Spaces:
Sleeping
Sleeping
Perfect7613
commited on
Delete main2.py
Browse files
main2.py
DELETED
@@ -1,288 +0,0 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
import google.generativeai as genai
|
3 |
-
from datetime import datetime
|
4 |
-
from dataclasses import dataclass
|
5 |
-
from typing import List, Dict, Optional, Tuple
|
6 |
-
import requests
|
7 |
-
import json
|
8 |
-
import os
|
9 |
-
from dotenv import load_dotenv
|
10 |
-
|
11 |
-
@dataclass
|
12 |
-
class Source:
|
13 |
-
"""Represents a source used for fact-checking."""
|
14 |
-
url: str
|
15 |
-
title: str
|
16 |
-
content: str
|
17 |
-
reputation_score: float
|
18 |
-
|
19 |
-
@dataclass
|
20 |
-
class FactCheckResult:
|
21 |
-
"""Represents the result of a fact check."""
|
22 |
-
claim: str
|
23 |
-
verdict: str
|
24 |
-
confidence_score: float
|
25 |
-
analysis_date: str
|
26 |
-
sources: List[Source]
|
27 |
-
evidence: List[Dict]
|
28 |
-
contradictions: List[Dict]
|
29 |
-
explanation: str
|
30 |
-
|
31 |
-
class GeminiFactChecker:
|
32 |
-
def __init__(self):
|
33 |
-
if not os.getenv("GOOGLE_API_KEY"):
|
34 |
-
raise ValueError("GOOGLE_API_KEY environment variable is required")
|
35 |
-
|
36 |
-
genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
|
37 |
-
generation_config = genai.types.GenerationConfig(
|
38 |
-
temperature=0.1,
|
39 |
-
top_p=0.8,
|
40 |
-
top_k=40,
|
41 |
-
)
|
42 |
-
|
43 |
-
self.model = genai.GenerativeModel(
|
44 |
-
model_name='gemini-1.5-pro',
|
45 |
-
generation_config=generation_config
|
46 |
-
)
|
47 |
-
self.search_api_key = os.getenv("SEARCH_API_KEY")
|
48 |
-
self.search_engine_id = os.getenv("SEARCH_ENGINE_ID")
|
49 |
-
self.jinai_api_key = os.getenv("JINA_AI_API_KEY")
|
50 |
-
self.jinai_reader_url = "https://r.jina.ai/"
|
51 |
-
|
52 |
-
def _search_sources(self, claim: str, num_sources: int = 3) -> List[str]:
|
53 |
-
try:
|
54 |
-
search_url = "https://www.googleapis.com/customsearch/v1"
|
55 |
-
params = {
|
56 |
-
'key': self.search_api_key,
|
57 |
-
'cx': self.search_engine_id,
|
58 |
-
'q': claim,
|
59 |
-
'num': num_sources
|
60 |
-
}
|
61 |
-
response = requests.get(search_url, params=params)
|
62 |
-
response.raise_for_status()
|
63 |
-
search_results = response.json()
|
64 |
-
return [item['link'] for item in search_results.get('items', [])]
|
65 |
-
except Exception as e:
|
66 |
-
print(f"Error searching sources: {str(e)}")
|
67 |
-
return []
|
68 |
-
|
69 |
-
def _fetch_webpage_content(self, url: str) -> Optional[dict]:
|
70 |
-
try:
|
71 |
-
headers = {
|
72 |
-
'Accept': 'application/json',
|
73 |
-
'Authorization': f'Bearer {self.jinai_api_key}'
|
74 |
-
}
|
75 |
-
response = requests.get(f"{self.jinai_reader_url}/{url}",
|
76 |
-
headers=headers,
|
77 |
-
timeout=10)
|
78 |
-
response.raise_for_status()
|
79 |
-
|
80 |
-
data = response.json()
|
81 |
-
if not data.get('data'):
|
82 |
-
return None
|
83 |
-
|
84 |
-
return {
|
85 |
-
"content": data['data'].get('content', '')[:5000],
|
86 |
-
"title": data['data'].get('title', ''),
|
87 |
-
"data": data['data']
|
88 |
-
}
|
89 |
-
except Exception as e:
|
90 |
-
print(f"Error fetching {url}: {str(e)}")
|
91 |
-
return None
|
92 |
-
|
93 |
-
def _analyze_evidence(self, claim: str, sources: List[Source]) -> List[Dict]:
|
94 |
-
all_evidence = []
|
95 |
-
|
96 |
-
for source in sources:
|
97 |
-
prompt = f"""
|
98 |
-
Analyze this content and return evidence as JSON array:
|
99 |
-
|
100 |
-
CLAIM: "{claim}"
|
101 |
-
SOURCE TITLE: {source.title}
|
102 |
-
CONTENT: {source.content[:2000]}
|
103 |
-
|
104 |
-
Return array of evidence objects with properties:
|
105 |
-
- text: exact quote or clear paraphrase
|
106 |
-
- type: "supporting" or "contradicting"
|
107 |
-
- relevance: number 0.0 to 1.0
|
108 |
-
- source: source title
|
109 |
-
"""
|
110 |
-
|
111 |
-
try:
|
112 |
-
response = self.model.generate_content(prompt)
|
113 |
-
if response.text:
|
114 |
-
clean_text = response.text.strip()
|
115 |
-
if clean_text.startswith('```json'):
|
116 |
-
clean_text = clean_text[7:-3]
|
117 |
-
elif clean_text.startswith('[') and clean_text.endswith(']'):
|
118 |
-
clean_text = clean_text
|
119 |
-
|
120 |
-
evidence_list = json.loads(clean_text)
|
121 |
-
for evidence in evidence_list:
|
122 |
-
evidence["source_score"] = source.reputation_score
|
123 |
-
all_evidence.extend(evidence_list)
|
124 |
-
|
125 |
-
except Exception as e:
|
126 |
-
print(f"Error analyzing source {source.url}: {str(e)}")
|
127 |
-
continue
|
128 |
-
|
129 |
-
return all_evidence
|
130 |
-
|
131 |
-
def check_fact(self, claim: str, num_sources: int = 3) -> Optional[FactCheckResult]:
|
132 |
-
try:
|
133 |
-
urls = self._search_sources(claim, num_sources)
|
134 |
-
if not urls:
|
135 |
-
return None
|
136 |
-
|
137 |
-
sources = []
|
138 |
-
for url in urls:
|
139 |
-
content_dict = self._fetch_webpage_content(url)
|
140 |
-
if content_dict:
|
141 |
-
sources.append(Source(
|
142 |
-
url=url,
|
143 |
-
title=content_dict.get("title", url),
|
144 |
-
content=content_dict["content"],
|
145 |
-
reputation_score=0.8 # Default score
|
146 |
-
))
|
147 |
-
|
148 |
-
if not sources:
|
149 |
-
return None
|
150 |
-
|
151 |
-
evidence = self._analyze_evidence(claim, sources)
|
152 |
-
|
153 |
-
supporting = [e for e in evidence if e["type"] == "supporting"]
|
154 |
-
contradicting = [e for e in evidence if e["type"] == "contradicting"]
|
155 |
-
|
156 |
-
total_support = sum(
|
157 |
-
float(e.get("relevance", 0.5)) * float(e.get("source_score", 1))
|
158 |
-
for e in supporting
|
159 |
-
)
|
160 |
-
|
161 |
-
total_contradiction = sum(
|
162 |
-
float(e.get("relevance", 0.5)) * float(e.get("source_score", 1))
|
163 |
-
for e in contradicting
|
164 |
-
)
|
165 |
-
|
166 |
-
if not evidence:
|
167 |
-
verdict = "Insufficient evidence"
|
168 |
-
confidence = 0.0
|
169 |
-
explanation = "No evidence found from analyzed sources."
|
170 |
-
else:
|
171 |
-
support_ratio = total_support / (total_support + total_contradiction) if (total_support + total_contradiction) > 0 else 0
|
172 |
-
confidence = max(support_ratio, 1 - support_ratio)
|
173 |
-
|
174 |
-
if support_ratio > 0.6:
|
175 |
-
verdict = "Likely True" if confidence >= 0.7 else "Somewhat True"
|
176 |
-
elif support_ratio < 0.4:
|
177 |
-
verdict = "Likely False" if confidence >= 0.7 else "Somewhat False"
|
178 |
-
else:
|
179 |
-
verdict = "Inconclusive"
|
180 |
-
|
181 |
-
explanation = f"Based on {len(supporting)} supporting and {len(contradicting)} contradicting pieces of evidence."
