, it would be more of a project[product ?] management role which requires deeper understanding, scrum master is more for supporting workflow
Erik: this is the kind of thing I have asked Arthur to support/advise on (team practices)
Lila: and sometimes it's ok to share that role within team
Toby: but I need my people (like Dario) focus on their skills
[slide 19]
Dario:
Formal collaborations/hiring
onboarded Ellery as FR analyst
started collaboration with Morten (external researcher) on task recommendations
Aaron: Morten is processing data on task recommendations right now
(Dario:)
collaboration with LANL on anonymizing pageview data - on hold
opened traffic research analyst position - on hold
[slide 20]
Outreach: Wikimania, showcases, ...
[slide 22]
Q2 goals
Lila: coming out of standardization process, what gaps we still have?
Dario: have specifications and raw data, not yet the numbers
[slide 23]
strategic research: close to our strategic priorities
APIs, also for external research
Lila: anonymized?
(Dario:)
This is mostly about data that is public anyway
product research:
discontinue "embedded" model
[slide 23]
strategic research
understanding reader behavior - is becoming a priority
detect knowledge gaps
value-based measures - go beyond mere edit count
microcontributions
uplevel the FR experimental strategy
Anasuya: reader behavior - will this differentiate between mature/ small wikis?
Dario: look for definitions that work across the board
but e.g. Wikidata differs a lot
Anasuya: I'm thinking about differences between Wikipedias, too
Lila: need to wrap up Q1 work
goals on this slide make a lot of sense to me
Dario:
[slide 24]
Data services
A/B testing
Lila: Core team leading this? yes
Public data sources:
this is making already public data more accessible - e.g. revert metadata: would need lots of manual work now
Lab tools (stretch goal)
[slide 25]
Product support
move toward consulting model
should allow team to allocate resources more flexibly
Lila: should require advance notice from Engineering / Product teams
avoid resetting priorities
Erik: already have a cross-team process
Toby: both Dan and Andrew have chaired this
Erik: moving from long-term (yearly) goal-setting to quarterly priority setting, have separate meetings for this
Toby: get into a data/metrics driven mindset
Lila: no project should start without having a way to measure impact
Erik: need to be aware on whether it's possible or not, which data expectation is realistic
it's already the case that all the current priorities have measurable goals
Toby: suggest separate conversation on this
[slide 26]
Dario:
some support for IEG
Outreach:
Leila: Aaron does a lot of this (outreach)
Lila: it's OK for experiments, but does it scale?
Anasuya: also, brings external researchers into contact with our community much faster
[slide 27]
[slide 28]
staff outlook
slide 29-35: appendix (skipped, can take this offline)
Engineering
editKevin:
[slide 39]
main goal was to implement dashboard - accomplished
break down by target site (desktop/mobile/apps) - not reached
blocked on ...
[slide 40]
UV: blocked on tech proposal
[slide 41]
[slide 42]
Vital Signs: Value prop
[slide 43]
demo https://metrics.wmflabs.org/static/public/dash/ , e.g. newly registered on Armenian WP, rolling surviving new editors
Lila: I'm going to play with this... can we note the definitions at the bottom?
Dario: they are on Meta, just need to cross-link
Toby: check out Wikidata, compare to ENWP
this may represent a future trend: a lot more smaller edits (some of them by machines)
(Kevin:)
[slide 44]
Vital Signs: backlog
lots of feedback coming in
[slide 45]
Wikimetrics:
builds the data that Vital Signs uses
value prop:...
[slide 46]
Wikimetrics accomplishments: ...
a lot more robust than cronjobs
Lila: allows to identify fast growing/slow growing projects? yes
(Kevin:)
[slide 47]
backlog:
new hire will take on backlog for Grantmaking
extend to CentralAuth (users with edits on several wikis)
Lila: so that's a filter by user name (instead of project)? yes
Kevin:
tagging cohorts - classify
Lila: tag reports too?
Kevin: yes, we could do that
Lila: just want to make sure we do this in a general way
Kevin: add ability to remove users from cohort for privacy reasons
Lila: can we also integrate usage (traffic) data?
Toby: Wikimetrics is not primarily about that
friction is between more functionality desired by Grantmaking and...
Anasuya: content metrics are really important for us too, but that's separate
Toby: this is pretty general (using off-the shelf components, e.g. Hadoop),
might be able to quickly plug it into other databases as well, e.g. pageviews
Dan: ...
also interesting: Yuvi's Quarry (allows db queries via web interface on Labs)
Oliver (on chat): +1. Quarry is awesome.
Jonathan (on chat): incredibly useful
[slide 48]
Kevin:
Eventlogging
publicly documented schemas
[slide 49]
Backlog
maintenance, sanitize data
[slide 50]
Refinery
data processing ecosystem, addresses dissatisfaction with data quality
[slide 51]
accomplishments: operational since July
a lot of cooperation with Ops team
turned off legacy data feeds (after contacting people) - no complaints so far
port Webstatscollector to Hadoop
significantly reduced error rate (from 3-6%)
Toby: decided to have Christian work very closely with Ops team on this
good partnership
[slide 52]
backlog
compile PVs per page and make data public
- community has been asking about this for years
Erik: we have the historical PV data, it has been used for many research projects
Oliver: We can backfill for 1.5 years, using the new definition and the sampled logs, but it'll be a bit inaccurate, because..sampled logs.
[slide 54]
Team
Lila: squares on the bottom mean what?
Kevin: blue = addition (e.g. Q4 : myself), red = losses
[slide 55]
Process & Metrics
Lila: using point system?
Kevin: yes, Fibonacci
defined 34 points = one developer with little or no distraction during the sprint
7/24 outlier = Dan and Nuria on vacation
metrics show team was more focused in later sprints
Toby: visualization helped team focus
[slide 57]
preliminary Q2 goals
[slide 58]
Out of scope:
need to put a brake on Vital Signs
Lila: tradeoffs?
Dario: ...
Lila: can we break this down, make a list
Toby: did a back-of the envelope velocity calculation, saw we could not do all of them
Lila: just make sure you get the things you commit to done, rest as stretch goals
[slide 59]
Toby: Visualization
Limn tried to do too much
Vital Signs as basis to make this available to community
ErikM: using "Vega" open source library
Dan: ...
Lila: managing another tool is also work...
Toby: ...
Erik: good to do a spike on it, if it's not worthwhile, we will find out quickly
Lila: I have worked myself with Saiku and Tableau[?]
do what you think works, be aware of "human" (work) cost
Toby: should be easy to put contractor on this
Kevin:
(skip "Unique Visitors: Proposed Milestones" slide)
Asks
edit
[slide 64]
Toby:
continue with space and support - keep developers focused
recruit / staffing: need to tweak job description to attract right people
promote data informed perspective