File size: 23,126 Bytes
0d5bda6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a0c88a6
 
0d5bda6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a0c88a6
 
 
0d5bda6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
import torch
import torch.nn as nn
from transformers import PegasusForConditionalGeneration, PegasusTokenizer, AutoTokenizer,AutoModelForSequenceClassification
from scipy.special import softmax
from tqdm.notebook import tqdm
from sklearn.metrics import accuracy_score, f1_score, confusion_matrix
from rouge_score import rouge_scorer
from rouge import Rouge
import streamlit as st
from ydata_profiling import ProfileReport
from streamlit_pandas_profiling import st_profile_report



import time
import io
import os
import pprint
from IPython.display import HTML
import traceback
import logging
import random
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
import tensorflow as tf

st.set_page_config(page_title="Review Summary App", page_icon=None, layout="centered", initial_sidebar_state="auto", menu_items=None)
# st.set_page_config(layout="wide")
st.title("Review Summarizer App")
st.write("This app summarises all the reviews of a product")


@st.cache_resource#(allow_output_mutation=True)
def load_roberta_model_and_tokenizer(model_name):
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSequenceClassification.from_pretrained(model_name)
    return tokenizer, model

@st.cache_resource#(allow_output_mutation=True)
def load_pegasus_model_and_tokenizer(model_name):
    tokenizer = PegasusTokenizer.from_pretrained(model_name)
    model = PegasusForConditionalGeneration.from_pretrained(model_name)
    return tokenizer, model

# =========================================================
# Define a function to assign labels based on star rating
# =========================================================
def assign_star_label(row):
    return 'positive' if row['star_rating'] > 3 else 'negative'

def showEda(df):
    pr = ProfileReport(df, explorative=True)
    st.header('**Pandas Profiling Report**')
    st_profile_report(pr) 
    

def dataset_load():

    with st.spinner("Importing modules................"):
        time.sleep(2)
    st.success("Imported Modules")    



    # ===================================================================================================================
    # ================================================= UTILITY FUNCTIONS ===============================================
    # ===================================================================================================================
    with st.spinner("Initialising methods ............"):
        # ==================
        # Load & Clean Data
        # ================== 
        @st.cache_data
        def data_load_clean_df():
            df = pd.read_csv('./amazon_reviews_us_Mobile_Electronics_v1_00.csv', on_bad_lines='skip')
        #     df = df.loc[df['product_id'].isin(['B00J46XO9U'])]
            df = df[['customer_id','product_title','star_rating','review_body','product_id']]
            df[~df.duplicated(subset='review_body')] #Remove duplicates
            df = df.apply(lambda row: row[df['star_rating'].isin(['1','2','3','4','5'])]) # Remove date fields inside star_rating
            df['star_rating']=df['star_rating'].astype('int64') # Convert data type for star_rating
            df['star_rating_label'] = df.apply(assign_star_label, axis=1) # Apply the function to create the 'label' column
            df['review_body'] = df['review_body'].apply(lambda x : str(x)) # Convert text inputs to STRING
            df['review_body'] = df['review_body'].apply(lambda x : x[:512]) # Limit length of string
            return df.reset_index(drop=True)


        # ======================
        # Assign Polarity Score
        # ======================    
        def polarity_scores_roberta(review):
            encoded_text = roberta_tokenizer(review, return_tensors='pt').to(device)
            with torch.no_grad():
                output = roberta_model(**encoded_text)
        #     scores = output[0][0].detach().numpy() # FOR CPU
            scores = softmax(output.logits.detach().cpu().numpy()) # CONVERT from GPU to CPU
            scores = softmax(scores[0])
            scores_dict = {
                'roberta_negative' : scores[0],
                'roberta_positive' : scores[1]
            }
            return scores_dict

        # ==================
        # Summarising Text
        # ================== 
        def text_summarizer(review):
            batch = pegasus_tokenizer(review, truncation=True, padding="longest", max_length=1024, return_tensors="pt").to(device)
            with torch.no_grad():
                translated = pegasus_model.generate(**batch)
                #translated = pegasus_model.module.generate(**batch) #When using Data Parallel
            tgt_text = pegasus_tokenizer.batch_decode(translated, skip_special_tokens=True)
            summary_dict = {"summary":tgt_text[0]}
            return summary_dict

        # =================
        # Rouge Score Check
        # =================   
        def rouge_score_viewer(original_text,generated_summary):
            # Create a Rouge object
            rouge = Rouge()

            # Calculate ROUGE scores
            scores = rouge.get_scores(generated_summary, original_text)

            # Print ROUGE scores
            return {"Rouge-1":scores[0]['rouge-1'],"Rouge-2":scores[0]['rouge-2'],"Rouge-L":scores[0]['rouge-l']}

