THE FINALS - Review Analytics Dashboard
One review can appear in multiple categories. Totals may exceed review count.
Top Negative Reasons
Top Positive Reasons
Review Volume Over Time
Player count overlay from SteamCharts (static snapshot, not live data).
Season Health Barometer
Season 0 (Beta) is not shown — the game was in closed beta (Mar–Nov 2023). Off-season = reviews posted in gaps between seasons.
Top Complaints
Top Praise
Recurring Problems
Complaints that persist across multiple seasons — not a one-time spike but a systemic issue.
Player Journey — How Sentiment Changes With Playtime
Negative Categories by Playtime
Color intensity shows how each category's share changes across playtime brackets (normalized per row). Numbers are actual % of reviews in that bracket.
Positive Categories by Playtime
Sentiment by Season & Playtime Cohort
Approval rate (%) for each season × playtime combination. Reveals whether newcomers vs veterans experience the game differently in each season.
Category Deep-Dive
Review Bombing Detection
Days and weeks with unusual spikes in negative reviews. Orange zones highlight periods where the rolling negative ratio exceeds the overall average + 1.5 standard deviations. Dashed orange lines mark patch dates.
Daily Negative Ratio
Worst Days (highest negative ratio, min 20 reviews)
Language Analysis
Grouped by review language, not player region. A Russian-speaking player may be in EU, a Portuguese speaker in US, etc.
Top Complaints
Top Praise
Deviation from Global Average
How this region's complaint distribution differs from the overall average. Only significant deviations (>3%) are shown.
Same Topic — Praised vs Complained
The same game aspects get both love and hate. We paired related negative and positive categories to show the contrast. Sorted by total combined mentions — biggest topics first.
Approval by Playtime Cohort Across Seasons
How each playtime bracket's sentiment evolves season-to-season.
Top Reviews
Community Insights
Specific issues extracted from reviews using AI (Gemini 2.5 Flash Lite). Unlike broad categories, these are granular — exact complaints, suggestions, and praise points.
Rankings are by mention count, not importance. A viral controversy may inflate some issues. Consider using season/playtime filters to control for this.
Top Pain Points
Top Requests
What Players Love
Entity Tracker
Track how specific weapons, gadgets, abilities, and game modes are discussed across seasons. Mentions are extracted from reviews by AI. Click any bar to view details.
Most Discussed Entities
Patch Notes & Community Reaction
Patch notes parsed from THE FINALS Wiki. See what was changed and how complaints shifted in response.
Entity Timeline
Season Patch Details
Category Cloud
Visual overview of all review categories. Size = frequency, color = sentiment (red = complaint, green = praise).
Show all categories as table
Review Explorer
Browse real player reviews by category and specific issue. Click any bar on other pages to jump here.
How samples are selected
Methodology
Full transparency on how this dashboard processes and classifies review data.
Analysis Pipeline
A two-stage AI pipeline processes every review:
- Stage 1 — Category Classification: Each review is classified into 42 categories (30 negative, 12 positive) using AI. Covers all 29 languages. 197,004 reviews (80%) assigned ≥1 category; remaining 20% are short, empty, or off-topic reviews with no classifiable content. Cost: $3.30
- Stage 2 — Issue Extraction: Specific complaints, suggestions, and praise extracted from each review with game entity normalization. 440,481 issues extracted (avg 1.78 per review). Cost: $6.00
- Post-processing: Issue texts normalized to merge duplicates (prefix/suffix synonym groups, context-dependent mapping). Patch notes parsed for correlation.
Total AI cost: $9.30 for all 247K reviews.
AI Model & Parameters
- Model:
google/gemini-2.5-flash-lite(temperature 0.1) - Stage 1 (Classification): Separate prompts for positive and negative reviews (sentiment-locked). Each review classified into 1+ of 42 categories. Batch size: 200. Low-confidence: ~29%.
- Stage 2 (Issue Extraction): Each review analyzed for specific complaints, suggestions, and praise. Few-shot examples provided. Entity normalization against canonical game data. Batch size: 80.
- API: OpenAI-compatible endpoint at PayPerQ ($0.07/M input, $0.28/M output). Total cost: $9.30.
- Post-processing: Synonym merging (prefix/suffix groups), context-dependent normalization. Source code available on GitHub.
Confidence & Coverage
Limitations & Caveats
- Survivorship bias: Reviews represent ~0.3% of the player base. The silent majority may have entirely different priorities.
- Volume ≠ importance: Rankings are by mention count. A viral controversy can inflate specific issues disproportionately. Use season/playtime filters to control for this.
- Correlation ≠ causation: Complaint changes after patches may reflect seasonal trends, player count shifts, or meta changes rather than patch effects.
- Language ≠ region: The Language Analysis page groups by review language, not player location. A Russian-speaking player may be in EU.
- AI is imperfect: Sarcasm, context, and nuance can be missed. There is no manual verification of individual classifications.
- ~20% uncategorized: Short, empty, or off-topic reviews yielded no category assignments. An additional ~0.7% had no AI response at all.
- Issue extraction coverage: 86% of reviews yielded extractable issues.
- Newcomer bias: Reviews from 0–10h players form the largest bracket but tend to be shorter and less informative. Rankings on the Overview page default to 100h+ players to surface experienced player perspective. Use the playtime filter to see all brackets.
- Season assignment uses review creation date, not update date.
- Entity normalization may miss unusual spellings or slang. The canonical entity list includes aliases in 7 languages (English, Russian, Chinese, Portuguese, German, Spanish, French, Turkish), but coverage varies by language.
About This Dashboard
What is this?
An independent, community-facing analytics dashboard built from all Steam reviews for THE FINALS — Embark Studios' free-to-play FPS.
Data Source
Game Data Credits
Game entity data (weapons, abilities, gadgets, maps) and patch notes sourced from THE FINALS Wiki. Thank you to all wiki contributors!
Tech Stack
- Data: Python, Steam API
- AI: Gemini 2.5 Flash Lite via PayPerQ API ($9.30 total)
- Dashboard: Static HTML + Plotly.js
- Hosting: GitHub Pages (CDN)
For detailed classification methodology, see Methodology.
Author
[VSRAT] ANDRUHA
Buy Me a Coffee if you found this useful!
Community Streamers
Check out these cool streamers: