comp-serp-data / README.md
goker's picture
init
8477df5
# Comprehensive SERP Data
This dataset contains comprehensive search engine ranking data collected from Google and Bing, along with extracted technical and content features for analyzing search engine ranking algorithms.
## 📊 Dataset Overview
- **Total Records**: 14,465 search results
- **Search Engines**: Google (5,895 results) and Bing (8,570 results)
- **Keywords**: 500 diverse search queries
- **Features**: 20 features including technical scores, content analysis, and ranking metadata
## 🎯 Research Purpose
This dataset was created to empirically characterize and compare the ranking environments of large-scale search engines (Google and Bing) through systematic data collection and feature extraction. It enables research on:
- Search engine ranking algorithm analysis
- Cross-engine comparison studies
- Technical SEO feature importance
- Content relevance analysis
- Ranking prediction modeling
## 📁 Dataset Structure
### Main Dataset File
- **File**: `datasets/dataset.csv`
- **Format**: CSV (Comma-separated values)
- **Encoding**: UTF-8
- **Size**: ~2MB
### Column Descriptions
#### Search Metadata
| Column | Type | Description |
|--------|------|-------------|
| `query` | string | The search query used |
| `engine` | string | Search engine (google/bing) |
| `position` | integer | Ranking position (1-20) |
| `url` | string | Full URL of the result |
| `hostname` | string | Domain name of the result |
| `file_name` | string | Unique identifier for the page |
#### Content Features
| Column | Type | Description | Range |
|--------|------|-------------|-------|
| `query_in_title` | integer | Query presence in page title (0/1) | 0-1 |
| `exact_query_in_title` | integer | Exact query match in title (0/1) | 0-1 |
| `query_in_h1` | integer | Query presence in H1 tag (0/1) | 0-1 |
| `exact_query_in_h1` | integer | Exact query match in H1 (0/1) | 0-1 |
| `query_density_body` | float | Query frequency in body content | 0.0-1.0 |
| `semantic_similarity_title_query` | float | Semantic similarity between title and query | 0.0-1.0 |
| `semantic_similarity_content_query` | float | Semantic similarity between content and query | 0.0-1.0 |
| `word_count` | integer | Number of words in page content | 0-∞ |
#### Technical Features (Lighthouse Scores)
| Column | Type | Description | Range |
|--------|------|-------------|-------|
| `performance_score` | float | Page load performance score | 0-100 |
| `accessibility_score` | float | Web accessibility compliance score | 0-100 |
| `best-practices_score` | float | Security and best practices score | 0-100 |
| `seo_score` | float | Search engine optimization score | 0-100 |
#### Analysis Features
| Column | Type | Description |
|--------|------|-------------|
| `rank_tier` | string | Ranking tier (High: 1-5, Medium: 6-10, Low: 11-20) |
| `cluster` | integer | K-means cluster assignment (0-5) |
## 🔬 Data Collection Methodology
### 1. Keyword Selection
- **Source**: `raw/keywords/keywords.csv`
- **Count**: 500 diverse search queries
- **Categories**: E-commerce, services, information, local search
- **Selection Criteria**: High search volume, diverse intent types
### 2. Search Engine Data Collection
- **Google**: Custom Search API (top 20 results per query)
- **Bing**: Bing Search API (top 20 results per query)
- **Collection Period**: Systematic collection with rate limiting
- **Error Handling**: Retry mechanisms and exception tracking
### 3. Web Page Processing
- **HTML Extraction**: Full page content capture using headless browsers
- **Screenshot Capture**: Visual page representation
- **Performance Measurement**: Lighthouse scores via PageSpeed Insights API
- **Content Analysis**: NLP-based feature extraction
### 4. Feature Extraction
- **Technical Features**: Automated Lighthouse scoring
- **Content Features**: Natural language processing and semantic analysis
- **Query Matching**: Exact and fuzzy matching algorithms
- **Semantic Similarity**: Sentence transformer-based relevance scoring
## 📈 Dataset Statistics
### Distribution by Search Engine
- **Google**: 5,895 results (40.7%)
- **Bing**: 8,570 results (59.3%)
### Distribution by Ranking Tier
- **High Tier** (positions 1-5): 3,784 results (26.2%)
- **Medium Tier** (positions 6-10): 3,837 results (26.5%)