        # =======================================================    
        # Define a function to assign labels based on star rating
        # =======================================================
        def assign_label(row):
            if row['roberta_positive'] > row['roberta_negative']:
                return 'positive'
            else:
                return 'negative'

        # =======================================================    
        # Summarise bunch of summaries together
        # =======================================================
        @st.cache_data
        def data_summarizer(df, marker, summary_count):
            summaries = []
            marker   = 'positive' if marker==1 else 'negative'
            df_new   = df[(df['star_rating_label']==marker) & (df['roberta_rating_label']==marker)]
            df_new = df_new[~df_new.duplicated(subset=["review_body","summary"])]
            sentence = df_new.sort_values(['roberta_positive','Rouge_1','Rouge_2','Rouge_L'],ascending=[False, False,False,False])['summary'].reset_index(drop=True) if marker==1 else df_new.sort_values(['roberta_negative','Rouge_1','Rouge_2','Rouge_L'],ascending=[False, False,False,False])['summary'].reset_index(drop=True)
            print(sentence)
            print(f"Sentence len :{len(sentence)}")
            count=0
            for i in range(0,len(sentence),10):
                if(count==summary_count):
                    break
                else:
                    chunk = sentence[i:i + 10]
                    joined_sentence = ' '.join(chunk)
                    print(f"JOINED SENTENCE :{joined_sentence}\n\n\n")
                    summaries.append(text_summarizer(joined_sentence[:512])["summary"])
                    count+=1
            print(f"SUMMARY IS:{summaries}\n")
            return summaries
        
        
        # ==========================================================   
        # Convert the array to a markdown string with bullet points
        # ==========================================================   
        def bullet_markdown(array):
            return "\n".join(f"- {item}" for item in array)
        
        # ==========================================================   
        # Get rows with same rating labels
        # ==========================================================        
        
        def getMatchCols(df,value):
            marker = "positive" if value == 1 else "negative"
            df_new = df[(df['star_rating_label']==marker) & (df['roberta_rating_label']==marker)]
            if df_new.shape[0]>0:
                return df_new.sort_values(['roberta_positive','Rouge_1','Rouge_2','Rouge_L'],ascending=[False,False,False,False])['review_body'].values
            else:
                return [f"No {marker} reviews available"]

    # =========================================================================================================================
    # ================================================= LOADING OF THE DATA ===================================================
    # =========================================================================================================================

    ## Load & Clean Data
    with st.spinner("Loading the data ............"):
#         st.header("Loaded Dataframe")
        df = data_load_clean_df()
    
    loaded_df = df.copy()
    # Controlling the sidebar for loaded DF and new DF with selected product
    ProductDataframeCheck = False    

    # TODO : Limit for demonstration only. Less rows to be analysed later
#     df = df.groupby('product_id').filter(lambda x: (len(x) <= 5)).reset_index(drop=True)
    st.header("The Dataframe loaded is shown below :")
    with st.spinner("Loading the data ............"):
        st.dataframe(df)
    # =========================================================================================================================
    # ================================================= LIST OF ALL PRODUCTS ==================================================
    # =========================================================================================================================
    with st.spinner("Loading list of products ............"):
        time.sleep(2)
        prod_ids = df['product_id'].unique()

    # =========================================================================================================================
    # ================================================= CHOOSE A PRODUCT ======================================================
    # =========================================================================================================================

    # Create a dual slider to select the range of product ids to display
    st.markdown("---")
    st.subheader("Step 0 : Choose a product")    
    
    
    # Group the dataframe by product_id and count the number of rows for each product_id
    grouped_df = df.groupby("product_id").size().reset_index(name="count")
    
#     st.dataframe(grouped_df)

    # Find the product_id with the maximum number of rows and store it in max_rows
    max_rows = grouped_df["count"].max()

    # Create a slider in streamlit with min value as 0, and max value as max_rows
#     slider_value = st.slider("Select the number of rows", min_value=1, max_value=max_rows)
    slider_value = st.select_slider("Select the number of rows", options=sorted(grouped_df['count'].unique()),value=max(grouped_df['count']))

    # Filter the grouped dataframe by the slider value and get the product_id column as a list
    filtered_df = grouped_df[grouped_df["count"] == slider_value]["product_id"].tolist()

    # Create a select box in streamlit with the filtered list of product_id
    st.write(f"There are {len(filtered_df)} products with {slider_value} rows")
    selected_product_id = st.selectbox("Select the product_id", filtered_df)
    
    

    preview_df = df.loc[df['product_id']==selected_product_id].reset_index(drop=True)
    
    if(not preview_df.empty):
        prod_name = preview_df['product_title'][0]

        # Display the selected product id
        st.markdown("---")
        st.subheader("Step 1 : Product Details :")
        st.write(f'Product Name : {prod_name}')
        st.write(f'Product ID   : {selected_product_id} ')
        st.write(f'Total Rows   : {preview_df.shape[0]}')
        device = "cuda" if torch.cuda.is_available() else "cpu"
        st.write(f"Selected device for processing is (CPU/GPU) : {device.upper()}")
    
    #================================================================
    # Use the condition to control the display of the radio buttons
    #================================================================
    if(not preview_df.empty):
        ProductDataframeCheck = True
    
    if (not ProductDataframeCheck):
        option = st.sidebar.radio("Select an option", ["None","Show EDA"])
    else:
        option = st.sidebar.radio("Select an option", ["None","Show EDA", "Product EDA"])
        
    if(option=="Show EDA"):
        showEda(loaded_df)
    elif option=="Product EDA":
        showEda(preview_df)


    
    
    if st.button('Confirm Product'):
        df = df.loc[df['product_id']==selected_product_id].reset_index(drop=True)
        st.markdown("---")
        st.subheader("Step 2 : Dataframe with chosen product :")
        st.dataframe(df)
#         st.success(f"Dataframe loaded with product_id:{selected_product_id}")
    #     st.write(f"Selected product is {selected_product_id}, named as \"{df['product_title']}\" with dataframe having {df.shape[0]} rows")
    
        df_rows = df.shape[0]

        # =========================================================================================================================
        # ================================================ PRE-TRAINED MODEL ======================================================
        # =========================================================================================================================
        st.markdown("---")
        st.subheader("Step 3 : Initialising the models & running operation")
        #with st.spinner("Initializing RoBERTa Model ............"):
            #device = "cuda" if torch.cuda.is_available() else "cpu"
            #st.write(f"Selected device for processing is (CPU/GPU) : {device.upper()}")

        # ROBERTA Model
        with st.spinner("Initializing RoBERTa Model ............"):
#             roberta_model_name = f"siebert/sentiment-roberta-large-english" 
#             roberta_tokenizer = AutoTokenizer.from_pretrained(roberta_model_name)
#             roberta_model = AutoModelForSequenceClassification.from_pretrained(roberta_model_name).to(device)

            roberta_model_name = "siebert/sentiment-roberta-large-english"
            roberta_tokenizer, roberta_model = load_roberta_model_and_tokenizer(roberta_model_name)
            roberta_model.to(device)

        
        # PEGASUS Model
        with st.spinner("Initializing Pegasus Model ............"):
#             pegasus_model_name = "google/pegasus-large"
#             pegasus_tokenizer = PegasusTokenizer.from_pretrained(pegasus_model_name)
#             pegasus_model = PegasusForConditionalGeneration.from_pretrained(pegasus_model_name).to(device)
            pegasus_model_name = "google/pegasus-large"
            pegasus_tokenizer, pegasus_model = load_pegasus_model_and_tokenizer(pegasus_model_name)
            pegasus_model.to(device)

        st.success("Models successfully loaded")
        # =========================================================================================================================
        # ================================================ RUN MODEL ON DATA ======================================================
        # =========================================================================================================================

        # Sentimental Analysis & Text Summarisation

        res = {}
        summaries = {}
        rouge_1 = {}
        rouge_2 = {}
        rouge_L = {}
        broken_ids = []

        with st.spinner("Operation in progress ............"):

            progress_bar_analysis = st.progress((0/len(df))*100, text="Please wait......... 0%")

            progress_percent = 0
            progress_text = f"Please wait......... {float(progress_percent):.2f}%"


            for i, row in tqdm(df.iterrows(), total=len(df)):

                progress_percent = (i/len(df))*100
                progress_text = f"Please wait......... {progress_percent:.2f}%"
                progress_bar_analysis.progress(int(progress_percent+1), text=progress_text)


                # Process Sentimental Analysis
                text = row['review_body']
                myid = row['customer_id']


                roberta_result = polarity_scores_roberta(text)
                both = {**roberta_result}
                res[myid] = both

                # Process Summaries
                summary_result = text_summarizer(text)
                summaries[myid] = {**summary_result}

                #Rouge SCore
                original_text      = row['review_body']
                generated_summary  = summary_result['summary']
                rouge_scores       = rouge_score_viewer(original_text,generated_summary)
                rouge_1[myid]={"rouge-1":rouge_scores['Rouge-1']['f']}
                rouge_2[myid]={"rouge-2":rouge_scores['Rouge-2']['f']}
                rouge_L[myid]={"rouge-L":rouge_scores['Rouge-L']['f']}
            progress_bar_analysis.progress(int(100), text="Completed......... 100%")
        st.success("Operation Completed")

        with st.spinner("Merging in progress ............"):