|
182 |
-
|
183 |
-
return FactCheckResult(
|
184 |
-
claim=claim,
|
185 |
-
verdict=verdict,
|
186 |
-
confidence_score=confidence,
|
187 |
-
analysis_date=datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
188 |
-
sources=sources,
|
189 |
-
evidence=supporting,
|
190 |
-
contradictions=contradicting,
|
191 |
-
explanation=explanation
|
192 |
-
)
|
193 |
-
|
194 |
-
except Exception as e:
|
195 |
-
print(f"Error during fact checking: {str(e)}")
|
196 |
-
return None
|
197 |
-
|
198 |
-
def format_fact_check_report(result: FactCheckResult) -> str:
|
199 |
-
report = f"""# Fact Check Report
|
200 |
-
|
201 |
-
## Claim
|
202 |
-
"{result.claim}"
|
203 |
-
|
204 |
-
## Verdict: {result.verdict}
|
205 |
-
Confidence Score: {result.confidence_score:.2f}
|
206 |
-
|
207 |
-
## Explanation
|
208 |
-
{result.explanation}
|
209 |
-
|
210 |
-
## Analysis Summary
|
211 |
-
- Number of sources analyzed: {len(result.sources)}
|
212 |
-
- Supporting evidence found: {len(result.evidence)}
|
213 |
-
- Contradicting points found: {len(result.contradictions)}
|
214 |
-
|
215 |
-
## Sources Analyzed
|
216 |
-
"""
|
217 |
-
for source in result.sources:
|
218 |
-
report += f"- [{source.title}]({source.url}) (Credibility: {source.reputation_score:.2f})\n"
|
219 |
-
|
220 |
-
if result.evidence:
|
221 |
-
report += "\n### Supporting Evidence:\n"
|
222 |
-
for e in result.evidence[:3]:
|
223 |
-
report += f"- {e['text']} (Source: {e['source']})\n"
|
224 |
-
|
225 |
-
if result.contradictions:
|
226 |
-
report += "\n### Contradicting Points:\n"
|
227 |
-
for c in result.contradictions[:3]:
|
228 |
-
report += f"- {c['text']} (Source: {c['source']})\n"
|
229 |
-
|
230 |
-
return report
|
231 |
-
|
232 |
-
def main():
|
233 |
-
load_dotenv()
|
234 |
-
fact_checker = GeminiFactChecker()
|
235 |
-
|
236 |
-
with gr.Blocks() as demo:
|
237 |
-
gr.Markdown("# AI-Powered Fact Checker")
|
238 |
-
gr.Markdown("Enter a claim to check its veracity against multiple sources.")
|
239 |
-
|
240 |
-
with gr.Row():
|
241 |
-
with gr.Column():
|
242 |
-
claim = gr.Textbox(
|
243 |
-
label="Claim to Check",
|
244 |
-
placeholder="Enter the claim you want to verify...",
|
245 |
-
lines=3
|
246 |
-
)
|
247 |
-
num_sources = gr.Slider(
|
248 |
-
label="Number of Sources to Check",
|
249 |
-
minimum=1,
|
250 |
-
maximum=5,
|
251 |
-
value=3,
|
252 |
-
step=1
|
253 |
-
)
|
254 |
-
check_button = gr.Button("Check Claim", variant="primary")
|
255 |
-
|
256 |
-
with gr.Column():
|
257 |
-
status = gr.Markdown("Ready to check claims...")
|
258 |
-
report = gr.Markdown()
|
259 |
-
|
260 |
-
def check_fact_wrapper(claim: str, num_sources: int):
|
261 |
-
status_value = "🔍 Searching and analyzing sources..."
|
262 |
-
yield status_value, ""
|
263 |
-
|
264 |
-
try:
|
265 |
-
result = fact_checker.check_fact(claim, int(num_sources))
|
266 |
-
if result:
|
267 |
-
status_value = "✅ Analysis complete!"
|
268 |
-
report_value = format_fact_check_report(result)
|
269 |
-
else:
|
270 |
-
status_value = "❌ Error occurred"
|
271 |
-
report_value = "Error occurred during fact checking."
|
272 |
-
except Exception as e:
|
273 |
-
status_value = "❌ Error occurred"
|
274 |
-
report_value = f"Error: {str(e)}"
|
275 |
-
|
276 |
-
yield status_value, report_value
|
277 |
-
|
278 |
-
check_button.click(
|
279 |
-
fn=check_fact_wrapper,
|
280 |
-
inputs=[claim, num_sources],
|
281 |
-
outputs=[status, report],
|
282 |
-
show_progress=True
|
283 |
-
)
|
284 |
-
|
285 |
-
demo.launch()
|
286 |
-
|
287 |
-
if __name__ == "__main__":
|
288 |
-
main()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|