= Data & Metrics =
https://sql.telemetry.mozilla.org/dashboard/activity-stream-metrics-summary - requires login
= Engineering =
Activity Stream [status: green]
Highlights
* v1.1.3 deployed Sept 1.
* Issue [https://github.com/mozilla/activity-stream/issues/1145 #1145] initially a chemSpill, but downgraded as root cause is quickly enable and disable of addon could result in crash. This potentially only happens on updates of large profiles. We are using Sentry to monitor it's frequency.
* Risks:
** Improve content relevance through scoring system - we're waiting on final review from PM on going live with an experiement
** Metadata parser - new server instances needs Sec/Ops review
Last Sprint
* [Done] - Share Button with Social API [https://github.com/mozilla/activity-stream/issues/892 #892]
* [Done] - Metadata service internal testing and rating system [https://github.com/mozilla/page-metadata-service/issues/56 #56]
* [Done] - Improve highlight UI and include synced tabs [https://github.com/mozilla/activity-stream/issues/1007 #1007]
* [Done] - Scoring/Ranking highlights landed but pref'd off [https://github.com/mozilla/activity-stream/commit/0cb496c commit]
* [Done] - bug fixes (youtube bgcolor), chores, and performance work
Current Sprint:
* Local page parsing internal testing
* Metadata service experiment [https://github.com/mozilla/activity-stream/issues/1084 #1084]
* Scoring highlights experiment [https://github.com/mozilla/activity-stream/issues/1085 #1085]
* Add Sentry support [https://github.com/mozilla/activity-stream/issues/747 #747]
Changelog: https://github.com/mozilla/activity-stream/blob/master/CHANGELOG.md
Mobile sprint tasks:
https://wiki.mozilla.org/Firefox/Activity_Stream/Mobile_Sprint_Planning
'''[[Firefox/Activity_Stream_Updates | Past Updates]]'''
== Milestones ==
Desktop:
https://github.com/mozilla/activity-stream/milestones
Current mobile team work:
https://wiki.mozilla.org/Firefox/Activity_Stream/Mobile_Sprint_Planning
== Cross Team Projects ==
==== Page MetaData Parsing ====
Problem: embedly proxy server has limitations in performance, client-side intelligence in display heuristics, and has a high cost.
Questions and Goals:
* How do we land and prioritize the work to use Local page meta data parsing?
* What dependencies and risks are there?
** Performance could hang the main thread for ~100ms per page till we get Loci/worker queue
** Quality of the parsing results and image identification could be lower than embedly
** Advanced storage of page indices is needed to improve scoring, de-duplication, and intelligence of displayed content.
=====Working Plan=====
Goals:
* Develop infrastructure to refine rulesets for the Fathom library for Activity Stream
* Experiment, debug, and evaluate client performance/stability of local parsing.
Approach:
* [server] test rulesets and resolve embedly licensing cost.
* [server] work on testing and improving rulesets (with improved feedback from client thumbs up/down)
* [client] Land local parsing pref'd off to evaluate performance for internal testing
* [client] Improve feedback on good/bad highlights to help develop analytics in understand good/bad rulesets.
Task right now:
* <strike>[Jared] email list with instructions on building addon that uses a dev instance of fathom</strike>
* More people test this via setting embedly server endpoint via about:config
* [K8] land PR pref'd off to test Fathom library/rules locally to eval performance and stability
==== Rank/Scoring heuristics for content relevance ====
Problem: current query of places.db has limitations in flexibility and inclusion of other signals.
Solution: How can we build system that could help us learn programmatically what types of content is more engaging for users measured by click-through-rate (CTR)?
Questions and Goals:
* How do we land a scoring system
* How can we maintain low latency of computation?
==== Signals ====
'''Problem''': the heuristics for ranking page value, frecency or scoring, could improve by adding more signals from user activity
'''Solution''': Let's experiment with a basic set of new signals that are already available (e.g. share events, etc...) and add them to a scoring system to test how we can improve relevance to users measured by CTR.
'''Hypothesis''': If we deploy a scoring/ranking system and feed it more signals making score more accurate, we will achieve higher interactions on page.
'''Work to do this:
'''* Andrei and team working on MVP of scoring and ranking system in Activity Stream as an experiment which can be preffed on/off.
* Nan/Olivier working on storage to store signals.
Two groups of signals here:
* A: What user signals can we record now and feed into Activity Stream's scoring system?
* B: What user signals can we enable in gecko and feed into Activity Stream's scoring system for the long term or understanding what signals matter? We could consider uplifting behind a pref, where AS can flip the pref.
'''Actions''':
* Identify what signals are in A vs B - @Brendan can you identify this and update the sheet?
* How do we get A into scoring system or the current query system.
* Do we want to do work in gecko now to enable B signals? If yes, who can do this?
==== Performance ====
Problem: If we parse pages locally, we need to preserve performance which could be impacted if we parse many pages. Storing parsed page meta data would also improve reduce duplicate computations on existing pages.
Solution:
* Loci, a worker queue that could parse pages without blocking the main thread.
* Advanced storage to store page meta data for fast queries by score.
=Team - RACI =
