Fatima Tariq

Academic

Career

Verbatim text

FASTalk (Academic Chatbot) 
 
FASTalk is an academic, domain-specific chat-bot, based on Deep Learning and Artificial 
Intelligence, that is designed to assist students at FAST NUCES in their queries related to university 
and academics. It can answer queries related to class timings, class location, information of 
different courses and teacher related information in natural language i.e. English. As it is a research 
based project, we have worked on three different models. Our main goal was to build a generative 
model and achieve good accuracy on a small dataset. 
The first model was Doc2Vec, which worked on similarity basis. The dataset used for this model was 
in the form of Question Answer pairs. The overall correctness we achieved on our test dataset was 
0.9 out of 1. We got a very good accuracy on this model, but it is a retrieval-based (non-generative) 
model. 
Our second model was Encoder-Decoder based LSTM, which is a deep learning model. The dataset 
for this model was also in the form of Question Answer pairs. It is generative in nature, but the BLEU score we achieved for this model was 0.61. 
Our final model is End to End Memory Network based on LSTM. It is a generative model that is 
divided into two sub-models. The first part of this model takes query as input from user and finds 
relevant answer keywords, while the other part generates a complete sentence from the extracted 
answer keywords, which is our final output. We have achieved BLEU score of 0.87 for this model, which has outperformed the previous models. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Technology Used: 
Python, Tensorflow, Keras, NLTK, Genism, 
 Jupyter Notebook, Google Colab 
Supervisor Name: 
Syed M. Hassan Mustafa 
Group Members:   
Maham Majid (i16 - 0186) 
Fatima Tariq (i16 - 0293) 
Sohla Mamoon (i16 - 0332)