Skip to content

Repository files navigation

🧠 ArguFormer

AI-Powered Communication Intelligence for Arguments, Debates & Conversations

Understand not just WHAT people say — but HOW, WHY, and WITH WHAT EMOTION they say it.


🚧 ARGUFORMER IS CURRENTLY UNDER ACTIVE DEVELOPMENT 🚧

🔨 Building the next generation of AI-powered argument & communication analysis...

████████████████░░░░░░░░░░ 60%

Current Status: 🟡 Under Construction


💡 What is ArguFormer?

ArguFormer is an evolving NLP-based communication intelligence system designed to analyze human arguments and conversations from multiple dimensions.

Instead of simply asking:

"Is this argument correct?"

ArguFormer aims to ask:

"What is being argued, how is it being argued, what emotions are involved, and how does the conversation evolve?"

The long-term goal is to build a system capable of analyzing:

  • 🧠 Logical fallacies
  • 💬 Argument quality
  • 🎭 Emotional patterns
  • 😊 Sentiment
  • ☣️ Toxicity
  • 🎯 Stance
  • 🔥 Emotional escalation
  • 🔎 Evidence & contextual support
  • 🌍 Multilingual conversations
  • 📊 Speaker-level communication patterns

🚀 Current Development

ArguFormer is being developed incrementally as a modular NLP system.

✅ Currently Working

  • 🧹 Text preprocessing
  • 👥 Speaker segmentation
  • 📚 TF-IDF + Logistic Regression fallacy detection
  • 🤖 Transformer-based fallacy analysis
  • 🎭 GoEmotions-based emotion detection
  • 🔗 Emotion analysis integrated with speaker segments
  • 📈 Model evaluation & latency benchmarking
  • 📝 Logging & exception handling
  • ⚙️ Configurable project structure

🔨 Currently Building

  • 🎭 Emotion evaluation & threshold tuning
  • 📊 Unified analysis output
  • 🧠 Argument-quality scoring
  • 💬 Sentiment analysis
  • ☣️ Toxicity detection
  • 🔎 Retrieval-Augmented Generation (RAG)
  • 🌍 Multilingual analysis
  • 📈 Communication & emotional timelines

🔮 Future Vision

  • 🎙️ Speech-to-text conversation analysis
  • ⚡ Real-time analysis
  • 📊 Interactive communication dashboard
  • 🧑‍🤝‍🧑 Speaker behavior profiling
  • 🔥 Emotional escalation detection
  • 🧠 Evidence-aware reasoning
  • 🤖 AI-powered communication feedback
  • 📝 Debate & meeting intelligence

🏗️ Architecture

                    ┌─────────────────────┐
                    │   Debate / Text     │
                    └──────────┬──────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │   Preprocessing     │
                    └──────────┬──────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │ Speaker Segmentation│
                    └──────────┬──────────┘
                               │
              ┌────────────────┼────────────────┐
              │                │                │
              ▼                ▼                ▼
        🧠 Fallacy       🎭 Emotion       😊 Sentiment
          Analysis         Analysis         Analysis
              │                │                │
              └────────────────┼────────────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │ Communication       │
                    │ Intelligence Layer  │
                    └──────────┬──────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │ Reports / Dashboard │
                    └─────────────────────┘

🧠 NLP Pipeline

ArguFormer is being designed as a modular pipeline, where individual NLP capabilities can be developed, evaluated, and replaced independently.

Input
  │
  ▼
Preprocessing
  │
  ▼
Speaker Segmentation
  │
  ├──► Fallacy Detection
  │       ├── Classical ML
  │       └── Transformer
  │
  ├──► Emotion Detection
  │       └── RoBERTa + GoEmotions
  │
  ├──► Sentiment Analysis
  │
  ├──► Toxicity Detection
  │
  └──► Future NLP Modules
          │
          ▼
      Aggregation
          │
          ▼
   Communication Intelligence

🎭 Emotion Intelligence

One of the newest additions to ArguFormer is emotion analysis using:

SamLowe/roberta-base-go_emotions

The model analyzes 28 emotion categories, allowing ArguFormer to distinguish between signals such as:

anger
annoyance
approval
disapproval
fear
nervousness
sadness
admiration
confusion
...

