app.py

"""
# MANIFESTO ANALYSIS
"""

##IMPORTING LIBRARIES
import random
import matplotlib.pyplot as plt
import nltk
from nltk.tokenize import word_tokenize,sent_tokenize
from nltk.corpus import stopwords
from nltk.stem.porter import PorterStemmer
from nltk.stem import WordNetLemmatizer
from nltk.corpus import stopwords 
from nltk.tokenize import word_tokenize
from nltk.probability import FreqDist
from cleantext import clean
import textract
import urllib.request
import nltk.corpus  
from nltk.text import Text
import io
from io import StringIO,BytesIO
import sys 
import pandas as pd
import cv2
import re
from wordcloud import WordCloud, STOPWORDS, ImageColorGenerator
from textblob import TextBlob
from PIL import Image
import os
import gradio as gr
from zipfile import ZipFile
import contractions
import unidecode
import groq
import json
from dotenv import load_dotenv
from sklearn.feature_extraction.text import TfidfVectorizer
from collections import Counter

# Load environment variables from .env file
load_dotenv()

nltk.download('punkt_tab')
nltk.download('stopwords')
nltk.download('punkt')
nltk.download('wordnet')
nltk.download('words')

# Initialize Groq client for LLM capabilities
try:
    groq_api_key = os.getenv("GROQ_API_KEY")
    if groq_api_key:
        groq_client = groq.Groq(api_key=groq_api_key)
    else:
        print("Warning: GROQ_API_KEY not found in environment variables. Summarization will be disabled.")
        groq_client = None
except Exception as e:
    print(f"Error initializing Groq client: {e}")
    groq_client = None


"""## PARSING FILES"""

#def Parsing(parsed_text):
  #parsed_text=parsed_text.name
  #raw_party =parser.from_file(parsed_text) 
 # raw_party = raw_party['content'],cache_examples=True
#  return clean(raw_party)
 
  
def Parsing(parsed_text):
  '''
  Process a PDF file and extract its text content
  parsed_text: Can be a file object with a 'name' attribute or a file path string
  '''
  try:
    # Handle different input types
    if hasattr(parsed_text, 'name'):
      file_path = parsed_text.name
    else:
      file_path = parsed_text
      
    # Extract text from PDF
    raw_party = textract.process(file_path, encoding='ascii', method='pdfminer')
    return clean(raw_party)
  except Exception as e:
    print(f"Error parsing PDF: {str(e)}")
    return f"Error parsing PDF: {str(e)}"


#Added more stopwords to avoid irrelevant terms
stop_words = set(stopwords.words('english'))
stop_words.update('ask','much','thank','etc.', 'e', 'We', 'In', 'ed','pa', 'This','also', 'A', 'fu','To','5','ing', 'er', '2')

"""## PREPROCESSING"""

def clean_text(text):
  '''
  The function which returns clean text
  '''
  text = text.encode("ascii", errors="ignore").decode("ascii")  # remove non-asciicharacters
  text=unidecode.unidecode(text)# diacritics remove
  text=contractions.fix(text) # contraction fix
  text = re.sub(r"\n", " ", text)
  text = re.sub(r"\n\n", " ", text)
  text = re.sub(r"\t", " ", text)
  text = re.sub(r"/ ", " ", text)
  text = text.strip(" ")
  text = re.sub(" +", " ", text).strip()  # get rid of multiple spaces and replace with a single
  
  text = [word for word in text.split() if word not in stop_words]
  text = ' '.join(text)
  return text

# text_Party=clean_text(raw_party)

def Preprocess(textParty):
  '''
  Removing special characters extra spaces
  '''
  text1Party = re.sub('[^A-Za-z0-9]+', ' ', textParty) 
  #Removing all stop words
  pattern = re.compile(r'\b(' + r'|'.join(stopwords.words('english')) + r')\b\s*')
  text2Party = pattern.sub('', text1Party)
  # fdist_cong = FreqDist(word_tokens_cong)
  return text2Party





