Service Quality AI-Based System

Introduction

The Service Quality AI-Based System is a state-of-the-art platform developed to revolutionize customer service and support, with a focus on Arabic-language industries. It achieves this through advanced natural language processing (NLP) techniques tailored to analyze customer interactions and generate solutions efficiently.

Key Features :

• Topic Classification: Automatically categorizes customer queries into predefined topics (F1 Score: 95.8%).
• Sentiment Analysis: Identifies the emotional tone of customer interactions (F1 Score: 96%).
• Solution Generation: Provides actionable responses to customer queries.

Technologies Used :

• Programming Language: Python
• Framework: Django
• AI Models: MARBERT, ARBERT, GPT, MT5
• Tools: Web scraping tools, tokenization libraries, hyperparameter tuning methods

---

Project Phases

1. Data Gathering

Data is collected using web scraping tools to ensure a comprehensive dataset covering diverse scenarios and linguistic nuances in Arabic. This supports robust training for the AI models.

2. Data Preprocessing

To prepare the data for training, it undergoes the following steps:

• Noise Removal: Cleaning special characters and irrelevant text.
• Tokenization and Normalization: Standardizing Arabic language data.
• Data Splitting: Dividing data into training, validation, and test sets.

3. Hyperparameter Tuning

Hyperparameter optimization ensures the models perform optimally. This involves fine-tuning configurations such as learning rates, batch sizes, and architecture parameters.

4. AI Models Utilized

Task	Model	Dataset	F1 Score
Topic Classification	MARBERT	Merged Topics dataset	95.8%
Sentiment Analysis	ARBERT	100k revs & glare dataset	96.0%
Solution Generation	GPT, MT5	Contextual solution dataset	N/A

5. Deployment

The system is deployed using a robust Django framework, providing:

• A seamless frontend and backend experience.
• A responsive and user-friendly interface for end-users.

---

How It Works

1. Data Input: Customer queries are ingested into the system.
2. Processing: The platform performs topic classification, sentiment analysis, and solution generation.
3. Output: Contextual responses or solutions are provided in real time.

---

Colab Notebooks

Explore the Colab notebooks for:

• Topic Classification Notebook

Click the links above to access the respective Colab notebooks and view the implementation details.

---

Benefits

• Accurate and efficient processing of Arabic-language customer interactions.
• Scalable and reliable system design.
• Simplifies customer support and enhances service quality.

---

About the Models:

ARBERT & MARBERT: Deep Bidirectional Transformers for Arabic

• introduce ARBERT and MARBERT, two powerful Transformer-based language models for Arabic;
• introduce ArBench, a multi-domain, multi-variety benchmark for Arabic naturaal language understanding based on 41 datasets across 5 different tasks and task clusters;
• evaluate ARBERT and MARBERT on ArBench and compare against available language models.

Model establish new state-of-the-art (SOTA) on all 5 tasks/task clusters on 37 out of the 41 datasets. Language models are publicaly available for research (see below). The rest of this repository provides more information about the language models, benchmark, and experiments.

How to use ARBERT and MARBERT

Loading directly from Huggingface

You can use ARBERT and MARBERT with Hugging Face's Transformers library as follow.

   from transformers import AutoTokenizer, AutoModel
   #load AEBERT model from huggingface
   ARBERT_tokenizer = AutoTokenizer.from_pretrained("UBC-NLP/ARBERT")
   ARBERT_model = AutoModel.from_pretrained("UBC-NLP/ARBERT")
 
   #load MAEBERT model from huggingface
   MARBERT_tokenizer = AutoTokenizer.from_pretrained("UBC-NLP/MARBERT")
   MARBERT_model = AutoModel.from_pretrained("UBC-NLP/MARBERT") 

4.2 Example: Fine-tuning MARBERT for Sentiment Analysis

MARBERT Fine-Tuning demo with PyTorch checkpoint for Sentiment Analysis on the AJGT dataset