Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,228 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import torch
|
3 |
+
from pinecone import Pinecone
|
4 |
+
from transformers import AutoTokenizer, AutoModel
|
5 |
+
import time
|
6 |
+
import requests
|
7 |
+
import json
|
8 |
+
|
9 |
+
# Page configuration
|
10 |
+
st.set_page_config(
|
11 |
+
page_title="Hukuki Döküman Arama (Detaylı Özet)",
|
12 |
+
page_icon="⚖️",
|
13 |
+
layout="wide",
|
14 |
+
initial_sidebar_state="expanded"
|
15 |
+
)
|
16 |
+
|
17 |
+
# App title and description
|
18 |
+
st.title("⚖️ Hukuki Döküman Semantik Arama Detaylı Özet")
|
19 |
+
st.markdown("Bu uygulama, 10.000 hukuki dökümanı içeren bir veritabanında semantik arama yapmanızı sağlar.")
|
20 |
+
|
21 |
+
# Initialize Pinecone connection
|
22 |
+
@st.cache_resource
|
23 |
+
def initialize_pinecone():
|
24 |
+
pinecone_client = Pinecone(api_key="pcsk_5s8hcC_2zwJTQthP5PSWE992iXmbRx6ykNQbnEWLhj3fDuR1Cw9eKRn31i2zsRyyCxCmgW")
|
25 |
+
return pinecone_client.Index("etikos2")
|
26 |
+
|
27 |
+
# Load the model and tokenizer
|
28 |
+
@st.cache_resource
|
29 |
+
def load_model():
|
30 |
+
model_name = "intfloat/multilingual-e5-large"
|
31 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
32 |
+
model = AutoModel.from_pretrained(model_name)
|
33 |
+
|
34 |
+
# Use GPU if available
|
35 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
36 |
+
model = model.to(device)
|
37 |
+
|
38 |
+
return tokenizer, model, device
|
39 |
+
|
40 |
+
# Function to get query embedding
|
41 |
+
def get_query_embedding(query_text, tokenizer, model):
|
42 |
+
# Prepare text with prefix required by e5 model
|
43 |
+
prefix = "query: "
|
44 |
+
query_text = prefix + query_text
|
45 |
+
|
46 |
+
# Tokenize
|
47 |
+
inputs = tokenizer(
|
48 |
+
query_text,
|
49 |
+
padding=True,
|
50 |
+
truncation=True,
|
51 |
+
return_tensors="pt",
|
52 |
+
max_length=1024
|
53 |
+
).to(model.device)
|
54 |
+
|
55 |
+
# Get embeddings
|
56 |
+
with torch.no_grad():
|
57 |
+
model_output = model(**inputs)
|
58 |
+
|
59 |
+
# Mean pooling
|
60 |
+
attention_mask = inputs['attention_mask']
|
61 |
+
token_embeddings = model_output[0]
|
62 |
+
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
|
63 |
+
embeddings = torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
|
64 |
+
|
65 |
+
# Normalize
|
66 |
+
embeddings = torch.nn.functional.normalize(embeddings, p=2, dim=1)
|
67 |
+
|
68 |
+
# Convert to list
|
69 |
+
embedding = embeddings[0].cpu().numpy().tolist()
|
70 |
+
return embedding
|
71 |
+
|
72 |
+
# Function to truncate text to a reasonable preview length
|
73 |
+
def get_text_preview(text, max_chars=1000):
|
74 |
+
if not text:
|
75 |
+
return "İçerik mevcut değil."
|
76 |
+
|
77 |
+
if len(text) <= max_chars:
|
78 |
+
return text
|
79 |
+
|
80 |
+
return text[:max_chars] + "..."
