app.py

import gradio as gr 
import pandas as pd 
import numpy as np 
import matplotlib.pyplot as plt
import plotly.express as px
from stop_words import get_stop_words
from wordcloud import WordCloud
from datasets import load_dataset
import re

## import data 

dataset = load_dataset("Santarabantoosoo/italian_long_covid_tweets")
data = pd.DataFrame.from_dict(dataset["train"])


# load stop words

it_stop_words = load_dataset("Santarabantoosoo/italian-stopwords")
it_stop = pd.DataFrame.from_dict(it_stop_words["train"])

it_stop = it_stop.text.to_list()

## Optimize stop words according to Luca's repo

def format_input(user_key, stopwords):
  '''
  format user input request to lookup in the database of frequencies

  input:  
    user_key is a string
    stopwords is a list of strings
  output: 
    key is a string
  '''

  key = user_key.lower()
  key = re.sub(r'[^\w\s]', ' ', key)

  key = ' '.join([el for el in key.split() if not (el in stopwords)])


  return key


### Loading TFIDF

TFIDF_21_Jul_Oct = load_dataset("Santarabantoosoo/Long_Covid_word_frequency_TFIDF_21_Jul_Oct")

TFIDF_22_Feb_Apr = load_dataset("Santarabantoosoo/Long_Covid_word_frequency_TFIDF_22_Feb_Apr")

TFIDF_22_May_Jul = load_dataset("Santarabantoosoo/Long_Covid_word_frequency_TFIDF_22_May_Jul")

TFIDF_21_Nov_22_Jan = load_dataset("Santarabantoosoo/Long_Covid_word_frequency_TFIDF_21_Nov_22_Jan")


## Loading whole_text

whole_text_21_Jul_Oct = load_dataset("Santarabantoosoo/whole_text_TF_21_Jul_Oct")

whole_text_22_Feb_Apr = load_dataset("Santarabantoosoo/whole_text_TF_22_Feb_Apr")

whole_text_22_May_Jul = load_dataset("Santarabantoosoo/whole_text_TF_22_May_Jul")

whole_text_21_Nov_22_Jan = load_dataset("Santarabantoosoo/whole_text_TF_21_Nov_22_Jan")

TFIDF_21_Jul_Oct = pd.DataFrame.from_dict(TFIDF_21_Jul_Oct["train"])

TFIDF_22_Feb_Apr = pd.DataFrame.from_dict(TFIDF_22_Feb_Apr["train"])

TFIDF_22_May_Jul = pd.DataFrame.from_dict(TFIDF_22_May_Jul["train"])

TFIDF_21_Nov_22_Jan = pd.DataFrame.from_dict(TFIDF_21_Nov_22_Jan["train"])

whole_text_21_Jul_Oct = pd.DataFrame.from_dict(whole_text_21_Jul_Oct["train"])

whole_text_22_Feb_Apr = pd.DataFrame.from_dict(whole_text_22_Feb_Apr["train"])

whole_text_22_May_Jul = pd.DataFrame.from_dict(whole_text_22_May_Jul["train"])

whole_text_21_Nov_22_Jan = pd.DataFrame.from_dict(whole_text_21_Nov_22_Jan["train"])

ser_TFIDF = []

ser_TFIDF.append(TFIDF_21_Jul_Oct.transpose()[0])
ser_TFIDF.append(TFIDF_22_Feb_Apr.transpose()[0])
ser_TFIDF.append(TFIDF_22_May_Jul.transpose()[0])
ser_TFIDF.append(TFIDF_21_Nov_22_Jan.transpose()[0])

ser_whole_text = []

ser_whole_text.append(whole_text_21_Jul_Oct.transpose()[0])
ser_whole_text.append(whole_text_22_Feb_Apr.transpose()[0])
ser_whole_text.append(whole_text_22_May_Jul.transpose()[0])
ser_whole_text.append(whole_text_21_Nov_22_Jan.transpose()[0])


def plot_time_series(choice, keyword, user_keys):
    
    x = np.arange(2,10,2)
    
    y = [[] for j in range(len(keyword))]
    
    for j in range(len(keyword)):
      i=0
      while i < len(choice):
        try:
          y[j].append(choice[i][keyword[j]])
          i += 1
        except:
          y[j].append(0.0)
          i += 1

      y[j] = np.array(y[j])


    x_ticks_labels = ['Q1','Q2','Q3','Q4']

    fig, ax = plt.subplots(1,1) 

    for j in range(len(keyword)):
      ax.plot(x,y[j], label = user_keys[j].lower())