'''
  Using Concordance, you can see each time a word is used, along with its 
  immediate context. It can give you a peek into how a word is being used
  at the sentence level and what words are used with it 
'''
def conc(text_Party,strng):
  word_tokens_party = word_tokenize(text_Party)
  moby = Text(word_tokens_party) 
  resultList = []
  for i in range(0,1):
      save_stdout = sys.stdout
      result = StringIO()
      sys.stdout = result
      moby.concordance(strng,lines=4,width=82)    
      sys.stdout = save_stdout      
  s=result.getvalue().splitlines()
  return result.getvalue()
  
def get_all_phases_containing_tar_wrd(target_word, tar_passage, left_margin = 10, right_margin = 10, numLins=4):
    """
        Function to get all the phases that contain the target word in a text/passage tar_passage.
        Workaround to save the output given by nltk Concordance function
         
        str target_word, str tar_passage int left_margin int right_margin --> list of str
        left_margin and right_margin allocate the number of words/pununciation before and after target word
        Left margin will take note of the beginning of the text
    """     
    # Handle empty or None search terms
    if not target_word or target_word.strip() == "":
        return "Please enter a search term"
        
    # Create list of tokens using nltk function
    tokens = nltk.word_tokenize(tar_passage)
     
    # Create the text of tokens
    text = nltk.Text(tokens)
 
    ## Collect all the index or offset position of the target word
    c = nltk.ConcordanceIndex(text.tokens, key = lambda s: s.lower())
 
    ## Collect the range of the words that is within the target word by using text.tokens[start;end].
    ## The map function is use so that when the offset position - the target range < 0, it will be default to zero
    concordance_txt = ([text.tokens[list(map(lambda x: x-5 if (x-left_margin)>0 else 0,[offset]))[0]:offset+right_margin] for offset in c.offsets(target_word)])
                         
    ## join the sentences for each of the target phrase and return it
    result = [''.join([x.replace("Y","")+' ' for x in con_sub]) for con_sub in concordance_txt][:-1]
    result=result[:numLins+1]
    
    res='\n\n'.join(result)
    return res  
  

def normalize(d, target=1.0):
   raw = sum(d.values())
   factor = target/raw
   return {key:value*factor for key,value in d.items()}


def generate_summary(text, max_length=1000):
    """
    Generate a summary of the manifesto text using Groq LLM
    """
    if not groq_client:
        return "Summarization is not available. Please set up your GROQ_API_KEY in the .env file."
    
    # Truncate text if it's too long to fit in context window
    if len(text) > 10000:
        text = text[:10000]
    
    try:
        # Use Groq's LLaMA 3 model for summarization
        completion = groq_client.chat.completions.create(
            model="llama3-8b-8192",  # Using LLaMA 3 8B model
            messages=[
                {"role": "system", "content": "You are a helpful assistant that summarizes political manifestos. Provide a concise, objective summary that captures the key policy proposals, themes, and promises in the manifesto."},
                {"role": "user", "content": f"Please summarize the following political manifesto text in about 300-500 words, focusing on the main policy areas, promises, and themes:\n\n{text}"}
            ],
            temperature=0.3,  # Lower temperature for more focused output
            max_tokens=800,   # Limit response length
        )
        
        return completion.choices[0].message.content
    except Exception as e:
        return f"Error generating summary: {str(e)}. Please check your API key and connection."

def fDistance(text2Party):
  '''
  Most frequent words search using TF-IDF to find more relevant words
  '''
  # Traditional frequency distribution
  word_tokens_party = word_tokenize(text2Party) #Tokenizing
  fdistance = FreqDist(word_tokens_party).most_common(10)
  mem={}
  for x in fdistance:
    mem[x[0]]=x[1]
    
  # Enhanced with TF-IDF for better relevance
  sentences = sent_tokenize(text2Party)
  
  # Use TF-IDF to find more relevant words
  vectorizer = TfidfVectorizer(max_features=15, stop_words='english')
  tfidf_matrix = vectorizer.fit_transform(sentences)
  
  # Get feature names (words)
  feature_names = vectorizer.get_feature_names_out()
  
  # Calculate average TF-IDF score for each word across all sentences
  tfidf_scores = {}
  for i, word in enumerate(feature_names):
      scores = [tfidf_matrix[j, i] for j in range(len(sentences)) if i < tfidf_matrix[j].shape[1]]
      if scores:
          tfidf_scores[word] = sum(scores) / len(scores)
  
  # Sort by score and get top words
  sorted_tfidf = dict(sorted(tfidf_scores.items(), key=lambda x: x[1], reverse=True)[:10])
  
