Model Card for EmotiTrack Emotion & Sarcasm Classifier (phi2_emotrack_finetuned)

A fine-tuned phi-2 language model for classifying emotional tone and detecting sarcasm in text messages. This model is part of the EmotiTrack project, designed for use in emotional journaling, mental health tech, and empathetic AI assistants.

---

Model Details

Model Description

• Developed by: Dennis Muturia for the EmotiTrack project
• Funded by: Self-funded
• Shared by: Dennis Muturia
• Model type: Causal Language Model
• Language(s): English
• License: Apache 2.0 (inherits from microsoft/phi-2)
• Finetuned from model: microsoft/phi-2 using PEFT (LoRA)

This model is fine-tuned to detect emotions (based on GoEmotions) and sarcasm (modeled after deepset/sarcasm-xlm-roberta-base) in short-form user text.

---

Model Sources

• Repository: https://huggingface.co/dmuturia/phi2_emotrack_finetuned

---

Uses

Direct Use

• Emotion & sarcasm classification for conversational agents
• Tagging emotional states in journaling or logging apps
• Enriching sentiment understanding in chatbots and voice assistants

Downstream Use

• Integrated into the EmotiTrack backend to drive real-time insight
• Supports longitudinal tracking of emotional states

Out-of-Scope Use

• Not suitable for diagnosis or clinical use
• Not designed for multilingual inputs
• Not ideal for long or formal documents

---

Bias, Risks, and Limitations

• May reflect bias from emotion/sarcasm datasets scraped from internet text
• Limited context window makes deep sarcasm or subtext detection unreliable
• Emotion misclassification may occur on neutral or ambiguous expressions

Recommendations

• Use in user-facing applications should include disclaimers
• Avoid using outputs for automated decision-making in sensitive domains

---

How to Get Started with the Model

from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model="dmuturia/phi2_emotrack_finetuned",
    tokenizer="dmuturia/phi2_emotrack_finetuned",
    device=0  # Use -1 for CPU
)

prompt = """### Instruction:
Classify the emotional tone and detect sarcasm in the following message.

### Input:
"I just love it when everything breaks right before a deadline."

### Response:
"""

output = pipe(prompt, max_new_tokens=60)[0]["generated_text"]
print(output)