A data-driven analysis of the Home Assistant YouTube community through comment network analysis, sentiment evaluation, and community structure detection on monthly release party live streams.
This project analyzes user engagement patterns and community dynamics within the Home Assistant YouTube ecosystem by examining comments from monthly release party live streams. The analysis applies social network analysis, natural language processing, and community detection algorithms to understand user behavior, sentiment trends, and network structure.
The analysis uses the YouTube Data API v3 to collect:
- All Home Assistant Release Party videos from the official channel
- Complete comment threads with metadata (author, timestamp, text, reply structure)
- Video metadata (publication dates, view counts, engagement metrics)
A user co-occurrence network is built where:
- Nodes represent individual commenters
- Edges connect users who commented on the same video
- Edge weights reflect the number of shared videos
- Degree centrality: Identifies most connected users
- Betweenness centrality: Detects bridge users connecting communities
- Eigenvector centrality: Highlights users connected to influential peers
- Clustering coefficient: Measures local community cohesion
- Network density: Evaluates overall connectivity
- Assortativity: Tests for homophily by sentiment and comment type
Three algorithms are applied to identify sub-communities:
- Louvain: Modularity optimization with coarse granularity
- Leiden: Improved Louvain variant with better partition quality
- Infomap: Information-theoretic approach yielding fine-grained communities
TextBlob polarity scoring (-1 to 1 scale) evaluates comment sentiment:
- Per-video aggregate sentiment
- Per-user average sentiment
- Temporal sentiment trends
- Correlation with comment types
Comments are classified into three categories:
- Technical: Problem-solving, feature discussions, configuration questions
- Enthusiastic: Positive reactions, appreciation, general excitement
- Neutral: Informational, observational, or balanced commentary
- 485 active users with 4,529 connections
- Low density (0.0358) indicates sparse but well-connected structure
- Very high clustering coefficient (0.911) suggests tight-knit communities
- Network modularity of 0.703 confirms well-defined community boundaries
- 12 distinct communities detected via Louvain algorithm
- Balanced community sizes ranging from 11 to 56 members
- Two largest communities contain 56 members each
- Strong positive correlation (0.66) between degree and eigenvector centrality
- Cross-type interaction ratio of 0.645 indicates substantial mixing between user types
Most central user across multiple metrics: @jmr
- Highest degree centrality
- Highest comment volume
- Highest eigenvector centrality
Other key users:
- @johnny-d8o8m: High engagement and bridge position
- @peterdeuchar2042: Strong community connector
- @Smoothy0911: Active technical contributor
Category leaders:
- Technical community: @NeilNatic
- Enthusiastic community: @jmr
- Neutral community: @evlo8059
Overall sentiment distribution:
- Average polarity: 0.237 (positive)
- Most positive video: 2022.3 Release Party (0.647)
- Most negative video: 2022.10 Release Party (0.006)
- All videos maintain net positive sentiment
By comment type:
- Enthusiastic comments: 0.432 average sentiment
- Neutral comments: moderate sentiment
- Technical comments: 0.109 average sentiment
Temporal patterns:
- Negative correlation between neutral comments and sentiment (-0.299)
- Positive correlation between enthusiastic comments and sentiment (0.117)
- Slight negative correlation between technical comments and sentiment (-0.182)
- Neutral: 41.9%
- Enthusiastic: 34.5%
- Technical: 23.6%
Network connectivity by type:
- Enthusiastic users: 20.68 average connections (highest)
- Neutral users: moderate connectivity
- Technical users: 16.43 average connections (lowest)
Clustering by type:
- Neutral users: 0.862 clustering coefficient
- Enthusiastic users: 0.840 clustering coefficient
- Technical users: 0.698 clustering coefficient
- Sentiment assortativity: -0.001 (very weak negative)
- Comment type assortativity: -0.004 (very weak negative)
- Result: Users do not cluster by similarity; the community exhibits strong heterogeneous mixing
- Louvain and Leiden produce similar partitions (coarse granularity)
- Infomap yields finer-grained communities
- Louvain communities encompass multiple Infomap partitions
- Leiden aligns well with Louvain but with improved partition quality
The combination of high clustering (0.911) with weak assortativity (-0.001) indicates a mature community structure where users form tight local groups while maintaining connections across different user types. This suggests the community avoids echo chambers despite strong local cohesion.
Different user types exhibit distinct engagement patterns:
- Enthusiastic users drive connectivity (highest degree)
- Technical users form tighter sub-groups (lower degree, still high clustering)
- Neutral users balance between both patterns
The 0.645 cross-type interaction ratio demonstrates that community health does not depend on user homogeneity.
All analyzed videos maintain positive net sentiment despite individual variation. The relatively narrow range (0.006 to 0.647) suggests consistent community satisfaction with Home Assistant releases, with sentiment driven more by release-specific features than overall project trajectory.
The 0.66 correlation between degree and eigenvector centrality indicates preferential attachment: well-connected users tend to connect with other well-connected users. This core-periphery structure is typical of healthy online communities.
High modularity (0.703) coexisting with high global clustering (0.911) represents a community that successfully balances:
- Strong sub-community identity
- Cross-community interaction and knowledge sharing
Current analysis includes potential bot accounts. Future work should:
- Implement comment pattern analysis for bot detection
- Apply time-series regularity tests
- Filter low-quality engagement before network construction
- Note: Current dataset is too small for reliable low-degree node filtering
Extend beyond sentiment to topic-level analysis:
- Group comments by feature mentions (integrations, UI changes, performance)
- Apply LLM-based multi-label topic classification
- Correlate topic prevalence with release sentiment
- Track topic evolution across release cycles
Current analysis is static; temporal extension could reveal:
- User lifecycle patterns (newcomers vs veterans)
- Community structure evolution
- Sentiment trends correlated with Home Assistant version milestones
- Churn prediction and retention factors
Current language detection captures 90% of comments. Remaining 10% could be:
- Translated using DeepL API for improved coverage
- Analyzed for non-English community dynamics
- Used to detect geographic sub-communities
Extend beyond YouTube to other Home Assistant community channels:
- Reddit discussions
- Discord chat networks
- GitHub issue/PR interactions
- Forum activity patterns
Compare network structures and sentiment across platforms to identify platform-specific dynamics.
- Python 3.10 or higher (tested with Python 3.12)
- YouTube Data API v3 key (obtain from Google Cloud Console)
- Clone the repository:
git clone https://github.com/gerolori/Home-Assistant-Community-Analysis.git
cd Home-Assistant-Community-Analysis- Install dependencies:
pip install -r requirements.txt- Configure API credentials:
Create a
.envfile in the project root:
YOUTUBE_API_KEY=your_api_key_here
- Download required NLTK data (for TextBlob):
python -c "import nltk; nltk.download('punkt'); nltk.download('brown'); nltk.download('punkt_tab')"Alternatively, use TextBlob's automatic download:
python -m textblob.download_corporaOpen the main analysis notebook:
jupyter notebook "Home assistant youtube.ipynb"Execute cells sequentially to:
- Fetch YouTube data
- Build the comment network
- Apply community detection algorithms
- Generate sentiment analysis
- Produce visualizations
The analysis generates:
- Network visualizations with community coloring
- Centrality distribution plots
- Sentiment time series graphs
- Community overlap Venn diagrams
- Top user rankings by various metrics
Core libraries:
- google-api-python-client: YouTube API interaction
- networkx: Graph construction and analysis
- igraph: High-performance network algorithms
- python-louvain, leidenalg, infomap: Community detection
- textblob: Sentiment analysis
- langdetect: Language identification
- pandas, numpy, scipy: Data manipulation
- matplotlib, seaborn: Visualization
- scikit-learn: Machine learning utilities
See requirements.txt for complete dependency list with version constraints.
.
├── Home assistant youtube.ipynb # Main analysis notebook
├── Init.ipynb # Environment initialization
├── requirements.txt # Python dependencies
├── .env # API credentials (not committed)
├── .gitignore # Git exclusions
└── README.md # This file
- Sample size: Analysis limited to release party videos (subset of total community activity)
- Bot presence: No filtering applied for automated accounts
- Temporal scope: Static snapshot rather than longitudinal study
- Language coverage: 10% of comments not detected by langdetect
- Platform limitation: YouTube-only analysis (no cross-platform comparison)
- Reply structure: Current analysis treats all comments equally (no thread hierarchy)
This project was completed as part of the Social Media Mining course (Gerosa 966516) in 2025. The analysis demonstrates practical application of network science, NLP, and community detection algorithms to real-world social media data.
This project is provided as-is for educational and research purposes.
- Home Assistant development team for creating an active community
- YouTube Data API for data access
- Open-source community detection and NLP libraries