Multi-Task Ticket Classifier 🧾⚡

A traditional NLP-based system that automates multi-label classification and entity extraction from customer support tickets, designed as part of a machine learning internship assignment.

🚀 Project Overview

This project builds an end-to-end pipeline to process raw customer support tickets and perform:

• Issue Type Classification
• Urgency Level Prediction
• Key Entity Extraction (product names, complaint keywords, and dates)

The goal is to simulate a real-world enterprise helpdesk setting using classical NLP and ML techniques, with a focus on interpretability, robustness, and modular design.

🔍 Problem Statement

Given raw customer support ticket texts, automate the following tasks:

1. Predict the issue type (multi-class classification)
2. Predict the urgency level (Low, Medium, High)
3. Extract key entities from the text such as:
   • Product names (from a fixed set)
   • Complaint-related keywords (e.g., "broken", "delay", "error")
   • Dates or temporal references

🧠 Approach

🔧 Data Preprocessing

• Removed noise and normalized text
• Tokenized and lemmatized sentences
• Handled missing entries and irrelevant characters

🧮 Feature Engineering

• TF-IDF and Bag-of-Words representations
• Custom features: ticket length, sentiment score

🧪 Model Building

• Logistic Regression, SVM, and Random Forests for classification
• Multi-task setup: trained issue type and urgency level classifiers separately

🗂️ Entity Extraction

• Used regex, rule-based matching, and dictionary lookups to extract:
  • Product names
  • Complaint indicators
  • Temporal patterns (dates)

🎛️ Unified Inference Pipeline

def process_ticket(ticket_text):
    return {
        "issue_type": predicted_label_1,
        "urgency_level": predicted_label_2,
        "entities": {
            "products": [...],
            "dates": [...],
            "keywords": [...]
        }
    }

---

🌐 Gradio Interface

A clean Gradio-based web app allows users to:

• Input ticket text

• View predicted labels and extracted entities

• Live Demo: .

• Interface Preview of the Gradio App

---

📊 Evaluation

• Issue Type Accuracy: 88.7%

• Urgency Level Accuracy: 85.3%

• Precision/Recall/F1: Provided in full report inside the notebook

• Entity Extraction: Verified against product dictionary and regex benchmarks

---

📌 Highlights

• Modular, interpretable ML pipeline

• Traditional NLP without heavy dependencies

• Practical utility in customer service automation

• Live demo for real-time ticket analysis

📄 License

This project is licensed under the MIT License.

📬 Contact

Naman Kumar

📧 Email: namankr24@gmail.com

🔗 GitHub: NamanKr24