Instead of reducing communication to:

Positive / Negative / Neutral

ArguFormer aims to understand the emotional nuance behind an argument.

For example:

"I completely disagree with your argument."

→ disapproval

while:

"I'm really worried about what will happen."

→ nervousness
→ fear

This will eventually contribute to emotional escalation and communication-pattern analysis.


⚙️ Technology Stack

Machine Learning / NLP

  • Python
  • Scikit-learn
  • Hugging Face Transformers
  • PyTorch
  • TF-IDF
  • Logistic Regression
  • RoBERTa
  • GoEmotions

Engineering

  • Modular Python architecture
  • CLI interface
  • YAML configuration
  • Logging
  • Custom exceptions
  • Evaluation & benchmarking

Planned

  • FAISS
  • Sentence Transformers
  • RAG
  • Whisper
  • FastAPI
  • Streamlit
  • Real-time processing

📂 Project Structure

Arguformer/
│
├── cli/
├── configs/
├── core/
│   ├── preprocessing.py
│   ├── segmentation.py
│   ├── emotion_analyzer.py
│   └── ...
│
├── models/
│   ├── fallacy_lr.py
│   ├── fallacy_transformer.py
│   └── ...
│
├── datasets/
├── evaluation/
├── rag/
├── tests/
├── utils/
├── data/
├── outputs/
└── main.py

📊 Development Philosophy

ArguFormer is being built using an incremental approach:

Experiment
    ↓
Validate
    ↓
Modularize
    ↓
Integrate
    ↓
Evaluate
    ↓
Improve

The goal isn't to throw dozens of AI models together.

The goal is to build a system where each component has a clear purpose, measurable behavior, and defined interface.


🗺️ Roadmap

PHASE 1 ─ Core NLP
████████████████████████████  ✅

PHASE 2 ─ Communication Intelligence
████████████████░░░░░░░░░░░░  🔨

PHASE 3 ─ RAG & Evidence Intelligence
██████░░░░░░░░░░░░░░░░░░░░░░  🔜

PHASE 4 ─ Real-Time Intelligence
███░░░░░░░░░░░░░░░░░░░░░░░░  🔮

PHASE 5 ─ Full Communication Platform
░░░░░░░░░░░░░░░░░░░░░░░░░░░  🔮

🧪 Project Status

Component Status
Project Architecture 🟢
Preprocessing 🟢
Speaker Segmentation 🟢
Classical Fallacy Model 🟢
Transformer Fallacy Model 🟢
Emotion Detection 🟢
Emotion Integration 🟢
Emotion Evaluation 🟡
Argument Quality 🔨
Sentiment Intelligence 🔜
Toxicity Analysis 🔜
RAG 🔜
Multilingual NLP 🔜
Real-Time Analysis 🔮
Interactive Dashboard 🔮

🚧 Why the Loading Bar?

Because ArguFormer isn't finished.

And that's intentional.

This repository documents the development of the system as it evolves from:

an argument/fallacy analyzer

into:

a broader AI communication-intelligence platform.

Expect experiments, refactoring, new models, benchmark improvements, architectural changes, and probably a few bugs along the way. 😄


🧠 Analyze Arguments.

🎭 Understand Emotions.

🔎 Find Evidence.

📊 Understand Communication.

ArguFormer — Still Building.

[████████████████░░░░░░░░░░]

⭐ Star the repository if you want to follow the build.

```

About

NLP debate analyzer comparing classical ML vs. transformers — TF-IDF+LogReg and DistilBERT for fallacy detection, plus sentiment and toxicity scoring on debate transcripts.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages