Contents

1. Repository Name
2. Title of the Project
3. Short Description of the Project
4. Objectives of the Project
5. Name of the Dataset
6. Description of the Dataset
7. Goal of the Project using this Dataset
8. Size of the dataset
9. Algorithms which are used as part of our investigation
10. Project Requirements
11. Usage of the Project
12. Which chatbot architecture should the users use
13. Authors

Repository Name

smartchat-conversational-chatbot

Title of the Project

SmartChat: A Context-Aware Conversational Agent

Short Description of the Project

Develop a chatbot that can effectively adapt to context and topic shifts in a conversation, leveraging the Stanford Question Answering Dataset to provide informed and relevant responses, and thereby increasing user satisfaction and engagement.

Objectives of the Project

Create a user-friendly web or app interface that enables users to have natural and coherent conversations with the chatbot, with high satisfaction rating.

Name of the Dataset

The dataset used in this project is Stanford Question Answering Dataset.

Data Source: Kaggle

Type of the Dataset: Text

Description of the Dataset

The Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset consisting of questions posed by crowdworkers on a set of Wikipedia articles. The answer to every question is a segment of text, or span, from the corresponding reading passage. There are 100,000+ question-answer pairs on 500+ articles. More information can be found at: https://rajpurkar.github.io/SQuAD-explorer/

Goal of the Project using this Dataset

• The goal of the project is to develop a chatbot that can carry out multi-turn conversations, adapt to context, and handle a variety of topics.

Size of the Dataset:

• The dataset has 2 JSON files. One is for training and the other is for testing
  • dev-v1.1.json – 4.9 MB
  • train-v1.1.json – 30.3 MB

Algorithms which are used as part of our investigation

• 2 different architectures are used:
  • GPT2-Medium architecture using LoRA and PEFT
  • BERT (bert-base-uncased)

Project Requirements

• python3
• datasets
• torch
• peft
• transformers
• evaluate
• safetensors
• numpy
• pandas
• matplotlib
• scikit-learn
• seaborn
• nltk
• rouge-score
• rouge
• gradio
• tqdm

Usage of the Project

• Goto SQuAD Dataset Preprocessing and ensure that you have train-v1.1.json and dev-v1.1.json files.
  • In case you don't have them, you can download them here and here
• Goto SQuAD Dataset Preprocessing file and run all the cells.
• To execute and view the results of BERT (bert-base-uncased) approach, please go through the instructions provided in the SQuAD_chatbot_using_bert-base-uncased_README.md file.
• To execute and view the results of GPT (gpt2-medium using LoRA and PEFT) approach, please go through the instructions provided in the SQuAD_chatbot_using_gpt2-medium_README.md file.

Which chatbot architecture should the users use

• Actually, both the chatbots are running well.
• SQuAD_using_gpt2-medium generates the answers but most of the times it has issues.
  • For more information on the observations and technical details, refer training and validation files.
• SQuAD_using_bert-base-uncased is working very well as expected.
• Final conclusion is: Users can use any chatbot. But for perfect generation of answers, make use of SQuAD_using_bert-base-uncased.

Authors

Author	Contact Details
Ravi Teja Kothuru	rkothuru@sandiego.edu
Soumi Ray	soumiray@sandiego.edu
Anwesha Sarangi	asarangi@sandiego.edu