File size: 58,923 Bytes
bceeb47 dc57ca1 bceeb47 eddf12a bceeb47 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 |
import re
import os
import faiss
import whisper
import ffmpeg
import tempfile
import requests
import numpy as np
import pandas as pd
import streamlit as st
from openai import OpenAI
from transformers import pipeline
from sentence_transformers import SentenceTransformer
from newsplease import NewsPlease
from streamlit_echarts import st_echarts
from streamlit_option_menu import option_menu
# NEWS to check
# https://fbe.unimelb.edu.au/newsroom/fake-news-in-the-age-of-covid-19 True Claim
# https://newssalutebenessere.altervista.org/covid-19-just-a-simple-flue-or-something-else/ False Claim
###### CONFIGURATIONS ######
# Debug mode
debug = False
# File paths
embeddings_file = r"./data/abstract_embeddings.npy"
pmid_file = r"./data/pmids.npy"
faiss_index_file = r"./data/faiss_index.index"
file_path = r'./data/parte_205.csv'
# Initialize OpenAI API client
client = OpenAI(
base_url="https://integrate.api.nvidia.com/v1",
api_key=st.secrets.nvidia
)
# Load data
data = pd.read_csv(file_path)
# Load the model
model = SentenceTransformer('all-MiniLM-L6-v2')
def get_article_data(url):
"""
Extracts article data from a specified URL.
Args:
url (str): URL of the article to analyze.
Returns:
dict: Structured article data, including: title, authors, publication date, and content.
"""
try:
# Make an HTTP request to the specified URL
response = requests.get(url)
# Check if the request was successful (i.e., status code 200)
response.raise_for_status()
# Extract the HTML content from the response
html_content = response.text
# Use NewsPlease to extract structured data from the HTML content
article = NewsPlease.from_html(html_content, url=url)
# Return the structured article data
return {
"title": article.title,
"authors": article.authors,
"date_publish": article.date_publish,
"content": article.maintext,
}
except requests.exceptions.RequestException as e:
return {"error": f"Error during URL retrieval: {e}"}
except Exception as e:
return {"error": f"Error processing the article: {e}"}
def extract_and_split_claims(claims):
"""
Extracts and splits claims from a given string.
Args:
claims (str): String containing claims.
Returns:
dict: Dictionary containing the extracted claims.
"""
start_index = claims.find("Claim 1:")
if start_index != -1:
claims = claims[start_index:]
claim_lines = claims.strip().split("\n\n")
claims_dict = {}
for i, claim in enumerate(claim_lines, start=1):
claims_dict[f"Claim_{i}"] = claim
for var_name, claim_text in claims_dict.items():
globals()[var_name] = claim_text
return claims_dict
def extract_label_and_score(result):
"""
Extracts the predicted label and score from the result string.
Args:
result (str): String containing the prediction result.
Returns:
tuple: Predicted label and score.
"""
# Extract the predicted label
label_match = re.search(r"'labels': \['(.*?)'", result)
predicted_label = label_match.group(1) if label_match else None
# Extract the score
score_match = re.search(r"'scores': \[(\d+\.\d+)", result)
score_label = float(score_match.group(1)) if score_match else None
return predicted_label, score_label
def clean_phrases(phrases, pattern):
"""
Clean and extract phrases from a list of strings using a specified pattern.
Args:
phrases (list): List of strings containing phrases.
pattern (str): Regular expression pattern to extract phrases.
Returns:
list: List of cleaned phrases as dictionaries with text and abstract keys
"""
cleaned_phrases = []
for phrase in phrases:
matches = re.findall(pattern, phrase)
cleaned_phrases.extend([{"text": match[0], "abstract": f"abstract_{match[1]}"} for match in matches])
return cleaned_phrases
def highlight_phrases(abstract_text, phrases, color, label):
"""
Highlight phrases in the abstract text with the specified background color.
Args:
abstract_text (str): Text of the abstract to highlight.
phrases (list): List of phrases to highlight.
color (str): Background color to use for highlighting.
label (str): Predicted label for the claim.
Returns:
str: Abstract text with highlighted phrases.
"""
# Switch colors if the label is "False"
if label.lower() == "false":
color = "lightgreen" if color == "red" else color
# Highlight each phrase in the abstract text
for phrase in phrases:
abstract_text = re.sub(
re.escape(phrase["text"]),
f'<span style="background-color: {color}; font-weight: bold; border: 1px solid black; border-radius: 5px;">{phrase["text"]}</span>',
abstract_text,
flags=re.IGNORECASE
)
return abstract_text
def parse_response(response):
"""
Parse the response from the model and extract the fields.
Args:
response (str): Response string from the model.
Returns:
tuple: Extracted fields from the response.
"""
# Initial values for the fields
first_label = "Non trovato"
justification = "Non trovato"
supporting = "Non trovato"
refusing = "Non trovato"
notes = "Non trovato"
# Regular expression patterns for extracting fields
patterns = {
"first_label": r"Label:\s*(.*?)\n",
"justification": r"Justification:\s*(.*?)(?=\nSupporting sentences)",
"supporting": r"Supporting sentences from abstracts:\n(.*?)(?=\nRefusing sentences)",
"refusing": r"Refusing sentences from abstracts:\n(.*?)(?=\nNote:)",
"notes": r"Note:\s*(.*)"
}
# Extract the fields using regular expressions
if match := re.search(patterns["first_label"], response, re.DOTALL):
first_label = match.group(1).strip()
if match := re.search(patterns["justification"], response, re.DOTALL):
justification = match.group(1).strip()
if match := re.search(patterns["supporting"], response, re.DOTALL):
supporting = [{"text": sentence.strip(), "abstract": f"abstract_{i+1}"} for i, sentence in enumerate(match.group(1).strip().split('\n'))]
if match := re.search(patterns["refusing"], response, re.DOTALL):
refusing = [{"text": sentence.strip(), "abstract": f"abstract_{i+1}"} for i, sentence in enumerate(match.group(1).strip().split('\n'))]
if match := re.search(patterns["notes"], response, re.DOTALL):
notes = match.group(1).strip()
# Return the extracted fields
return first_label, justification, supporting, refusing, notes
def load_embeddings(embeddings_file, pmid_file, faiss_index_file, debug=False):
"""
Load embeddings, PMIDs, and FAISS index from the specified files.
Args:
embeddings_file (str): File path for the embeddings.
pmid_file (str): File path for the PMIDs.
faiss_index_file (str): File path for the FAISS index.
Returns:
tuple: Tuple containing the embeddings, PMIDs, and FAISS index.
"""
# Check if the files exist
if not (os.path.exists(embeddings_file) and os.path.exists(pmid_file) and os.path.exists(faiss_index_file)):
raise FileNotFoundError("One or more files not found. Please check the file paths.")
# Load the embeddings and PMIDs
embeddings = np.load(embeddings_file)
pmids = np.load(pmid_file, allow_pickle=True)
# Load the FAISS index
index = faiss.read_index(faiss_index_file)
if debug:
print("Embeddings, PMIDs, and FAISS index loaded successfully.")
return embeddings, pmids, index
def retrieve_top_abstracts(claim, model, index, pmids, data, top_k=5):
"""
Retrieve the top abstracts from the FAISS index for a given claim.
Args:
claim (str): Claim to fact-check.
model (SentenceTransformer): Sentence transformer model for encoding text.
index (faiss.IndexFlatIP): FAISS index for similarity search.
pmids (np.ndarray): Array of PMIDs for the abstracts.
data (pd.DataFrame): DataFrame containing the abstract data.
top_k (int): Number of top abstracts to retrieve.
Returns:
list: List of tuples containing the abstract text, PMID, and distance.
"""
# Encode the claim using the SentenceTransformer model
claim_embedding = model.encode([claim])
faiss.normalize_L2(claim_embedding) # Normalize the claim embedding (with L2 norm)
distances, indices = index.search(claim_embedding, top_k)
# Retrieve the top abstracts based on the indices
results = []
for j, i in enumerate(indices[0]):
pmid = pmids[i]
abstract_text = data[data['PMID'] == pmid]['AbstractText'].values[0]
distance = distances[0][j]
results.append((abstract_text, pmid, distance))
return results
def generate_justification(query, justification):
"""
Generate a justification for the claim using the Zero-Shot Classification model.
Args:
query (str): Claim to fact-check.
justification (str): Justification for the claim.
Returns:
str: Final justification for the claim.
"""
# Define the classes for the Zero-Shot Classification model
Class = ["True", "False","NEI"]
# Generate the justification text
justification_text = (
f'Justification: "{justification}"'
)
# Limit the justification text to a maximum length
max_length = 512
if len(justification_text) > max_length:
justification_text = justification_text[:max_length]
# Generate the final justification using the Zero-Shot Classification model
output = zeroshot_classifier(
query,
Class,
hypothesis_template=f"The claim is '{{}}' for: {justification_text}",
multi_label=False
)
# Prepare the final justification text
final_justification = f'{output}.'
return final_justification
def llm_reasoning_template(query):
"""
Generate a template for the prompt used for justification generation by the LLM model.
Args:
query (str): Claim to fact-check.
Returns:
str: Reasoning template for the claim.
"""
llm_reasoning_prompt = f"""<<SYS>> [INST]
You are a helpful, respectful and honest Doctor. Always answer as helpfully as possible using the context text provided.
Use the information in Context.
Elaborate the Context to generate a new information.
Use only the knowledge in Context to answer.
Answer describing in a scentific way. Be formal during the answer. Use the third person.
Answer without mentioning the Context. Use it but don't refer to it in the text.
To answer, use max 300 word.
Create a Justification from the sentences given.
Use the structure: Justification: The claim is (label) because... (don't use the word "context")
Write as an online doctor to create the Justification.
After, give some sentences from Context from scientific papers: that supports the label and reject the label.
Supporting sentences from abstracts:
information sentence from abstract_1:
information sentence from abstract_2:
..
Refusing sentences from abstracts:
information sentence from abstract_1:
information sentence from abstract_2:
..
Add where it comes from (abstract_1, abstract_2, abstract_3, abstract_4, abstract_5)
With the answer, gives a line like: "Label:". Always put Label as first. After Label, give the Justification.
The justification will be always given as Justification:
Label can be yes, no, NEI, where yes: claim is true. no: claim is false. NEI: not enough information.
The Label will be chosen with a voting system of support/refuse before.
[/INST] <</SYS>>
[INST] Question: {query} [/INST]
[INST] Context from scientific papers:
"""
return llm_reasoning_prompt
def claim_detection_template(full_text):
"""
Generate a template for the prompt used for claim detection by the LLM model.
Args:
full_text (str): Full text to analyze.
Returns:
str: Template for claim detection.
"""
claim_detection_prompt = f"""<<SYS>> [INST]
Your task is to extract from the text potential health related question to verify their veracity.
The context extracted from the online where to take the claim is: {full_text}
Create simple claim of single sentence from the context.
Dont's use *
Give just the claim. Don't write other things.
Extract only health related claim.
Rank eventual claim like:
Claim 1:
Claim 2:
Claim 3:
Use always this structure.
Start every claim with "Claim " followed by the number.
The number of claims may go from 1 to a max of 5.
The claims have to be always health related. [/INST] <</SYS>>
"""
return claim_detection_prompt
# Page and Title Configuration
st.set_page_config(page_title="CER - Combining Evidence and Reasoning Demo", layout="wide", initial_sidebar_state="collapsed")
st.markdown("<h1 style='text-align: center; color: inherit;'>✔️✨ CER - Biomedical Fact Checker</h1>", unsafe_allow_html=True)
# Horizontal option menu for selecting the page
page = option_menu(None, ["Single claim check", "Page check", "Video check"],
icons=['check', 'ui-checks'],
menu_icon="cast", default_index=0, orientation="horizontal")
# Sidebar Configuration
st.sidebar.title("🔬 Combining Evidence and Reasoning Demo")
st.sidebar.caption("🔍 Fact-check biomedical claims using scientific evidence and reasoning.")
st.sidebar.markdown("---")
st.sidebar.caption("#### ℹ️ About")
st.sidebar.caption("This is a demo application for fact-checking biomedical claims using scientific evidence and reasoning. It uses a combination of language models, scientific literature, and reasoning to provide explanations for the predictions.")
# Load embeddings, PMIDs, and FAISS index
if 'embeddings_loaded' not in st.session_state:
embeddings, pmids, index = load_embeddings(embeddings_file, pmid_file, faiss_index_file, debug)
st.session_state.embeddings = embeddings
st.session_state.pmids = pmids
st.session_state.index = index
st.session_state.embeddings_loaded = True
else:
embeddings = st.session_state.embeddings
pmids = st.session_state.pmids
index = st.session_state.index
# Check if the claim and top_abstracts are in the session state
if 'claim' not in st.session_state:
st.session_state.claim = ""
if 'top_abstracts' not in st.session_state:
st.session_state.top_abstracts = []
#### Single claim check PAGE ####
if page == "Single claim check":
st.subheader("Single claim check")
st.caption("✨ Enter a single claim to fact-check and hit the button to see the results! 🔍")
st.session_state.claim = st.text_input("Claim to fact-check:")
if st.button("✨ Fact Check"):
if st.session_state.claim:
# Retrieve the top abstracts for the claim
top_abstracts = retrieve_top_abstracts(st.session_state.claim, model, index, pmids, data, top_k=5)
st.session_state.top_abstracts = top_abstracts
st.markdown("### **Results**")
with st.container():
for i, (abstract, pmid, distance) in enumerate(st.session_state.top_abstracts, 1):
pubmed_url = f"https://pubmed.ncbi.nlm.nih.gov/{pmid}/"
globals()[f"abstract_{i}"] = abstract
globals()[f"reference_{i}"] = pubmed_url
globals()[f"distance_{i}"] = distance
with st.spinner('🔍 We are checking...'):
try:
# Retrieve the question from the DataFrame
query = st.session_state.claim
# Generate the reasoning template
prompt_template = llm_reasoning_template(query)
# Add the abstracts to the prompt
for i in range(1, len(st.session_state.top_abstracts)):
prompt_template += f"{globals()[f'abstract_{i}']} ; "
prompt_template += f"{globals()[f'abstract_{i+1}']} [/INST]"
# Call the API
completion = client.chat.completions.create(
model="meta/llama-3.1-405b-instruct",
messages=[{"role": "user", "content": prompt_template}],
temperature=0.1,
top_p=0.7,
max_tokens=1024,
stream=True
)
# Collect the response
answer = ""
for chunk in completion:
if chunk.choices[0].delta.content:
answer += chunk.choices[0].delta.content
# Debug: Check the answer
if debug:
print(f"{answer}")
except Exception as e:
st.write(f"Error processing index: {e}")
with st.spinner('🤔💬 Justifying the check...'):
# Perform parsing and separate variables
zeroshot_classifier = pipeline(
"zero-shot-classification", model="MoritzLaurer/deberta-v3-large-zeroshot-v1.1-all-33"
)
first_label, justification, supporting, refusing, notes = parse_response(answer)
with st.spinner('🕵️♂️📜 We are finding evidence...'):
# Generate the justification for the claim
result = generate_justification(st.session_state.claim, justification)
predicted_label, score_label = extract_label_and_score(result)
if predicted_label == "True":
color = f"rgba(0, 204, 0, {score_label})" # Green
elif predicted_label == "False":
color = f"rgba(204, 0, 0, {score_label})" # Red
elif predicted_label == "NEI":
color = f"rgba(255, 255, 0, {score_label})" # Yellow
else:
color = "black" # Default color
# Calculate the confidence score
confidence = f"{score_label * 100:.2f}%"
st.caption(f"📝 The Claim: {st.session_state.claim}")
st.markdown(
f"**Prediction of claim:** Most likely <span style='color: {color}; font-weight: bold;'>{predicted_label}</span> with a confidence of <span style='color: {color}; font-weight: bold;'>{confidence}</span>",
unsafe_allow_html=True
)
st.markdown("### **Justification**")
st.markdown(f'<p> {justification}</p>', unsafe_allow_html=True)
# Extract the abstracts and references
abstracts = {}
for i in range(1, len(st.session_state.top_abstracts) + 1):
abstracts[f"abstract_{i}"] = globals()[f"abstract_{i}"]
pattern = r'"\s*(.*?)\s*"\s*\(abstract_(\d+)\)'
supporting_texts = []
for item in supporting:
try:
supporting_texts.append(item["text"])
except (TypeError, KeyError):
continue
supporting = clean_phrases(supporting_texts, pattern)
refusing_text = []
for item in refusing:
try:
refusing_text.append(item["text"])
except (TypeError, KeyError):
continue
refusing = clean_phrases(refusing_text, pattern)
if debug:
print(supporting)
print(refusing)
processed_abstracts = {}
for abstract_name, abstract_text in abstracts.items():
# Highlight supporting phrases in green
supporting_matches = [phrase for phrase in supporting if phrase["abstract"] == abstract_name]
abstract_text = highlight_phrases(abstract_text, supporting_matches, "lightgreen", predicted_label)
# Highlight refusing phrases in red
refusing_matches = [phrase for phrase in refusing if phrase["abstract"] == abstract_name]
abstract_text = highlight_phrases(abstract_text, refusing_matches, "red", predicted_label)
# Add only if supporting matches are found
if supporting_matches:
# Add the reference if a corresponding variable exists
reference_variable = f"reference_{abstract_name.split('_')[1]}"
if reference_variable in globals():
reference_value = globals()[reference_variable]
abstract_text += f"<br><br><strong>🔗 Reference:</strong> {reference_value}"
# Add the processed abstract
processed_abstracts[abstract_name] = abstract_text
# Iterate over the processed abstracts and remove duplicates
seen_contents = set() # Set to track already seen contents
evidence_counter = 1
# Display the results of the processed abstracts with numbered expanders
st.markdown("### **Scientific Evidence**")
# Add a legend for the colors
legend_html = """
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: lightgreen; margin-right: 10px; border-radius: 5px;"></div>
<div>Positive Evidence</div>
</div>
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: red; margin-right: 10px; border-radius: 5px;"></div>
<div>Negative Evidence</div>
</div>
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: yellow; margin-right: 10px; border-radius: 5px;"></div>
<div>Dubious Evidence</div>
</div>
</div>
"""
col1, col2 = st.columns([0.8, 0.2])
with col1:
if processed_abstracts:
tabs = st.tabs([f"Scientific Evidence {i}" for i in range(1, len(processed_abstracts) + 1)])
for tab, (name, content) in zip(tabs, processed_abstracts.items()):
if content not in seen_contents: # Check for duplicates
seen_contents.add(content)
with tab:
# Switch colors if the label is "False"
if predicted_label.lower() == "false":
content = content.replace("background-color: lightgreen", "background-color: tempcolor")
content = content.replace("background-color: red", "background-color: lightgreen")
content = content.replace("background-color: tempcolor", "background-color: red")
# Use `st.write` to display HTML directly
st.write(content, unsafe_allow_html=True)
else:
st.markdown("No relevant Scientific Evidence found")
with col2:
st.caption("Legend")
st.markdown(legend_html, unsafe_allow_html=True)
#### Web page check PAGE ####
elif page == "Page check":
st.subheader("Page check")
st.caption("✨ Enter a URL to fact-check the health-related claims on the page and hit the button to see the results! 🔍")
url = st.text_input("URL to fact-check:")
if st.button("✨ Fact Check") and url:
st.session_state.true_count = 0
st.session_state.false_count = 0
st.session_state.nei_count = 0
with st.spinner('🌐🔍 Extracting claims...'):
article_data = get_article_data(url)
try:
# Retrieve the claims from the article data
prompt_template = claim_detection_template(article_data)
# Call the API
completion = client.chat.completions.create(
model="meta/llama-3.1-405b-instruct",
messages=[{"role": "user", "content": prompt_template}],
temperature=0.1,
top_p=0.7,
max_tokens=1024,
stream=True
)
# Collect the response
answer = ""
for chunk in completion:
if chunk.choices[0].delta.content:
answer += chunk.choices[0].delta.content
# Debug: Controlla la risposta
print(f"{answer}")
except Exception as e:
print(f"Error {e}")
claims_dict = extract_and_split_claims(answer)
# Display the extracted claims
st.markdown("### **Claims Extracted**")
st.caption("🔍 Here are the health-related claims extracted from the page:")
cols = st.columns(3)
for i, (claim_key, claim_text) in enumerate(claims_dict.items(), 1):
col = cols[(i - 1) % 3]
with col.expander(f"Claim {i} 📝", expanded=True):
st.write(claim_text)
# Display the results for the extracted claims
st.markdown("### **Results**")
st.caption("🔍 Here are the results for the extracted claims:")
for claim_key, claim_text in claims_dict.items():
st.session_state.claim = claim_text
if st.session_state.claim:
top_abstracts = retrieve_top_abstracts(st.session_state.claim, model, index, pmids, data, top_k=5)
st.session_state.top_abstracts = top_abstracts # Salva i risultati
with st.expander(f"✔️ **Results for {claim_key}**", expanded=True):
for i, (abstract, pmid, distance) in enumerate(st.session_state.top_abstracts, 1):
pubmed_url = f"https://pubmed.ncbi.nlm.nih.gov/{pmid}/"
globals()[f"abstract_{i}"] = abstract
globals()[f"reference_{i}"] = pubmed_url
globals()[f"distance_{i}"] = distance
with st.spinner('🔍 We are checking...'):
try:
# Retrieve the question from the DataFrame
query = st.session_state.claim
# Generate the reasoning template
prompt_template = llm_reasoning_template(query)
# Add the abstracts to the prompt
for i in range(1, len(st.session_state.top_abstracts)):
prompt_template += f"{globals()[f'abstract_{i}']} ; "
prompt_template += f"{globals()[f'abstract_{i+1}']} [/INST]"
# Call the API
completion = client.chat.completions.create(
model="meta/llama-3.1-405b-instruct",
messages=[{"role": "user", "content": prompt_template}],
temperature=0.1,
top_p=0.7,
max_tokens=1024,
stream=True
)
# Collect the response
answer = ""
for chunk in completion:
if chunk.choices[0].delta.content:
answer += chunk.choices[0].delta.content
# Debug: Check the answer
if debug:
print(f"{answer}")
except Exception as e:
st.write(f"Error processing index: {e}")
with st.spinner('🤔💬 Justifying the check...'):
# Perform parsing and separate variables
zeroshot_classifier = pipeline(
"zero-shot-classification", model="MoritzLaurer/deberta-v3-large-zeroshot-v1.1-all-33"
)
first_label, justification, supporting, refusing, notes = parse_response(answer)
with st.spinner('🕵️♂️📜 We are finding evidence...'):
# Generate the justification for the claim
result = generate_justification(st.session_state.claim, justification)
predicted_label, score_label = extract_label_and_score(result)
# Update the counts based on the predicted label
if predicted_label == "True":
color = f"rgba(0, 204, 0, {score_label})" # Green
st.session_state.true_count += 1
elif predicted_label == "False":
color = f"rgba(204, 0, 0, {score_label})" # Red
st.session_state.false_count += 1
elif predicted_label == "NEI":
color = f"rgba(255, 255, 0, {score_label})" # Yellow
st.session_state.nei_count += 1
else:
color = "black" # Default color
confidence = f"{score_label * 100:.2f}%"
st.caption(f"📝 The Claim: {st.session_state.claim}")
st.markdown(
f"**Prediction of claim:** Most likely <span style='color: {color}; font-weight: bold;'>{predicted_label}</span> with a confidence of <span style='color: {color}; font-weight: bold;'>{confidence}</span>",
unsafe_allow_html=True
)
st.markdown("### **Justification**")
st.markdown(f'<p> {justification}</p>', unsafe_allow_html=True)
abstracts = {}
for i in range(1, len(st.session_state.top_abstracts) + 1):
abstracts[f"abstract_{i}"] = globals()[f"abstract_{i}"]
pattern = r'"\s*(.*?)\s*"\s*\(abstract_(\d+)\)'
supporting_texts = []
for item in supporting:
try:
supporting_texts.append(item["text"])
except (TypeError, KeyError):
continue
supporting = clean_phrases(supporting_texts, pattern)
refusing_text = []
for item in refusing:
try:
refusing_text.append(item["text"])
except (TypeError, KeyError):
continue
refusing = clean_phrases(refusing_text, pattern)
if debug:
print(supporting)
print(refusing)
processed_abstracts = {}
for abstract_name, abstract_text in abstracts.items():
# Highlight supporting phrases in green
supporting_matches = [phrase for phrase in supporting if phrase["abstract"] == abstract_name]
abstract_text = highlight_phrases(abstract_text, supporting_matches, "lightgreen", predicted_label)
# Highlight refusing phrases in red
refusing_matches = [phrase for phrase in refusing if phrase["abstract"] == abstract_name]
abstract_text = highlight_phrases(abstract_text, refusing_matches, "red", predicted_label)
# Add only if supporting matches are found
if supporting_matches:
# Add the reference if a corresponding variable exists
reference_variable = f"reference_{abstract_name.split('_')[1]}"
if reference_variable in globals():
reference_value = globals()[reference_variable]
abstract_text += f"<br><br><strong>🔗 Reference:</strong> {reference_value}"
# Add the processed abstract
processed_abstracts[abstract_name] = abstract_text
# Iterate over the processed abstracts and remove duplicates
seen_contents = set() # Set to track already seen contents
evidence_counter = 1
# Display the results of the processed abstracts with numbered expanders
st.markdown("### **Scientific Evidence**")
# Add a legend for the colors
legend_html = """
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: lightgreen; margin-right: 10px; border-radius: 5px;"></div>
<div>Positive Evidence</div>
</div>
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: red; margin-right: 10px; border-radius: 5px;"></div>
<div>Negative Evidence</div>
</div>
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: yellow; margin-right: 10px; border-radius: 5px;"></div>
<div>Dubious Evidence</div>
</div>
</div>
"""
col1, col2 = st.columns([0.8, 0.2])
with col1:
if processed_abstracts:
tabs = st.tabs([f"Scientific Evidence {i}" for i in range(1, len(processed_abstracts) + 1)])
for tab, (name, content) in zip(tabs, processed_abstracts.items()):
if content not in seen_contents: # Check for duplicates
seen_contents.add(content)
with tab:
# Switch colors if the label is "False"
if predicted_label.lower() == "false":
content = content.replace("background-color: lightgreen", "background-color: tempcolor")
content = content.replace("background-color: red", "background-color: lightgreen")
content = content.replace("background-color: tempcolor", "background-color: red")
# Use `st.write` to display HTML directly
st.write(content, unsafe_allow_html=True)
else:
st.markdown("No relevant Scientific Evidence found")
with col2:
st.caption("Legend")
st.markdown(legend_html, unsafe_allow_html=True)
st.markdown("### **Page Summary**")
st.caption("📊 Here is a summary of the results for the extracted claims:")
# Labels and Colors
labels = ['True', 'False', 'NEI']
colors = ['green', 'red', 'yellow']
# Sizes of the pie chart
sizes = [
st.session_state.true_count,
st.session_state.false_count,
st.session_state.nei_count
]
# Configure the Pie Chart Options
options = {
"tooltip": {"trigger": "item"},
"legend": {"top": "5%", "left": "center"},
"series": [
{
"name": "Document Status",
"type": "pie",
"radius": ["40%", "70%"],
"avoidLabelOverlap": False,
"itemStyle": {
"borderRadius": 10,
"borderColor": "#fff",
"borderWidth": 2,
},
"label": {"show": True, "position": "center"},
"emphasis": {
"label": {"show": True, "fontSize": "20", "fontWeight": "bold"}
},
"labelLine": {"show": False},
"data": [
{"value": sizes[0], "name": labels[0], "itemStyle": {"color": colors[0]}},
{"value": sizes[1], "name": labels[1], "itemStyle": {"color": colors[1]}},
{"value": sizes[2], "name": labels[2], "itemStyle": {"color": colors[2]}},
],
}
],
}
# Display the Pie Chart
st1, st2 = st.columns([0.6, 0.4])
with st1:
st.markdown("#### The page is :")
true_count = st.session_state.true_count
false_count = st.session_state.false_count
nei_count = st.session_state.nei_count
if true_count > 0 and false_count == 0:
reliability = '<span style="color: darkgreen; font-weight: bold;">Highly Reliable</span>'
elif true_count > false_count:
reliability = '<span style="color: lightgreen; font-weight: bold;">Fairly Reliable</span>'
elif true_count == 0:
reliability = '<span style="color: darkred; font-weight: bold;">Strongly Considered Unreliable</span>'
elif false_count > true_count:
reliability = '<span style="color: lightcoral; font-weight: bold;">Unlikely to be Reliable</span>'
elif (true_count == false_count) or (nei_count > true_count and nei_count > false_count and true_count != 0 and false_count != 0):
reliability = '<span style="color: yellow; font-weight: bold;">NEI</span>'
else:
reliability = '<span style="color: black; font-weight: bold;">Completely Reliable</span>'
st.markdown(f"The page is considered {reliability} because it contains {true_count} true claims, {false_count} false claims, and {nei_count} claims with not enough information.", unsafe_allow_html=True)
with st.popover("ℹ️ Understanding the Truthfulness Ratings"):
st.markdown("""
The reliability of the page is determined based on the number of true and false claims extracted from the page.
- If the page contains only true claims, it is considered **Highly Reliable**.
- If the page has more true claims than false claims, it is considered **Fairly Reliable**.
-If the page has more false claims than true claims, it is considered **Unlikely to be Reliable**.
- If the page contains only false claims, it is considered **Strongly Considered Unreliable**.
- If the page has an equal number of true and false claims, it is considered **NEI**.
""")
with st2:
st_echarts(
options=options, height="500px",
)
#### Video check PAGE ####
elif page == "Video check":
st.subheader("Video claim check")
st.caption("✨ Upload a video to fact-check and hit the button to see the results! 🔍")
video = st.file_uploader("Choose a video...", type=["mp4"])
video_box, text_box = st.columns([0.6, 0.4])
if video is not None:
with video_box:
with st.expander("▶️ See uploaded video", expanded=False):
st.video(video)
if st.button("✨ Fact Check") and video is not None:
with st.spinner('🎥🔄 Processing video...'):
# Save the video to a temporary file
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as temp_video:
temp_video.write(video.read())
temp_video_path = temp_video.name
# Extract the audio from the video
temp_audio_path = tempfile.NamedTemporaryFile(delete=False, suffix=".wav").name
ffmpeg.input(temp_video_path).output(temp_audio_path, acodec="pcm_s16le", ar=16000, ac=1).run(overwrite_output=True)
# Transcribe the audio
model1 = whisper.load_model("small")
result = model1.transcribe(temp_audio_path)
# Extract the final text
transcribed_text = result["text"]
with text_box:
with st.expander("📝 Transcribed Text", expanded=False):
st.caption("🔍 Here is the transcribed text from the uploaded video:")
container = st.container(height=322)
container.write(transcribed_text)
st.session_state.true_count = 0
st.session_state.false_count = 0
st.session_state.nei_count = 0
with st.spinner('🌐🔍 Extracting claims from video...'):
try:
# Retrieve the claims from the video
prompt_template = claim_detection_template(transcribed_text)
# Call the API
completion = client.chat.completions.create(
model="meta/llama-3.1-405b-instruct",
messages=[{"role": "user", "content": prompt_template}],
temperature=0.1,
top_p=0.7,
max_tokens=1024,
stream=True
)
# Collect the response
answer = ""
for chunk in completion:
if chunk.choices[0].delta.content:
answer += chunk.choices[0].delta.content
# Debug: Check the answer
if debug:
print(f"{answer}")
except Exception as e:
print(f"Error {e}")
claims_dict = extract_and_split_claims(answer)
# Display the extracted claims
st.markdown("### **Claims Extracted**")
st.caption("🔍 Here are the health-related claims extracted from the video:")
cols = st.columns(3)
for i, (claim_key, claim_text) in enumerate(claims_dict.items(), 1):
col = cols[(i - 1) % 3]
with col.expander(f"Claim {i} 📝", expanded=True):
st.write(claim_text)
# Display the results for the extracted claims
st.markdown("### **Results**")
st.caption("🔍 Here are the results for the extracted claims:")
for claim_key, claim_text in claims_dict.items():
st.session_state.claim = claim_text
if st.session_state.claim:
top_abstracts = retrieve_top_abstracts(st.session_state.claim, model, index, pmids, data, top_k=5)
st.session_state.top_abstracts = top_abstracts # Salva i risultati
with st.expander(f"✔️ **Results for {claim_key}**", expanded=True):
for i, (abstract, pmid, distance) in enumerate(st.session_state.top_abstracts, 1):
pubmed_url = f"https://pubmed.ncbi.nlm.nih.gov/{pmid}/"
globals()[f"abstract_{i}"] = abstract
globals()[f"reference_{i}"] = pubmed_url
globals()[f"distance_{i}"] = distance
with st.spinner('🔍 We are checking...'):
try:
# Retrieve the question from the DataFrame
query = st.session_state.claim
# Generate the reasoning template
prompt_template = llm_reasoning_template(query)
# Add the abstracts to the prompt
for i in range(1, len(st.session_state.top_abstracts)):
prompt_template += f"{globals()[f'abstract_{i}']} ; "
prompt_template += f"{globals()[f'abstract_{i+1}']} [/INST]"
# Call the API
completion = client.chat.completions.create(
model="meta/llama-3.1-405b-instruct",
messages=[{"role": "user", "content": prompt_template}],
temperature=0.1,
top_p=0.7,
max_tokens=1024,
stream=True
)
# Collect the response
answer = ""
for chunk in completion:
if chunk.choices[0].delta.content:
answer += chunk.choices[0].delta.content
# Debug: Check the answer
if debug:
print(f"{answer}")
except Exception as e:
st.write(f"Error processing index: {e}")
with st.spinner('🤔💬 Justifying the check...'):
# Perform parsing and separate variables
zeroshot_classifier = pipeline(
"zero-shot-classification", model="MoritzLaurer/deberta-v3-large-zeroshot-v1.1-all-33"
)
first_label, justification, supporting, refusing, notes = parse_response(answer)
with st.spinner('🕵️♂️📜 We are finding evidence...'):
# Generate the justification for the claim
result = generate_justification(st.session_state.claim, justification)
predicted_label, score_label = extract_label_and_score(result)
# Update the counts based on the predicted label
if predicted_label == "True":
color = f"rgba(0, 204, 0, {score_label})" # Green
st.session_state.true_count += 1
elif predicted_label == "False":
color = f"rgba(204, 0, 0, {score_label})" # Red
st.session_state.false_count += 1
elif predicted_label == "NEI":
color = f"rgba(255, 255, 0, {score_label})" # Yellow
st.session_state.nei_count += 1
else:
color = "black" # Default color
confidence = f"{score_label * 100:.2f}%"
st.caption(f"📝 The Claim: {st.session_state.claim}")
st.markdown(
f"**Prediction of claim:** Most likely <span style='color: {color}; font-weight: bold;'>{predicted_label}</span> with a confidence of <span style='color: {color}; font-weight: bold;'>{confidence}</span>",
unsafe_allow_html=True
)
st.markdown("### **Justification**")
st.markdown(f'<p> {justification}</p>', unsafe_allow_html=True)
abstracts = {}
for i in range(1, len(st.session_state.top_abstracts) + 1):
abstracts[f"abstract_{i}"] = globals()[f"abstract_{i}"]
pattern = r'"\s*(.*?)\s*"\s*\(abstract_(\d+)\)'
supporting_texts = []
for item in supporting:
try:
supporting_texts.append(item["text"])
except (TypeError, KeyError):
continue
supporting = clean_phrases(supporting_texts, pattern)
refusing_text = []
for item in refusing:
