Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Improving sticker keyboard with data analysis

Improving sticker keyboard with data analysis

Eebedc2ee7ff95ffb9d9102c6d4a065c?s=128

LINE DevDay 2020

November 27, 2020
Tweet

Transcript

  1. None
  2. Self Introduction › Joined LINE Corporation as a Data Scientist

    in May 2019 › Working on data analysis for LINE Sticker Service › Naho Tanigawa
  3. Agenda › Sticker Keyboard Project › Challenging Task 1 :

    Definition of Session › Challenging Task 2 : Identification of KPI › Case Studies of Analysis
  4. Sticker Keyboard Project < Goal > Increase the frequency of

    sending stickers < Approach > Reduce time to search and send stickers Sticker Sticker Keyboard
  5. Data-Driven Improvement User Activity logs Analysis Collecting Data feedback

  6. Data-Driven Improvement User Activity logs Analysis Collecting Data feedback We

    have no data of sent contents nor user identity.
  7. Sticker Keyboard Update new functions released in June 2020

  8. Sticker Keyboard Update new functions released in June 2020 New

    panel for searching stickers Notification of time-limited sticker’s expiration
  9. 2 Challenging Tasks Definition of Session Identification of KPI

  10. Definition of Session

  11. Problem Logs of 1 LINE App. Launch = sending stickers

    E Need to group sequences for understanding user behavior 1 LINE App launch: Too long 1 action on Sticker keyboard: Too short
  12. 1 Session on sending a sticker Problem Logs of 1

    LINE App. Launch = sending stickers E E 1st session 2nd session 3rd session E E Need to group sequences for understanding user behavior 1 LINE App launch: Too long 1 action on Sticker keyboard: Too short
  13. Solution Define session by splitting log based on rules S

    Action to start session Action to end session (Sending stickers) E Click E S 1st session 2nd session 3rd session Logs of 1 LINE App. Launch E S Click We have no data of sent contents nor user identity.
  14. Solution User Activity logs Randomly sample the logs of 1%

    of all users Splitting logs by rule-based Estimated Statistics By × 100 Action to start session Action to end session (sending stickers) E Pipeline of processing session logs Session Logs S 1st session 2nd session 3rd session Logs of 1 LINE App. Launch E E S S 10 billion in a day
  15. Identification of KPI

  16. Problem Identify KPI to represent both Goal & Approach <

    Goal > Increase the frequency of sending stickers < Approach > Reduce time to search and send stickers
  17. Solution # of Sending Sticker Count # of Unique Sending

    Users Average Sending Sticker Count per Users Goal
  18. Solution # of Sending Sticker Count # of Unique Sending

    Users Average Sending Sticker Count per Users Average Sending Sticker Count in 1 App Launch per Users Average App Launch Count per Users Goal
  19. Solution Sending Sticker Count in 1 App Launch per Users

     2 # of Sending Sticker Count = sending stickers E # of Unique Sending Users Average Sending Sticker Count per Users Average Sending Sticker Count in 1 App Launch per Users Average App Launch Count per Users Goal Approach E S 1st session 2nd session 3rd session Logs of 1 LINE App. Launch E S
  20. Solution ≒ Elapsed Time of 1 App Launch Elapsed Time

    of Sending 1 sticker Sending Sticker Count in 1 App Launch per Users  2 # of Sending Sticker Count = sending stickers E # of Unique Sending Users Average Sending Sticker Count per Users Average Sending Sticker Count in 1 App Launch per Users Average App Launch Count per Users Goal Approach E S 1st session 2nd session 3rd session Logs of 1 LINE App. Launch E S
  21. Case Studies of Analysis

  22. Case 1 : UI of New sticker panel Reduce time

    to send stickers by allowing users to do the most common way TO-BE Input method in new sticker panel: double-click a sticker … AS-IS Most frequent sending pattern: double-click a sticker 26%
  23. Case 2 : Notification Timing of time- limited sticker’s expiration

    expiration Download date Reduce time to search stickers by highlighting stickers near expiration AS-IS Last use date of time-limited stickers: Increase in 10 days before expiration TO-BE Notification timing: 10 days before expiration
  24. Data-Driven Improvement User Activity logs Analysis Collecting Data feedback

  25. Thank you