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

KPI Visualization, Prediction for LINE Fukuoka Service Ops

KPI Visualization, Prediction for LINE Fukuoka Service Ops

Kazuhiro Maeda
LINE Fukuoka DataLabs Data Analysis Team Data Scientist
https://linedevday.linecorp.com/2020/ja/sessions/0556
https://linedevday.linecorp.com/2020/en/sessions/0556

LINE DevDay 2020

November 27, 2020
Tweet

More Decks by LINE DevDay 2020

Other Decks in Technology

Transcript

  1. About me
    Self introduction
    › Analysis to drive service/project growth.
    › Analysis to promote business improvement within LINE Fukuoka Corp.
    › Manage Analytical Projects / Products.
    My Roles
    Skills
    › Data analysis processing using the R language in general.
    MAEDA Kazuhiro
    › Develop center – DataLabs – Data Analysis Team
    › Team Manager, Data Scientist

    View full-size slide

  2. Agenda
    › Introduction of LINE Fukuoka Corporation
    › Issues in Operation Tasks of LINE Fukuoka
    › VIIM Project
    › Centralized Prediction Project
    › In future direction

    View full-size slide

  3. Introduction of
    LINE Fukuoka Corporation

    View full-size slide

  4. The 4 Roles of LINE Fukuoka
    Our Business
    Business Planning
    Engineering
    Creative Works
    Operations

    View full-size slide

  5. Issues in Operation Tasks of
    LINE Fukuoka
    for ”always data driven”

    View full-size slide

  6. for “always data driven”
    Three basic phases
    the data
    Collect
    the data
    Apply
    the data
    Analyse

    View full-size slide

  7. for “always data driven”
    Three basic phases
    Collect
    Consolidate and manage
    relevant data in one place
    › Automatically collect
    diverse data
    › convert to an easy-to-
    use format for analysis
    Analyse
    Analyze aggregated
    data
    › Calculate metrics to
    deal with
    › Visualization
    › Extract issues
    etc…
    Apply
    Implement Analysis
    Results
    › Planning of business
    plans based on time-
    series forecasts
    › Apply machine learning
    etc…

    View full-size slide

  8. The three NOs we had
    Issues that Prevent Data Driven in LINE Fukuoka
    NO integrated data collection for operations
    NO metric measurement and/or visualization framework
    NO cross-organizational projects in data application

    View full-size slide

  9. From NO to BUILD
    Three approaches for data driven
    BUILD integrated data collection for operations
    BUILD metric measurement and/or visualization framework
    BUILD cross-organizational projects in data application

    View full-size slide

  10. VIIM Project
    VIsualization – Improvement - Management

    View full-size slide

  11. Purposes of this project
    › Consolidate and visualize data that
    management and field members
    can use to make decisions
    › Based on visualized data, the
    environment/system for
    improvement measures and future
    directions is constructed.

    View full-size slide

  12. Cross-Organizational Project Structure
    › PM/PdM, requirements definition
    › dashboard construction
    Digital Planning Team
    › application development
    › construction of data acquisition systems
    Global Operation Team
    › overall system construction
    › in-house IT support
    Work Improvement Team
    › data pipeline construction
    › dashboard construction
    Data Analysis Team

    View full-size slide

  13. Architecture of this system
    Referred
    DB
    other
    Products
    Data Lake Data Ware House
    CMS
    CMS
    Log data
    DB
    Files Storages
    LINE Fukuoka DB

    View full-size slide

  14. Dashboard Visualization

    View full-size slide

  15. Achievements
    Consolidate data from many departments into one place
    Build a visualization platform for KPI metrics
    Reduce field operational costs

    View full-size slide

  16. Time series forecast of workload
    Centralized Prediction
    Project

    View full-size slide

  17. Purposes of this project
    › Established a framework for
    predicting workload that affects
    the operation
    › Delivering more accurate
    predictions using data science
    techniques
    Centralized
    Prediction
    Project

    View full-size slide

  18. Dashboard
    Architecture of this system - batch
    Input data
    storage
    manipulating
    data
    training, predicting
    Raw data
    Events
    model: DeepAR, prophet,
    etc…
    predicted
    result
    Dashboard

    View full-size slide

  19. Online time series prediction tool

    View full-size slide

  20. Achievements
    Provide accurate predictions
    Flexible design for predictive systems
    Both online and batch systems are available

    View full-size slide

  21. In future…

    View full-size slide

  22. Toward true data driven
    Three approaches for data driven
    Building a more efficient data management and operational structure
    Building an integrated framework for time series forecasting
    Streaming other operations using machine learning

    View full-size slide