🧠 TriTaskNLP

Unified Multi-Task Learning for Topic, Sentiment, and Author Analysis

---

Overview

TriTaskNLP is a multi-task Natural Language Processing system that performs:

• Topic Classification
• Sentiment Analysis
• Author Identification

within a single unified architecture.

Rather than training separate models, the system uses shared representation learning, allowing a single encoder to extract linguistic features that generalize across tasks. This reduces redundancy and improves both efficiency and performance.

---

Why This Project Exists

Most NLP systems treat tasks in isolation. In practice, however:

• sentiment depends on topic context
• writing style influences both tone and structure
• semantic features overlap across tasks

TriTaskNLP explores how a shared latent space can capture these relationships more effectively than independent pipelines.

---

Architecture

Input Text → Preprocessing → Embedding Layer → Shared Encoder
                ↓
        Shared Representation
                ↓
     ┌────────────┬────────────┬────────────┐
     │ Topic      │ Sentiment  │ Author     │
     └────────────┴────────────┴────────────┘

Key Design Decisions

• Shared Encoder (BiLSTM / Transformer) Learns contextual, semantic, and stylistic features once

• Task-Specific Heads Independent classifiers for each objective

• Weighted Multi-Loss Optimization Balances contribution of each task

---

Demo

---

Interface

CLI Dashboard

The system provides:

• interactive menu (number-based)
• command-based execution
• animated transitions
• structured outputs using tables and panels

---

Features

• Multi-task learning with shared representations
• GPU acceleration (CUDA support)
• Interactive CLI with Rich UI
• Live training visualization
• PCA, t-SNE, and clustering analysis
• Confusion matrix evaluation
• Modular and extensible architecture

---

Installation

git clone https://github.com/SanyogSingh07/TriTaskNLP.git
cd TriTaskNLP
pip install -r requirements.txt

---

Usage

Launch Interactive CLI

python main.py

Or run commands directly

python main.py train
python main.py evaluate
python main.py predict "This model is impressive"
python main.py visualize

---

Evaluation & Results

Task	Accuracy Range
Topic Classification	92–96%
Sentiment Analysis	90–94%
Author Identification	85–92%

---

Representation Analysis

These visualizations demonstrate how the shared embedding space separates different classes and tasks.

---

Project Structure

TriTaskNLP/
│
├── cli/                # CLI commands
├── data/               # Data loaders and preprocessing
├── models/             # Neural architectures
├── utils/              # Visualization, metrics, dashboard
│
├── train.py            # Training pipeline
├── evaluate.py         # Evaluation logic
├── main.py             # CLI entry point

---

Trade-offs

Advantages

• Shared learning improves generalization
• Reduced computational redundancy
• Unified architecture

Limitations

• Task interference (negative transfer)
• Requires careful loss balancing
• Increased system complexity

---

Future Work

• Transformer-based architecture (BERT / RoBERTa)
• Web-based dashboard (Streamlit)
• API deployment (FastAPI)
• Automated report generation

---

License

This project is licensed under the MIT License.

---

Author

Sanyog Kumar Singh B.Tech Data Science

---

Final Note

This project is not just about building a model — it’s about exploring how multiple dimensions of language can be understood through a shared representation. The goal is to move closer to systems that learn language holistically, rather than as isolated tasks.

---