Opencast Analytics @ DCE

Jay Luker - Senior Software Engineer, Harvard DCE

Using Opencast API data to...

• track workflow performance
• understand workflow performance
• (nearly) real-time user activity monitoring
• provide user insight to instructors & researchers (future)

Workflow Browser

• d3-based workflow visualization
• Each workflow rendered as a timeline showing duration of individual operations
• Pan, zoom and filter
• Mouse-over operations to display details
• Goal was to allow visual id of problem workflows and/or operations

WPI Dashboard

• Workflow Performance Index
• WPI = media length divided by workflow run time
• run time calculation excludes operations not related to actual video production

• Time-series graphs of active user sessions
• Filter by live vs on demand, identified vs anonymous
• Top terms: course, IP address, browser, device, OS
• User location mapping based on geoip lookups

User Activity

Dashboard

Usertracking Event Pipeline

Paella

• Standard event types: PLAY, PAUSE, SEEK, etc
• HEARTBEAT events every 30s
• Custom param: playing=<true|false>
• Additional interesting event types:
  • setVolume
  • enterFullscreen
  • setplaybackrate
  • caption:enabled

Usertracking Service

• Usertracking service adds user id based on auth
• Current DB size: ~73G
• Mostly mh_user_action: 131m records
• This is a problem
• We are exploring solutions

Harvester

• Runs */2 minutes
• Fetches actions in batches using start/end params
• Builds event record from action +  episode metadata
• Publishes to SQS

ELK Elastic Stack

• Logstash - multi-purpose data pipeline component
  • geoip lookups
  • browser useragent parsing
• Elasticsearch - data store for search & analytics
  • automated snapshots to s3
  • indexes older than x days: archive & delete (future)
• Kibana - analytics & visualization platform
  • internal dashboards
  • embed graphs in LMS course pages (future)

{
  "mpid": "f21fcc8f-93...",
  "ip": "12.34.56.78",
  "huid": "12346789",
  "is_live": 0,
  "ua": {...},
  "session_id": "15fe8z...",
  "action_id": 72922555,
  "@version": "1",
  "@timestamp": "2016-01-01...",
  "episode": {...},
  "host": "127.0.0.1",
  "geoip": {...},
  "ua": {...}
}

{
  "inpoint": 4790,
  "type": "HEARTBEAT",
  "is_playing": true
}

{
  "series": "20160114328",
  "course": "Data Science",
  "type": "L10",
  "title": "Lecture 10",
  "year": "2016",
  "term": "01",
  "cdn": "14328",
  "duration": 5978537,
  "start": "2017-04-03T21:29:00Z"
}

{
  "ip": "12.34.56.78",
  "country_code2": "SA",
  "country_code3": "SAU",
  "country_name": "Saudi Arabia",
  "continent_code": "AS",
  "timezone": "Asia/Riyadh",
  "location": [45,25]
}

{
  "name": "Chrome",
  "os": "Windows 7",
  "os_name": "Windows 7",
  "device": "Other",
  "major": "47",
  "minor": "0",
  "patch": "2526"
}

Next Steps (in no particular order)

• Convert Workflow Browser & WPI Dashboard to new Elastic Stack deployment
• Query API for user event data
• Embeddable visualizations
• Separate usertracking collection from engage

Thanks!

Here's some links!

• These slides: http://bit.ly/dce-analytics
• DCE @ Github: https://github.com/harvard-dce
  • mh-opsworks-recipes (jluker-elk branch)
  • mh-user-action-harvester
  • dce-paella-extensions
  • mh-workflowbrowser
• Elastic Stack: https://www.elastic.co/