- **Low Tier** (positions 11-20): 6,844 results (47.3%)
### Cluster Distribution
- **Cluster 0**: 2,314 results (16.0%)
- **Cluster 1**: 2,122 results (14.7%)
- **Cluster 2**: 1,522 results (10.5%)
- **Cluster 3**: 5,142 results (35.6%)
- **Cluster 4**: 2,311 results (16.0%)
- **Cluster 5**: 1,054 results (7.3%)
## 🛠️ Technical Implementation
### Data Collection Pipeline
The dataset was created using the SERP Profiler Kit, a comprehensive research framework with the following components:
1. **Data Collection Modules**
- Google Custom Search API integration
- Bing Search API integration
- Web scraping with headless browsers
- PageSpeed Insights API integration
2. **Feature Extraction**
- Natural language processing with sentence transformers
- Technical SEO analysis using Lighthouse
- Content relevance scoring
- Query matching algorithms
3. **Data Processing**
- Quality validation and outlier detection
- Feature normalization and standardization
- Cluster analysis using K-means
- Statistical analysis and validation
### Quality Assurance
- **Exception Tracking**: Comprehensive error logging
- **Data Validation**: Multi-stage quality checks
- **Outlier Detection**: Statistical anomaly identification
- **Reproducibility**: Deterministic processing with fixed random seeds
## 📊 Research Applications
### 1. Clustering Analysis (RQ1)
- Identify distinct ranking profiles using K-means clustering
- Analyze feature patterns across different ranking strategies
- Validate cluster quality using multiple metrics
### 2. Feature Importance Analysis (RQ2)
- Determine which features most strongly predict ranking positions
- Compare feature importance across ranking tiers
- Identify engine-specific ranking factors
### 3. Cross-Engine Comparison (RQ3)
- Compare ranking characteristics between Google and Bing
- Analyze feature distribution differences
- Identify engine-specific ranking patterns
### 4. Ranking Prediction (RQ4)
- Build ordinal logistic regression models for ranking prediction
- Analyze feature coefficients and significance
- Validate model assumptions and performance
## 🔧 Usage Examples
### Python Usage
```python
import pandas as pd
# Load the dataset
df = pd.read_csv('datasets/dataset.csv')
# Basic statistics
print(f"Dataset shape: {df.shape}")
print(f"Engines: {df['engine'].value_counts()}")
print(f"Rank tiers: {df['rank_tier'].value_counts()}")
# Filter by search engine
google_results = df[df['engine'] == 'google']
bing_results = df[df['engine'] == 'bing']
# Analyze technical features
tech_features = ['performance_score', 'accessibility_score',
'best-practices_score', 'seo_score']
print(df[tech_features].describe())
# Analyze content features
content_features = ['query_in_title', 'query_in_h1',
'semantic_similarity_title_query', 'word_count']
print(df[content_features].describe())
```
### R Usage
```r
# Load the dataset
df <- read.csv('datasets/dataset.csv')
# Basic analysis
summary(df)
table(df$engine)
table(df$rank_tier)
# Technical features analysis
tech_cols <- c('performance_score', 'accessibility_score',
'best.practices_score', 'seo_score')
summary(df[tech_cols])
```
## 🔗 Related Resources
- **Source Code**: [SERP Profiler Kit Repository](https://github.com/gokerDEV/serp-profiler-kit)
- **Documentation**: Complete methodology and implementation details
- **Analysis Results**: Statistical analysis and visualization outputs
## 📄 License
This dataset is released under the MIT License. Please ensure compliance with search engine terms of service and website robots.txt files when using this data.
## 🤝 Contributing
For questions, issues, or contributions:
- Open an issue on the repository
- Review the documentation for methodology details
- Check the analysis results for statistical insights
## ⚠️ Important Notes
1. **API Compliance**: This dataset was collected following search engine API terms of service
2. **Ethical Scraping**: All web scraping was performed with respect to robots.txt files
3. **Data Freshness**: Search results may change over time; this dataset represents a snapshot
4. **Usage Limitations**: This dataset is for research purposes only
---
**Note**: This dataset enables systematic analysis of search engine ranking algorithms and provides a foundation for understanding how different factors influence search result positioning across major search engines.