            # Merge dataframes
            results_df = pd.DataFrame(res).T    
            results_df['summary'] = (pd.DataFrame(summaries).T)['summary'].values #Add summary column
            results_df['Rouge_1'] = pd.DataFrame(rouge_1).T[:].values
            results_df['Rouge_2'] = pd.DataFrame(rouge_2).T[:].values
            results_df['Rouge_L'] = pd.DataFrame(rouge_L).T[:].values
            results_df = results_df.reset_index().rename(columns={'index': 'customer_id'})
            results_df = results_df.merge(df, how='left')

            results_df['roberta_rating_label'] = results_df.apply(assign_label, axis=1) # Apply the function to create the 'label' column
            st.markdown("---")
            st.subheader("Step 4 : Dataframe after operation")
#             st.dataframe(results_df)
#         st.success("Merge Completed")


        with st.spinner("Matching Columns in progress ............"):
            # prod_a = results_df.loc[results_df['product_id']=='B00J46XO9U']
            prod_a = results_df.copy()
            prod_a = prod_a[prod_a['star_rating_label'] == prod_a['roberta_rating_label']]
            prod_a.reset_index(drop=True)

#         st.success("Matching columns Completed")

#         st.header("Dataframe with matching labels")
        st.dataframe(prod_a)
        
        # =========================================================================================================================
        # ============================================= HISTOGRAM CHECK ======================================================
        # =========================================================================================================================        
        
        # # Create a histogram using matplotlib
        # plt.figure(figsize=(8, 6))
        # plt.hist(prod_a['Rouge_1'], bins=30, alpha=0.7, color='blue')  # Adjust bins and color as needed
        # plt.title('Histogram of Random Data')
        # plt.xlabel('Values')
        # plt.ylabel('Frequency')
        # plt.grid(True)
        # plt.show()
        
        # # Create a histogram using matplotlib
        # plt.figure(figsize=(8, 6))
        # plt.hist(prod_a['Rouge_2'], bins=30, alpha=0.7, color='blue')  # Adjust bins and color as needed
        # plt.title('Histogram of Random Data')
        # plt.xlabel('Values')
        # plt.ylabel('Frequency')
        # plt.grid(True)
        # plt.show()
        
        # # Create a histogram using matplotlib
        # plt.figure(figsize=(8, 6))
        # plt.hist(prod_a['Rouge_L'], bins=30, alpha=0.7, color='blue')  # Adjust bins and color as needed
        # plt.title('Histogram of Random Data')
        # plt.xlabel('Values')
        # plt.ylabel('Frequency')
        # plt.grid(True)
        # plt.show()


        # =========================================================================================================================
        # ============================================= CHECKING THE METRICS ======================================================
        # =========================================================================================================================
        
        # RUN only if NUMBER OF ROWS > 4
        if(df_rows>4):
            with st.spinner("Creating confusion matrix ............"):
                st.markdown("---")
                st.subheader("Step 5. - Confusion Matrix")
                # Sample confusion matrix (replace this with your actual data)
                conf_df = results_df.copy()
                actual_labels = conf_df['star_rating_label']
                predicted_labels = conf_df['roberta_rating_label']

                # Create the confusion matrix
                cm_a = confusion_matrix(actual_labels, predicted_labels)

                # Display the confusion matrix using seaborn
                st.set_option('deprecation.showPyplotGlobalUse', False)
                sns.heatmap(cm_a, annot=True, fmt='d')
                st.pyplot()


                # Extract true positives, false positives, false negatives, true negatives
                tn, fp, fn, tp = cm_a.ravel()

                # Calculate accuracy
                accuracy = accuracy_score(actual_labels, predicted_labels)

                # Calculate precision, recall, and F1 score
                precision = tp / (tp + fp)
                recall = tp / (tp + fn)
                f1 = 2 * (precision * recall) / (precision + recall)

                st.write(f"Accuracy :{accuracy*100:.2f} | Precision :{precision:.2f} | Recall:{recall:.2f} | F1-Score:{f1:.2f}")

        # =========================================================================================================================
        # ============================================= SUMMARRY OF PRODUCT =======================================================
        # =========================================================================================================================
        st.markdown("---")
        st.subheader("Step 6 : Summary of product")
        choice = 10#st.number_input("Choose number of summaries", 0, 10)

        # POSITIVE SUMMARIES
        st.header("Positive Reviews Summary")
        if(df_rows<=10):
            st.markdown(bullet_markdown(getMatchCols(prod_a,1)))
        else:
            with st.spinner("Generating Positive Summaries ............"):
                sum_list_pos = data_summarizer(prod_a,1,choice)
                st.markdown(bullet_markdown(sum_list_pos))

        # NEGATIVE SUMMARIES 
        st.header("Negative Reviews Summary")
        if(df_rows<=10):
            st.markdown(bullet_markdown(getMatchCols(prod_a,0)))
        else:
            with st.spinner("Generating Negative Summaries ............"):
                sum_list_neg =data_summarizer(prod_a,0,choice)
                st.markdown(bullet_markdown(sum_list_neg))
                
                
dataset_load()