|
81 |
+
|
82 |
+
# Function to process query through Dify AI
|
83 |
+
def process_with_dify(query):
|
84 |
+
# Replace with your actual Dify API details
|
85 |
+
dify_api_key = "app-0UV1vRHHnChGssQ2Kc5UK9gg" # Replace with your actual API key
|
86 |
+
dify_api_endpoint = "https://api.dify.ai/v1/chat-messages" # Replace with your actual endpoint
|
87 |
+
|
88 |
+
headers = {
|
89 |
+
"Authorization": f"Bearer {dify_api_key}",
|
90 |
+
"Content-Type": "application/json"
|
91 |
+
}
|
92 |
+
|
93 |
+
payload = {
|
94 |
+
"inputs": {},
|
95 |
+
"query": f"{query}",
|
96 |
+
"response_mode": "blocking",
|
97 |
+
"user": "user"
|
98 |
+
}
|
99 |
+
|
100 |
+
try:
|
101 |
+
response = requests.post(dify_api_endpoint, headers=headers, json=payload)
|
102 |
+
if response.status_code == 200:
|
103 |
+
data = response.json()
|
104 |
+
return data.get("answer", "")
|
105 |
+
else:
|
106 |
+
st.warning(f"Dify AI ile iletişim kurulurken hata oluştu: {response.status_code}")
|
107 |
+
return ""
|
108 |
+
except Exception as e:
|
109 |
+
st.warning(f"Dify AI işlemi sırasında hata: {str(e)}")
|
110 |
+
return ""
|
111 |
+
|
112 |
+
# Sidebar configuration
|
113 |
+
st.sidebar.header("Arama Ayarları")
|
114 |
+
top_k = st.sidebar.slider("Gösterilecek sonuç sayısı:", 1, 30, 5)
|
115 |
+
preview_length = st.sidebar.slider("Ön izleme uzunluğu (karakter):", 500, 3000, 1000)
|
116 |
+
|
117 |
+
# Initialize resources with status indicators
|
118 |
+
with st.sidebar:
|
119 |
+
st.subheader("Sistem Durumu")
|
120 |
+
|
121 |
+
with st.status("Pinecone bağlantısı kuruluyor...", expanded=True) as status:
|
122 |
+
try:
|
123 |
+
index = initialize_pinecone()
|
124 |
+
status.update(label="Pinecone bağlantısı kuruldu ✅", state="complete", expanded=False)
|
125 |
+
except Exception as e:
|
126 |
+
status.update(label=f"Pinecone bağlantı hatası ❌: {str(e)}", state="error", expanded=True)
|
127 |
+
st.error("Veritabanına bağlanılamadı. Lütfen daha sonra tekrar deneyin.")
|
128 |
+
st.stop()
|
129 |
+
|
130 |
+
with st.status("Model yükleniyor...", expanded=True) as status:
|
131 |
+
try:
|
132 |
+
tokenizer, model, device = load_model()
|
133 |
+
status.update(label=f"Model yüklendi ✅ ({device.upper()} kullanılıyor)", state="complete", expanded=False)
|
134 |
+
except Exception as e:
|
135 |
+
status.update(label=f"Model yükleme hatası ❌: {str(e)}", state="error", expanded=True)
|
136 |
+
st.error("Model yüklenemedi. Lütfen daha sonra tekrar deneyin.")
|
137 |
+
st.stop()
|
138 |
+
|
139 |
+
# Main search interface
|
140 |
+
query = st.text_area("Aramak istediğiniz konuyu yazın:", height=100,
|
141 |
+
placeholder="Örnek: Mülkiyet hakkı ile ilgili davalar")
|
142 |
+
|
143 |
+
# Search button
|
144 |
+
search_button = st.button("🔍 Ara", type="primary", use_container_width=True)
|
145 |
+
|
146 |
+
# Execute search when button is clicked
|
147 |
+
if search_button and query:
|
148 |
+
# Process with Dify AI
|
149 |
+
with st.spinner("Sorgunuz Dify AI ile analiz ediliyor..."):
|
150 |
+
dify_output = process_with_dify(query)
|
151 |
+
|
152 |
+
if not dify_output:
|
153 |
+
st.error("Dify AI'dan yanıt alınamadı. Lütfen tekrar deneyin.")