try:
refusing_text.append(item["text"])
except (TypeError, KeyError):
continue
refusing = clean_phrases(refusing_text, pattern)
processed_abstracts = {}
for abstract_name, abstract_text in abstracts.items():
# Highlight supporting phrases in green
supporting_matches = [phrase for phrase in supporting if phrase["abstract"] == abstract_name]
abstract_text = highlight_phrases(abstract_text, supporting_matches, "lightgreen", predicted_label)
# Highlight refusing phrases in red
refusing_matches = [phrase for phrase in refusing if phrase["abstract"] == abstract_name]
abstract_text = highlight_phrases(abstract_text, refusing_matches, "red", predicted_label)
if supporting_matches:
# Add the reference if a corresponding variable exists
reference_variable = f"reference_{abstract_name.split('_')[1]}"
if reference_variable in globals():
reference_value = globals()[reference_variable]
abstract_text += f"<br><br><strong>🔗 Reference:</strong> {reference_value}"
# Add the processed abstract
processed_abstracts[abstract_name] = abstract_text
# Iterate over the processed abstracts and remove duplicates
seen_contents = set() # Set to track already seen contents
evidence_counter = 1
# Display the results of the processed abstracts with numbered expanders
st.markdown("### **Scientific Evidence**")
# Add a legend for the colors
legend_html = """
<div style="display: flex; flex-direction: column; align-items: flex-start;">
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: lightgreen; margin-right: 10px; border-radius: 5px;"></div>
<div>Positive Evidence</div>
</div>
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: red; margin-right: 10px; border-radius: 5px;"></div>
<div>Negative Evidence</div>
</div>
<div style="display: flex; align-items: center; margin-bottom: 5px;">
<div style="width: 20px; height: 20px; background-color: yellow; margin-right: 10px; border-radius: 5px;"></div>
<div>Dubious Evidence</div>
</div>
</div>
"""
col1, col2 = st.columns([0.8, 0.2])
with col1:
if processed_abstracts:
tabs = st.tabs([f"Scientific Evidence {i}" for i in range(1, len(processed_abstracts) + 1)])
for tab, (name, content) in zip(tabs, processed_abstracts.items()):
if content not in seen_contents: # Check for duplicates
seen_contents.add(content)
with tab:
# Switch colors if the label is "False"
if predicted_label.lower() == "false":
content = content.replace("background-color: lightgreen", "background-color: tempcolor")
content = content.replace("background-color: red", "background-color: lightgreen")
content = content.replace("background-color: tempcolor", "background-color: red")
# Use `st.write` to display HTML directly
st.write(content, unsafe_allow_html=True)
else:
st.markdown("No relevant Scientific Evidence found")
with col2:
st.caption("Legend")
st.markdown(legend_html, unsafe_allow_html=True)
st.markdown("### **Video Summary**")
st.caption("📊 Here is a summary of the results for the extracted claims:")
# Labels and Colors
labels = ['True', 'False', 'NEI']
colors = ['green', 'red', 'yellow']
# Sizes of the pie chart
sizes = [
st.session_state.true_count,
st.session_state.false_count,
st.session_state.nei_count
]
# Configure the Pie Chart Options
options = {
"tooltip": {"trigger": "item"},
"legend": {"top": "5%", "left": "center"},
"series": [
{
"name": "Document Status",
"type": "pie",
"radius": ["40%", "70%"],
"avoidLabelOverlap": False,
"itemStyle": {
"borderRadius": 10,
"borderColor": "#fff",
"borderWidth": 2,
},
"label": {"show": True, "position": "center"},
"emphasis": {
"label": {"show": True, "fontSize": "20", "fontWeight": "bold"}
},
"labelLine": {"show": False},
"data": [
{"value": sizes[0], "name": labels[0], "itemStyle": {"color": colors[0]}},
{"value": sizes[1], "name": labels[1], "itemStyle": {"color": colors[1]}},
{"value": sizes[2], "name": labels[2], "itemStyle": {"color": colors[2]}},
],
}
],
}
# Display the Pie Chart
st1, st2 = st.columns([0.6, 0.4])
with st1:
st.markdown("#### The Video is :")
true_count = st.session_state.true_count
false_count = st.session_state.false_count
nei_count = st.session_state.nei_count
if true_count > 0 and false_count == 0:
reliability = '<span style="color: darkgreen; font-weight: bold;">Highly Reliable</span>'
elif true_count > false_count:
reliability = '<span style="color: lightgreen; font-weight: bold;">Fairly Reliable</span>'
elif true_count == 0:
reliability = '<span style="color: darkred; font-weight: bold;">Strongly Considered Unreliable</span>'
elif false_count > true_count:
reliability = '<span style="color: lightcoral; font-weight: bold;">Unlikely to be Reliable</span>'
elif (true_count == false_count) or (nei_count > true_count and nei_count > false_count and true_count != 0 and false_count != 0):
reliability = '<span style="color: yellow; font-weight: bold;">NEI</span>'
else:
reliability = '<span style="color: black; font-weight: bold;">Completely Reliable</span>'
st.markdown(f"The video is considered {reliability} because it contains {true_count} true claims, {false_count} false claims, and {nei_count} claims with not enough information.", unsafe_allow_html=True)
with st.popover("ℹ️ Understanding the Truthfulness Ratings"):
st.markdown("""
The reliability of the video is determined based on the number of true and false claims extracted from the video.
- If the video contains only true claims, it is considered **Highly Reliable**.
- If the video has more true claims than false claims, it is considered **Fairly Reliable**.
- If the video has more false claims than true claims, it is considered **Unlikely to be Reliable**.
- If the video contains only false claims, it is considered **Strongly Considered Unreliable**.
- If the video has an equal number of true and false claims, it is considered **NEI**.
""")
with st2:
st_echarts(
options=options, height="500px",
)
|