    # Set number of ticks for x-axis
    ax.set_xticks(x)
    ax.set_xticklabels(x_ticks_labels, fontsize=12)

    leg = plt.legend(loc='best')
    plt.xlabel('Time')
    plt.title("keywords quartely analysis (July 2021 - July 2022)")
    plt.ylabel(f'Freq. from {user_keys}')
    return fig

# Wordcloud with anger tweets
angry_tweets = data['tweet'][data["emotion"] == 'anger']
angry_tweets = angry_tweets.apply(format_input, args = [it_stop])
stop_words = ["https", 'http', "co", "RT"] + list(it_stop)
anger_wordcloud = WordCloud(max_font_size=50, max_words=50, background_color="white", stopwords = stop_words).generate(str(angry_tweets))


# Wordcloud with sad tweets
sad_tweets = data['tweet'][data["emotion"] == 'sadness']
sad_tweets = sad_tweets.apply(format_input, args = [it_stop])
stop_words = ["https", 'http', "co", "RT"] + list(it_stop)
sad_wordcloud = WordCloud(max_font_size=50, max_words=50, background_color="white", stopwords = stop_words).generate(str(sad_tweets))


 # Wordcloud with joy tweets
joy_tweets = data['tweet'][data["emotion"] == 'joy']
joy_tweets = joy_tweets.apply(format_input, args = [it_stop])
stop_words = ["https", 'http', "co", "RT"] + list(it_stop)
joy_wordcloud = WordCloud(max_font_size=50, max_words=50, background_color="white", stopwords = stop_words).generate(str(joy_tweets))


 # Wordcloud with fear tweets
fear_tweets = data['tweet'][data["emotion"] == 'fear']
fear_tweets = fear_tweets.apply(format_input, args = [it_stop])
stop_words = ["https", 'http', "co", "RT"] + list(it_stop)
fear_wordcloud = WordCloud(max_font_size=50, max_words=50, background_color="white", stopwords = stop_words).generate(str(fear_tweets))

## COmbine all plots in a single plot

wc_fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(2,2)

# fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(10, 5))

wc_fig.tight_layout()

ax1.imshow(sad_wordcloud, interpolation="bilinear")

ax1.axis("off")

ax1.set_title('Sadness', {'fontsize': 30})


ax2.imshow(joy_wordcloud, interpolation="bilinear")

ax2.axis("off")

ax2.set_title('Joy', {'fontsize': 30})


ax3.imshow(fear_wordcloud, interpolation="bilinear")

ax3.axis("off")

ax3.set_title('Fear', {'fontsize': 30})



ax4.imshow(anger_wordcloud, interpolation="bilinear")

ax4.axis("off")

ax4.set_title('Anger', {'fontsize': 30})


# plot a pie plot for emotions' distribution 

number_tweets_per_day = data.groupby(['date', 'emotion']).agg({'id': 'count'}).reset_index()

number_tweets_per_day["tweet_date"] = pd.to_datetime(number_tweets_per_day["date"])

time_fig = px.line(number_tweets_per_day, x = 'tweet_date', y = 'id', labels = {'id': 'count'}, color = 'emotion', 
                  color_discrete_sequence=px.colors.qualitative.G10)

# create a lineplot for emotions 

sentiment_counts = data.groupby('emotion').agg({'id' : 'size'}).reset_index()
sentiment_counts.rename(columns = {'id':'count'}, inplace = True)
sent_fig = px.pie(sentiment_counts, values='count', names='emotion', title='Tweets within each emotion', labels = {'id': 'count'}, 
                 color_discrete_sequence=px.colors.qualitative.G10)
sent_fig

def display_plot(image_choice):
    
    if image_choice == 'Sentiment distribution':
        return sent_fig
    
    elif image_choice == 'Time series':
        return time_fig
    
    elif image_choice == 'Word clouds':
        return wc_fig
    
def display_freq_plot(choice, *args):
    
    user_keys = [arg for arg in args]
    
