01 · DATA ANALYSIS & DASHBOARD
Real data.
Prepared with discipline.
Data analysis, cleaning, preparation, dashboarding, and integrity validation for the STRONGR OS recommendation-intelligence workflow.
WHY MIND-SMALL
Public data for real recommendation mechanics.
MIND is a public Microsoft News recommendation dataset based on genuine user-content interaction logs. This capstone uses official MIND-small train/dev data to validate recommendation mechanics without claiming that the model already recommends Strongr Daily content.
Its official chronology supports a real held-out development evaluation rather than a random split.
DATA SCALE
| Table | Rows | Purpose |
|---|---|---|
| BehaviorSummary | 230,117 | Interaction and history summary |
| NewsContent | 65,238 | Content and metadata context |
| CategoryEngagementSummary | 417 | Category/subcategory engagement view |
| ContextSummary | 79 | Time and candidate-set context |
| DataQualitySummary | 15 | Integrity and readiness checks |
DATA PREPARATION
Preserve the source.
Derive the context.
Candidate-level records
Original TSVs were preserved; candidate-level rows were created, metadata joined, and historical context derived.
Leakage controls
Raw IDs and candidate position were excluded. Train/dev chronology was preserved, with scalers and encoders fit on training only.
Explicit missingness
Missing histories and abstracts are represented explicitly rather than silently treated as ordinary content.
DATA QUALITY
Key integrity checks: 0 duplicate impression IDs · 0 duplicate news IDs · 0 malformed candidate tokens.
KEY DASHBOARD INSIGHTS
Rare positive engagement
Positive engagement is about 4%, making accuracy alone misleading.
Stable held-out prevalence
Train and development prevalence are closely aligned.
Context matters
Candidate-set size, history, affinity, category/subcategory, and time are important modeling contexts.
Preparation passed
Final preparation passed metadata and integrity validation.
POWER BI EVIDENCE
Power BI was used to validate data scale, engagement patterns, recommendation context and model readiness. These presentation cards reproduce the accepted analytical evidence from the validated MIND-small reporting tables.
Scale of the prepared recommendation dataset.
Stable train/dev engagement prevalence.
Candidate-set context observed in Power BI.
Interaction-history context observed in Power BI.
Content-category engagement comparison.
Validated data-integrity checks.
Original Power BI project retained as the technical source of record.