Sefer Tasdemir

← All projects

IMDb

A complete decision-support solution on IMDb data: Python ETL, OLAP warehouse, Power BI and sentiment analysis.

  • Python
  • SQL Server
  • SSIS
  • Power BI
  • DAX
  • scikit-learn
  • PowerAMC

View the GitHub repository

Context

SAE5.C.01 of the BUT in Computer Science (2025-26, IUT d’Annecy). Starting from IMDb’s public flat files, the brief was to design an optimized SQL Server database and Power BI reports, then AI models able to assess a review from its text alone, all on a standard PC.

Team and my role

Group project, 5 people.

I took the lead on the project and handled the technical side.

Approach

  • Modeling: conceptual, logical and physical models in PowerAMC for a normalized, constrained OLTP schema.
  • ETL: multithreaded Python script with manual memory management and disk offloading to fit on a standard PC, followed by bulk loading with BCP.
  • Optimization: indexes on foreign keys and composite indexes, backed by execution plan analysis.
  • Decision support: move to a star schema loaded with SSIS, then Power BI reports.
  • AI: TF-IDF and linear models compared with 3-fold cross-validation, using a restricted hyperparameter grid to stay frugal.

Visuals

  • OLTP relational schema with the TITLE, PERSON and TITLE_PRINCIPAL tables and their reference tables
    Normalized OLTP model with 12 tables.
  • OLAP star schema with the T_FACT_MOVIE_STATS and T_FACT_CASTING fact tables and their dimensions
    Star schema with 2 fact tables, 9 dimensions and a bridge table for genres.
  • Data model view in Power BI, showing relationships between facts and dimensions
    Semantic model in Power BI.
  • General Statistics page of the Power BI report, with KPIs, a historical curve and a world map
    Catalog overview.
  • World map with a tooltip showing the five most popular genres of a country
    Tooltip on hover showing a country’s leading genres.
  • Statistics of movies page with a key influencers visual, a ranking and a scatter plot
    Movie analysis, with the Key Influencers AI visual.
  • Casting page with search filters and rankings of actors and directors
    Casting analysis.
  • TV Shows vs Movies page with histograms, stacked areas and a Sankey diagram
    TV shows and movies compared, down to a Sankey diagram.
  • Focus on Video Game page with engagement by genre between Europe and North America
    Focus on video games.
  • Confusion matrix of the binary model, with a strongly dominant diagonal
    Model 1, positive or negative polarity.
  • Scatter plot comparing actual and predicted ratings, with the perfect prediction line
    Model 2, actual and predicted rating out of 10.
  • Confusion matrix of the grade model, across four classes
    Model 3, grade classification.

Facts and figures

7 TSV input files
Over 12 GB
processed by the ETL in 25 minutes
222M rows
OLTP and OLAP models
12 + 12 tables
join query after indexing
62 to 7 ms
polarity accuracy on 25,000 test reviews
90.2%
on the rating out of 10, with an R² of 0.657
MAE of 1.61
grade accuracy, with the Mixed class assigned by a rule
65%
OLAP Power BI report and its tooltips
5 + 3 pages

What I took away

The main challenges and lessons came from managing the project and from setting up the initial requirements.