    # clean input strings to match keywords in the database
    keyword = []
    for key in user_keys:
        keyword.append(format_input(key, it_stop))
        
    if choice == "TFIDF":
        return plot_time_series(ser_TFIDF, keyword, user_keys)
   
    elif choice == "Whole_text":
        return plot_time_series(ser_whole_text, keyword, user_keys)
    
def display_output(tweet_index):
    topics = "<ol>\
    <li>Discussion about scientific studies</li>\
    <li>Anxiety about pandemic and the information about it OR Specific people in the context of LC</li>\
    <li>Discussion about LC impact in terms of time periods</li>\
    <li>Discussion about LC impact on patient life (impact on life so far or scope for lifelong impact)</li>\
    <li>Treatment scenario</li>\
    <li>Impact/Consequences of LC on children</li>\
    </ol>"
    item = topic_dist_list[tweet_index]
    distribution = f'<html><body><h3>Topics Distribution</h3>({item[0][0]+1}, {item[0][1]}), ({item[1][0]+1}, {item[1][1]}), ({item[2][0]+1}, {item[2][1]}), ({item[3][0]+1}, {item[3][1]}), ({item[4][0]+1}, {item[4][1]}), ({item[5][0]+1}, {item[5][1]})\
    </body></html>'
    return gr.HTML.update(distribution, visible=True)

def display_output_Q2_Q4(tweet_index):
    item = topic_dist_list_Q2_Q4[tweet_index]
    distribution = f'<html><body><h3>Topics Distribution</h3>({item[0][0]+1}, {item[0][1]}), ({item[1][0]+1}, {item[1][1]}), ({item[2][0]+1}, {item[2][1]}), ({item[3][0]+1}, {item[3][1]}), ({item[4][0]+1}, {item[4][1]}), ({item[5][0]+1}, {item[5][1]})\
    </body></html>'
    return gr.HTML.update(distribution, visible=True)

# with gr.Blocks() as demo:
#     gr.Markdown("## Choose your adventure")
    
#     with gr.Tabs():
        
#         with gr.TabItem("Topic modeling"):
#             gr.Markdown("Nothing here yet")

#         with gr.TabItem("Word frequency"):

#             inputs =  [gr.Radio(choices = ['TFIDF', 'Whole_text'], label = 'Choose ur method'), 
#                       gr.Textbox(label = 'word 1'),
#                       gr.Textbox(label = 'word 2'),
#                       gr.Textbox(label = 'word 3'),
#                       gr.Textbox(label = 'word 4')]
#             plot_output = gr.Plot(elem_id = 1)
#             freq_button = gr.Button("Submit")
            

#         with gr.TabItem("Sentiment analysis"):
#             text_input =  gr.Radio(choices = ['Sentiment distribution', 'Word clouds', 'Time series'], label = 'Choose ur plot')
#             sent_plot = gr.Plot()
#             sent_button = gr.Button("Submit")          
            
        
#     sent_button.click(display_plot, inputs=text_input, outputs= sent_plot)
#     freq_button.click(display_freq_plot, inputs=inputs, outputs=plot_output)


with gr.Blocks() as demo:
    gr.Markdown("## Choose your adventure")
    
    with gr.Tabs():
        
        with gr.TabItem("Topic modeling"):
             gr.Markdown(
                """
            ## <div style="text-align: center;">Topic modeling analysis on Twitter</div>
            """
            )
             with gr.Tabs():
                 with gr.TabItem("July-Semptember 2021"):
                     with gr.Row():
                         gr.Image("./wordclouds_Q1 data.png", label="July-September 2021")