  # Combine traditional frequency with TF-IDF for better results
  combined_scores = {}
  for word in set(list(mem.keys()) + list(sorted_tfidf.keys())):
      # Normalize and combine both scores (with more weight to TF-IDF)
      freq_score = mem.get(word, 0) / max(mem.values()) if mem else 0
      tfidf_score = sorted_tfidf.get(word, 0) / max(sorted_tfidf.values()) if sorted_tfidf else 0
      combined_scores[word] = (freq_score * 0.3) + (tfidf_score * 0.7)  # Weight TF-IDF higher
  
  # Get top 10 words by combined score
  top_words = dict(sorted(combined_scores.items(), key=lambda x: x[1], reverse=True)[:10])
  
  return normalize(top_words)

def fDistancePlot(text2Party,plotN=15):
  '''
  Most Frequent Words Visualization
  '''
  word_tokens_party = word_tokenize(text2Party) #Tokenizing
  fdistance = FreqDist(word_tokens_party)
  plt.title('Frequency Distribution')
  plt.axis('off')
  plt.figure(figsize=(4,3))
  fdistance.plot(plotN)
  plt.tight_layout()
  buf = BytesIO()
  plt.savefig(buf)
  buf.seek(0)
  img1 = Image.open(buf)
  plt.clf() 
  return img1


def DispersionPlot(textParty):
  '''
  Dispersion PLot
  '''
  word_tokens_party = word_tokenize(textParty) #Tokenizing
  moby = Text(word_tokens_party) 
  fdistance = FreqDist(word_tokens_party)
  word_Lst=[]
  for x in range(5):
    word_Lst.append(fdistance.most_common(6)[x][0]) 
  
  plt.axis('off')
  plt.figure(figsize=(4,3))
  plt.title('Dispersion Plot')
  moby.dispersion_plot(word_Lst)
  plt.plot(color="#EF6D6D")
  plt.tight_layout()
  buf = BytesIO()
  plt.savefig(buf)
  buf.seek(0)
  img = Image.open(buf)
  plt.clf() 
  return img


def getSubjectivity(text):
  
  '''
  Create a function to get the polarity
  '''
  return TextBlob(text).sentiment.subjectivity


def getPolarity(text):
  '''
  Create a function to get the polarity
  '''
  return  TextBlob(text).sentiment.polarity
  
  
def getAnalysis(score):
  if score < 0:
    return 'Negative'
  elif score == 0:
    return 'Neutral'
  else:
    return 'Positive'
def Original_Image(path):
  img= cv2.imread(path)
  img= cv2.cvtColor(img, cv2.COLOR_BGR2RGB)  
  return img

def Image_Processed(path):
  '''
  Reading the image file
  '''
  img= cv2.imread(path)
  img= cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
  
  #Thresholding
  ret, bw_img = cv2.threshold(img, 124, 255, cv2.THRESH_BINARY)

  return bw_img

def word_cloud(orgIm,mask_img,text_Party_pr,maxWord=2000,colorGener=True,
    contCol='white',bckColor='white'):
  '''
  #Generating word cloud
  '''  
  mask =mask_img
  # Create and generate a word cloud image:
  wordcloud = WordCloud(max_words=maxWord, background_color=bckColor,
                        mask=mask,
                        colormap='nipy_spectral_r',
                        contour_color=contCol,
                        width=800, height=800,
                        margin=2,
                        contour_width=3).generate(text_Party_pr)

  # create coloring from image

  
  plt.axis("off")
  if colorGener==True:
    image_colors = ImageColorGenerator(orgIm)
    plt.imshow(wordcloud.recolor(color_func= image_colors),interpolation="bilinear")
    
  
  else:    
    plt.imshow(wordcloud)
    
  
  
 
def word_cloud_generator(parsed_text_name,text_Party):
  parsed=parsed_text_name.lower()

  if 'bjp' in parsed:
    orgImg=Original_Image('bjpImg2.jpeg')
    bwImg=Image_Processed('bjpImg2.jpeg')
    plt.figure(figsize=(6,5))
    word_cloud(orgImg,bwImg,text_Party,maxWord=3000,colorGener=True,
    contCol='white', bckColor='black')
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf)
    buf.seek(0)
    img1 = Image.open(buf)
    plt.clf() 
    return img1

  
  elif 'congress' in parsed:
    orgImg=Original_Image('congress3.jpeg')
    bwImg=Image_Processed('congress3.jpeg')
    plt.figure(figsize=(5,4))
    word_cloud(orgImg,bwImg,text_Party,maxWord=3000,colorGener=True)
    
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf)
    buf.seek(0)
    img2 = Image.open(buf)
    plt.clf() 
    return img2
    #congrsMain.jpg
    
  
  elif 'aap' in parsed:
    orgImg=Original_Image('aapMain2.jpg')
    bwImg=Image_Processed('aapMain2.jpg')
    plt.figure(figsize=(5,4))
    word_cloud(orgImg,bwImg,text_Party,maxWord=3000,colorGener=False,contCol='black')
    
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf)
    buf.seek(0)
    img3 = Image.open(buf)
    plt.clf() 
    return img3
  
  else :
    wordcloud = WordCloud(max_words=2000, background_color="white",mode="RGB").generate(text_Party)
    plt.figure(figsize=(5,5))
    plt.imshow(wordcloud, interpolation="bilinear")
    plt.axis("off")   
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf)
    buf.seek(0)
    img4 = Image.open(buf)
    plt.clf()
    return img4



'''
url = "http://library.bjp.org/jspui/bitstream/123456789/2988/1/BJP-Election-english-2019.pdf"
path_input = "./Bjp_Manifesto_2019.pdf"
urllib.request.urlretrieve(url, filename=path_input)

url="https://drive.google.com/uc?id=1BLCiy_BWilfVdrUH8kbO-44DJevwO5CG&export=download"
path_input = "./Aap_Manifesto_2019.pdf"
urllib.request.urlretrieve(url, filename=path_input)

url="https://drive.google.com/uc?id=1HVZvTtYntl0YKLnE0cwu0CvAIRhXOv60&export=download"
path_input = "./Congress_Manifesto_2019.pdf"
urllib.request.urlretrieve(url, filename=path_input)
'''
def analysis(Manifesto, Search): 
  '''
  Main analysis function that processes the manifesto and generates all outputs
  Manifesto: PDF file uploaded by the user
  Search: Search term entered by the user
  '''
  try:
    print(f"Analysis function called with: Manifesto={Manifesto}, Search={Search}")
    
    # Check if a file was uploaded
    if Manifesto is None:
      print("No file uploaded")
      return "Please upload a PDF file", {}, None, None, None, None, None, "No file uploaded"
      
    # Handle empty search term
    if Search is None or Search.strip() == "":
      Search = "government"  # Default search term
      print(f"Using default search term: {Search}")
    else:
      print(f"Using provided search term: {Search}")
      
    # Process the uploaded PDF
    print(f"Processing file: {Manifesto.name if hasattr(Manifesto, 'name') else Manifesto}")
    raw_party = Parsing(Manifesto)
    
    # Check if parsing was successful
    if isinstance(raw_party, str) and raw_party.startswith("Error"):
      print(f"Parsing error: {raw_party}")
      return raw_party, {}, None, None, None, None, None, "Error generating summary due to parsing failure"
    
    print("Parsing successful, cleaning text...")
    text_Party = clean_text(raw_party)
    text_Party_processed = Preprocess(text_Party)

    # Generate summary using LLM
    print("Generating summary...")
    summary = generate_summary(raw_party)

    # Sentiment analysis
    print("Performing sentiment analysis...")
    df = pd.DataFrame(raw_party.split('\n'), columns=['Content'])
    df['Subjectivity'] = df['Content'].apply(getSubjectivity)
    df['Polarity'] = df['Content'].apply(getPolarity)
    df['Analysis on Polarity'] = df['Polarity'].apply(getAnalysis)
    df['Analysis on Subjectivity'] = df['Subjectivity'].apply(getAnalysis)
    
    # Generate sentiment analysis plot
    print("Generating sentiment analysis plot...")
    plt.title('Sentiment Analysis')
    plt.xlabel('Sentiment')
    plt.ylabel('Counts')
    plt.figure(figsize=(4,3))
    df['Analysis on Polarity'].value_counts().plot(kind ='bar',color="#FF9F45")
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf)
    buf.seek(0)
    img1 = Image.open(buf)
    plt.clf() 
    
