app.py

import streamlit as st
from transformers import pipeline
from transformers import AutoModelForQuestionAnswering, AutoTokenizer


st.set_page_config(page_title="Automated Question Answering System")    # set page title

# heading 
st.markdown("<h2 style='text-align: center;'>Question Answering on Academic Essays</h2>", unsafe_allow_html=True)   
# description
st.markdown("<h3 style='text-align: left; color:#F63366; font-size:18px;'><b>What is extractive question answering about?<b></h3>", unsafe_allow_html=True)     
st.write("Extractive question answering is a Natural Language Processing task where text is provided for a model so that the model can refer to it and make predictions about where the answer to a question is.")

# store the model in cache resources to enhance efficiency (ref: https://docs.streamlit.io/library/advanced-features/caching)
@st.cache_resource(show_spinner=True)
def question_model():
    # call my model for question answering
    model_name = "kxx-kkk/FYP_deberta-v3-base-squad2_mrqa"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForQuestionAnswering.from_pretrained(model_name)
    question_answerer = pipeline("question-answering", model=model, tokenizer=tokenizer)
    return question_answerer

# get the answer by passing the context & question to the model
def question_answering(context, question):
    with st.spinner(text="Loading question model..."):
        question_answerer = question_model()
    with st.spinner(text="Getting answer..."):
        answer = question_answerer(context=context, question=question)
        answer_score = str(answer["score"])
        answer = answer["answer"]
        # display the result in container
        container = st.container(border=True)
        container.write("<h5><b>Answer:</b></h5>"+answer+"<p><small>(F1 score: "+answer_score+")</small></p><br>", unsafe_allow_html=True)

# choose the source with different tabs
tab1, tab2 = st.tabs(["Input text", "Upload File"])

'''----- input text -----'''
# if type the text as input
with tab1:  
    # set the example  
    sample_question = "What is NLP?"
    with open("sample.txt", "r") as text_file:
        sample_text = text_file.read()

    # Get the initial values of context and question
    context = st.session_state.get("contextInput", "")
    question = st.session_state.get("questionInput", "")

    # Button to try the example
    example = st.button("Try example")

    # Update the values if the "Try example" button is clicked
    if example:
        context = sample_text
        question = sample_question

    # Display the text area and text input with the updated or default values
    context = st.text_area("Enter the essay below:", value=context, key="contextInput", height=330)
    question = st.text_input(label="Enter the question: ", value=question, key="questionInput")
    
    # perform question answering when "get answer" button clicked
    button = st.button("Get answer", key="textInput")
    if button:
        if context=="" or question=="":
            st.error ("Please enter BOTH the context and the question", icon="🚨")
        else:
            question_answering(context, question)

'''----- upload file -----'''
# if upload file as input  
with tab2:
    # provide upload place
    uploaded_file = st.file_uploader("Choose a .txt file to upload", type=["txt"])

    # transfer file to context and allow ask question, then perform question answering 
    if uploaded_file is not None:
        raw_text = str(uploaded_file.read(),"utf-8")
        context = st.text_area("Your essay context: ", value=raw_text, height=330)
        question = st.text_input(label="Enter your question", value="Enter question here")

        # perform question answering when "get answer" button clicked
        button2 = st.button("Get answer", key="fileInput")
        if button2:
            if context=="" or question=="":
                st.error ("Please enter BOTH the context and the question", icon="🚨")
            else:
                question_answering(context, question)

st.markdown("<br><br><br><br><br>", unsafe_allow_html=True)