                     tweets_list = ['C\'è uno studio a riguardo condotto proprio sui più giovani che identifica il long covid alla stregua di ogni strascico di malattie infettive polmonari. Il long covid è dannoso come una polmonite in quanto a effetti a lungo termine.  Se lo ritrovo te lo passo, ora sono fuori...',
                    'Mio cugino è guarito dal covid dopo 4 mesi di ospedale,  di cui  più di 2 intubato, grazie alla testardaggine dei medici che hanno fatto di tutto per salvargli la vita a 57 anni. Ora è nella fase long covid per recuperare i danni fisici riportati',
                    'È importante parlare di #LongCovid e sensibilizzare tutti, giovani compresi, che non è un gioco ma una malattia debilitante/invalidante che può stravolgere la vita. Io 39 anni e #LongCovid da 18 mesi (con 4 figli piccoli). #countlongcovid',
                    'Il Long Covid è una diretta conseguenza di quelli che nei primi tempi sono stati abbandonati a se stessi giorni e giorni e curati solo quando molto aggravati, in ospedale. Se ti curi tempestivamente non hai nessuna conseguenza.',
                    'Non sai di cosa parli sono stato un mese attaccato ad un respiratore e sono salvo per miracolo. Ma questo è niente in confronto con il #LongCovid che mi porto dietro da mesi e mesi. Siete dei criminali a pensare ch\'è meglio curare che prevenire. Dei pazzi da rinchiudere',
                    'A chi dice ""Il COVID è innocuo per i bambini"".   Oltre ad alcuni decessi 500+ bambini sono morti di COVID negli USA 2020)  c\'è #LongCOVID.  Se ne parla in questo studio:   ""Studio inglese rileva che il COVID a lungo colpisce fino a 1 bambino su 7 mesi dopo l\'infezione']

                     q1_data_topic_list=['0. Discussion about scientific studies','1. Anxiety about pandemic and the information about it OR Specific people in the context of LC',
                    '2. Discussion about LC impact in terms of time periods','3. Discussion about LC impact on patient life (impact on life so far or scope for lifelong impact)' ,
                    '4. Treatment scenario', '5. Impact/Consequences of LC on children']


                     topic_dist_list=[[(0, 0.2181524), (1, 0.13380228), (2, 0.021277282), (3, 0.48123622), (4, 0.01883339), (5, 0.12669843)],
                    [(0, 0.0145399235), (1, 0.01287178), (2, 0.43158862), (3, 0.24750596), (4, 0.264914), (5, 0.028579665)],
                    [(0, 0.016303344), (1, 0.014450405), (2, 0.36162496), (3, 0.48426068), (4, 0.023487965), (5, 0.09987263)],
                    [(0, 0.018612841), (1, 0.016472807), (2, 0.44922927), (3, 0.033633586), (4, 0.026889767), (5, 0.45516175)],
                    [(0, 0.016305258), (1, 0.014453228), (2, 0.7628153), (3, 0.029092493), (4, 0.14613572), (5, 0.031198042)],
                    [(0, 0.016303508), (1, 0.014449066), (2, 0.15605325), (3, 0.029179793), (4, 0.023376595), (5, 0.7606378)]]

                     topics = '<html><body>\
                        <h3><b>Topics July to Sept, 2021</b></h3>\
                        <ol type="1">\
                        <li>1. Discussion about scientific studies</li>\
                        <li>2. Anxiety about pandemic and the information about it OR Specific people in the context of LC</li>\
                        <li>3. Discussion about LC impact in terms of time periods</li>\
                        <li>4. Discussion about LC impact on patient life (impact on life so far or scope for lifelong impact)</li>\
                        <li>5. Treatment scenario</li>\
                        <li>6. Impact/Consequences of LC on children</li>\
                        </ol>\
                        </body></html>'

                     Q1_topics = gr.HTML(topics, visible=True)

                     gr.Markdown(
                        """
                    ### Test our topic modeling model : select a tweet and check the topics distribution ! 
                    """
                    )

                     tweet = gr.Dropdown(tweets_list, label="Example tweets", interactive=True, type="index")

                     model_output = gr.HTML("", visible=False)
                     tweet.change(display_output, tweet, model_output)

                 with gr.TabItem("October 2021-July 2022"):

                     topic_dist_list_Q2_Q4=[[(0, 0.4377157), (1, 0.05924045), (2, 0.1525337), (3, 0.1941842), (4, 0.075339705), (5, 0.08098622)],
                    [(0, 0.16064012), (1, 0.063850455), (2, 0.08664099), (3, 0.2870743), (4, 0.081202514), (5, 0.32059166)],
                    [(0, 0.14904374), (1, 0.059243646), (2, 0.08039133), (3, 0.26638654), (4, 0.07534457), (5, 0.36959016)],
                    [(0, 0.14897935), (1, 0.059245925), (2, 0.08039324), (3, 0.41068354), (4, 0.14752874), (5, 0.15316921)],
                    [(0, 0.089826144), (1, 0.069229595), (2, 0.09393969), (3, 0.5643193), (4, 0.08804329), (5, 0.09464199)],
                    [(0, 0.08284077), (1, 0.29718927), (2, 0.08663448), (3, 0.36485678), (4, 0.08119658), (5, 0.08728213)]]

                     with gr.Row():
                            gr.Image("./wordclouds_Q2-Q2 data.png", label="October 2021-July 2022")