    # Generate subjectivity analysis plot
    print("Generating subjectivity analysis plot...")
    plt.figure(figsize=(4,3))
    df['Analysis on Subjectivity'].value_counts().plot(kind ='bar',color="#B667F1")
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf)
    buf.seek(0)
    img2 = Image.open(buf)
    plt.clf()
      
    # Generate word cloud
    print("Generating word cloud...")
    img3 = word_cloud_generator(Manifesto.name, text_Party_processed)
    
    # Generate frequency distribution and dispersion plots
    print("Generating frequency distribution...")
    fdist_Party = fDistance(text_Party_processed)
    img4 = fDistancePlot(text_Party_processed)
    
    print("Generating dispersion plot...")
    img5 = DispersionPlot(text_Party_processed)
    
    # Search for the term in the text
    print(f"Searching for term: {Search}")
    searChRes = get_all_phases_containing_tar_wrd(Search, text_Party_processed)
    searChRes = searChRes.replace(Search, "\u0332".join(Search))
    
    plt.close('all') 
    print("Analysis completed successfully")
    return searChRes, fdist_Party, img1, img2, img3, img4, img5, summary
    
  except Exception as e:
    error_message = f"Error analyzing manifesto: {str(e)}"
    print(error_message)
    import traceback
    traceback.print_exc()
    # Return placeholder values in case of error
    return error_message, {}, None, None, None, None, None, "Error generating summary. Please check the console for details."


Search_txt= "text"
filePdf = "file"
text = gr.Textbox(label='Context Based Search')
mfw=gr.Label(label="Most Relevant Topics (LLM Enhanced)")
plot1=gr.Image(label='Sentiment Analysis')
plot2=gr.Image(label='Subjectivity Analysis')
plot3=gr.Image(label='Word Cloud')
plot4=gr.Image(label='Frequency Distribution')
plot5=gr.Image(label='Dispersion Plot')
summary_output = gr.Textbox(label='AI-Generated Summary', lines=10)

with gr.Blocks(title='Manifesto Analysis') as demo:
    gr.Markdown("# Manifesto Analysis with LLM Enhancement")
    gr.Markdown("### Analyze political manifestos with advanced NLP and LLM techniques")
    
    with gr.Row():
        with gr.Column(scale=1):
            file_input = gr.File(label="Upload Manifesto PDF", file_types=[".pdf"])
            search_input = gr.Textbox(label="Search Term", placeholder="Enter a term to search in the manifesto")
            submit_btn = gr.Button("Analyze Manifesto")
    
    with gr.Tabs():
        with gr.TabItem("Summary"):
            summary_output
        
        with gr.TabItem("Search Results"):
            text
        
        with gr.TabItem("Key Topics"):
            mfw
        
        with gr.TabItem("Visualizations"):
            with gr.Row():
                with gr.Column(scale=1):
                    plot3
                with gr.Column(scale=1):
                    plot4
            
            with gr.Row():
                with gr.Column(scale=1):
                    plot1
                with gr.Column(scale=1):
                    plot2
            
            with gr.Row():
                plot5
    
    submit_btn.click(
        fn=analysis,
        inputs=[file_input, search_input],
        outputs=[text, mfw, plot1, plot2, plot3, plot4, plot5, summary_output]
    )
    
    # Add a debug print to verify the button is connected
    print("Button connected to analysis function")
    
    gr.Examples(
        examples=[
            ['Example/AAP_Manifesto_2019.pdf', 'government'],
            ['Example/Bjp_Manifesto_2019.pdf', 'environment'],
            ['Example/Congress_Manifesto_2019.pdf', 'safety']
        ],
        inputs=[file_input, search_input]
    )

demo.launch(debug=True, share=False, show_error=True)


# Old interface code replaced by the Blocks implementation above
# io=gr.Interface(fn=analysis, inputs=[filePdf,Search_txt], outputs=[text,mfw,plot1,plot2,plot3,plot4,plot5], title='Manifesto Analysis',examples=[['Example/AAP_Manifesto_2019.pdf','government'],['Example/Bjp_Manifesto_2019.pdf','environment'],['Example/Congress_Manifesto_2019.pdf','safety']])
# io.launch(debug=True,share=False)