Iterative-Classification

KO | EN

gitlite — search

#javascript #python #typescript #llm #golang #android #machine-learning #rust #macos #nodejs #windows #cli #openai #mcp

Iterative-Classification

(★ 8)

Implemention of an iterative model for making appropriate predictions using the contents of the links and relations that exist between the links in order to classify unlabeled data more accurately as compared to Content-based Bayesian classification.

Iterative-Classification Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Iterative Classification Algorithm (ICA) for Text Document Classification

A Python implementation of the Iterative Classification Algorithm that improves text categorization accuracy by exploiting relational information between linked documents.

## Project Overview

This project implements the Iterative Classification Algorithm (ICA) for text document classification. Unlike traditional content-based classification methods, ICA leverages both document content and the relationships between documents to improve classification accuracy. The algorithm is particularly effective for networked data such as scientific publications, where citation relationships provide valuable information for determining document categories.

## Features

- **Hybrid Classification Approach**: Utilizes both content features and relational information
- **Iterative Refinement**: Progressively improves classification accuracy through multiple iterations
- **Multiple Classifier Support**: Implements various machine learning algorithms including:
  - Support Vector Machines (SVM)
  - Multinomial Naive Bayes
  - Logistic Regression
  - Decision Tree Classifier
- **Cross-Validation**: Implements k-fold cross-validation for reliable performance evaluation
- **Aggregation Operators**: Supports different relational feature representations (count, mode, exists)

## Datasets

The implementation was tested on two standard datasets:

1. **CiteSeer**: 
   - 3,312 scientific papers
   - 3,703 unique words (appearing at least 10 times)
   - 4,732 citation connections (avg. 2.77 links per paper)
   - 6 topic classes

2. **CoRA**:
   - 2,708 machine learning papers
   - 1,433 unique words (appearing at least 10 times)
   - 5,429 citation connections (avg. 4.01 links per paper)
   - 7 topic classes

## Installation

```bash
# Clone the repository
git clone https://github.com/username/iterative-classification-algorithm.git
cd iterative-classification-algorithm

# Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt
```

## Dependencies

- NumPy: Efficient numerical array operations
- SciPy: Scientific computing functions
- scikit-learn: Machine learning algorithms and tools

## Usage

```python
from ica import IterativeClassificationAlgorithm
from sklearn.model_selection import train_test_split

# Load your dataset
X, y, links = load_dataset("path/to/dataset")

# Split the data
X_train, X_test, y_train, y_test, links_train, links_test = train_test_split(
    X, y, links, test_size=0.2, random_state=42
)

# Initialize and train the ICA model
ica = IterativeClassificationAlgorithm(
    base_classifier="decision_tree",
    max_iterations=10
)
ica.fit(X_train, y_train, links_train)

# Make predictions
predictions = ica.predict(X_test, links_test)

# Evaluate performance
accuracy = ica.score(X_test, y_test, links_test)
print(f"Accuracy: {accuracy * 100:.2f}%")
```

## Algorithm

The Iterative Classification Algorithm follows these steps:

1. **Bootstrapping**: Assign initial labels to all nodes using only content features
2. **Iteration**:
   - Generate a random ordering of nodes
   - For each node, recalculate labels using both content and relational features
   - Update node labels
3. **Termination**: Continue until labels stabilize or maximum iterations reached

```
Iterative Classification Algorithm (ICA)
for all Yi ∈ Y do
    Compute g(yi|vN(Yi);w) only using OA(Yi|xN(Yi))
    ∀yi ∈ C:
    Set yi = arg max g(y|vN(Yi))
end for
repeat
    Generate Ordering O over nodes Y
    for all Yi ∈ O do
        Compute g(yi|vN(Yi);w) ∀yi ∈ C:
        yi = arg max g(y|vN(Yi))
    end for
until labels are stabilized
```

## Results

Performance comparison of different classifiers on the datasets:

| Classifier | CiteSeer Accuracy | CoRA Accuracy |
|------------|-------------------|---------------|
| Decision Tree | 74.98% | 93.17% |
| Logistic Regression | 16.78% | 30.21% |
| Multinomial NB | 50.64% | 71.31% |
| SVM | 56.04% | 88.85% |

## Future Work

- Extend the iterative procedure for prediction of multiple object types
- Implement collaborative classification using multiple classifiers
- Explore additional aggregation operators for relational feature representation
- Optimize performance for larger network datasets

## References

1. Jennifer Neville, David Jensen. "Iterative Classification in Relational Data." Knowledge Discovery Laboratory, Department of Computer Science, University of Massachusetts.
2. Sofus A. Macskassy, Foster Provost. "A Simple Relational Classifier." NYU Stern School of Business.
3. Qing Lu, Lise Getoor. "Link Based Classification." Department of Computer Science, University of Maryland, College Park.
4. Prithviraj Sen, et al. "Collective Classification in Network Data."

## Author

Saurabh Katkar (skatkar@hawk.iit.edu)

## License

This project is licensed under the MIT License - see the LICENSE file for details.

Iterative-Classification

Iterative-Classification Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Iterative-Classification

Iterative-Classification Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README