                     Q2_Q4_topics = '<html><body>\
                            <h3><b>Topics October 2021 to July 2022</b></h3>\
                            <ol type="1">\
                            <li>1. Variants</li>\
                            <li>2. Vaccine side-effects (and general anti-vax/ anti-LC narrative)</li>\
                            <li>3. Aftermath of LC or vaccine</li>\
                            <li>4. Impact of LC in terms of time OR Risks/Symptoms of LC</li>\
                            <li>5.  Anger or anxiety about LC information</li>\
                            <li>6. Discussion or Information about the science/knowledge surrounding LC</li>\
                            </ol>\
                            </body></html>'


                     Q2_Q4_topics_html = gr.HTML(Q2_Q4_topics, visible=True)

                     tweet_list_Q2_Q4=["Omicron e Long Covid: palpitazioni e perdita d'udito tra i sintomi - #Omicron #Covid: #palpitazioni ", 
                    'Long Covid e trombosi. La correlazione è spiegata da Giovanni Esposito, Presidente GISE, in un articolo sul sito  https://t.co/8TdI9nhDHY e avvalorata da uno studio svedese pubblicato sul British Medical Journal.  https://t.co/UebaXUtfbz',
                    'Peccato che il ""long COVID"" che è proprio ciò di cui parla l\'esimio dottore citato determini una alterazione o soppressione del sistema immunitario di cui si sa ancora poco ma che può portare a conseguenze fatali per il paziente.',
                    'Il Long covid rappresentava un problema solo fino ad aprile 2021, i vaccini hanno molto ridotto l\'impatto e la gravità delle patologie a lungo termine, in pratica si può dire che il long covid non esiste più',
                    'Sicuro, 100-150 morti al giorno, 6 ondate l anno, rischio long covid, rischio evoluzionario, e via dicendo — finitissimo',
                    'le cure le fai giorno dopo giorno... ci sono casi di long-covid dopo 6 mesi dall\'infezione. [Vaccino > >Cure] è un dato di fatto',
                    'A parte il rischio di sviluppare il #longcovid, il pericolo grave di lasciar circolare il virus e di farlo diventare endemico come preconizza il governo e lo sciagurato #speranza non è nel decorso del singolo caso ma nell\'aumento proporzionale dell\'insorgere di nuove varianti']

                     gr.Markdown(
                        """
                    ### Test our topic modeling model : select a tweet and check the topics distribution ! 
                    """
                    )

                     tweet_Q2_Q4 = gr.Dropdown(tweet_list_Q2_Q4, label="Example tweets", interactive=True, type="index")

                     model_output_Q2_Q4 = gr.HTML("", visible=False)
                     tweet_Q2_Q4.change(display_output_Q2_Q4, tweet_Q2_Q4, model_output_Q2_Q4)
        with gr.TabItem("Word frequency"):

            inputs =  [gr.Radio(choices = ['TFIDF', 'Whole_text'], label = 'Choose ur method'), 
                      gr.Textbox(label = 'word 1'),
                      gr.Textbox(label = 'word 2'),
                      gr.Textbox(label = 'word 3')]
            plot_output = gr.Plot()
            freq_button = gr.Button("Submit")

            freq_button.click(display_freq_plot, inputs=inputs, outputs=plot_output)
            gr.Examples(
        examples= [['Stanchezza', "l'età", '#LongCovidKids'], ['nebbia cognitiva', 'mal di testa', 'Dura PROVA']], 
        inputs= [gr.Textbox(),gr.Textbox(),gr.Textbox()],
        outputs=plot_output,
        fn=display_freq_plot)

        with gr.TabItem("Sentiment analysis"):
            text_input =  gr.Radio(choices = ['Sentiment distribution', 'Word clouds', 'Time series'], label = 'Choose ur plot')
            sent_plot = gr.Plot()
            sent_button = gr.Button("Submit")          

            sent_button.click(display_plot, inputs=text_input, outputs= sent_plot)


demo.launch(debug=True, show_error = True);

demo.launch(debug=True, show_error = True);