|
154 |
+
st.stop()
|
155 |
+
|
156 |
+
# Show the query transformation
|
157 |
+
with st.expander("Sorgu Dönüşümü", expanded=True):
|
158 |
+
st.write("Orijinal sorgu:")
|
159 |
+
st.info(query)
|
160 |
+
st.write("Dify AI çıktısı (arama için kullanılacak):")
|
161 |
+
st.success(dify_output)
|
162 |
+
|
163 |
+
# Perform the search with ONLY the Dify AI output
|
164 |
+
with st.spinner("Arama yapılıyor..."):
|
165 |
+
try:
|
166 |
+
# Get query embedding from Dify AI output
|
167 |
+
start_time = time.time()
|
168 |
+
query_embedding = get_query_embedding(dify_output, tokenizer, model)
|
169 |
+
|
170 |
+
# Search Pinecone
|
171 |
+
search_results = index.query(
|
172 |
+
vector=query_embedding,
|
173 |
+
top_k=top_k,
|
174 |
+
include_metadata=True
|
175 |
+
)
|
176 |
+
|
177 |
+
elapsed_time = time.time() - start_time
|
178 |
+
|
179 |
+
# Display results
|
180 |
+
st.success(f"Arama tamamlandı! ({elapsed_time:.2f} saniye)")
|
181 |
+
|
182 |
+
if not search_results.matches:
|
183 |
+
st.info("Aramanıza uygun sonuç bulunamadı.")
|
184 |
+
else:
|
185 |
+
st.subheader(f"Arama Sonuçları ({len(search_results.matches)} döküman)")
|
186 |
+
|
187 |
+
# Display each result in a card
|
188 |
+
for i, match in enumerate(search_results.matches):
|
189 |
+
with st.container():
|
190 |
+
col1, col2 = st.columns([4, 1])
|
191 |
+
|
192 |
+
with col1:
|
193 |
+
st.markdown(f"### {i+1}. {match.metadata.get('daire', 'Bilinmeyen Daire')}")
|
194 |
+
|
195 |
+
with col2:
|
196 |
+
st.metric(label="Benzerlik", value=f"{match.score*100:.1f}%")
|
197 |
+
|
198 |
+
st.markdown("**Döküman Bilgileri:**")
|
199 |
+
st.markdown(f"""
|
200 |
+
- **Karar No:** {match.metadata.get('karar_no', 'Belirtilmemiş')}
|
201 |
+
- **Esas No:** {match.metadata.get('esas_no', 'Belirtilmemiş')}
|
202 |
+
- **Tarih:** {match.metadata.get('tarih', 'Belirtilmemiş')}
|
203 |
+
""")
|
204 |
+
|
205 |
+
# Get full text content from metadata
|
206 |
+
text_content = match.metadata.get('text', match.metadata.get('text_snippet', ''))
|
207 |
+
|
208 |
+
# Display text content in an expandable section
|
209 |
+
with st.expander("Döküman İçeriği", expanded=True):
|
210 |
+
st.markdown(get_text_preview(text_content, preview_length))
|
211 |
+
|
212 |
+
# Add download button if text content exists
|
213 |
+
if text_content:
|
214 |
+
st.download_button(
|
215 |
+
label="Tam Metni İndir",
|
216 |
+
data=text_content,
|
217 |
+
file_name=f"karar_{match.metadata.get('karar_no', 'bilinmeyen')}.txt",
|
218 |
+
mime="text/plain"
|
219 |
+
)
|
220 |
+
|
221 |
+
st.divider()
|
222 |
+
|
223 |
+
except Exception as e:
|
224 |
+
st.error(f"Arama sırasında bir hata oluştu: {str(e)}")
|
225 |
+
|
226 |
+
# Footer
|
227 |
+
st.sidebar.markdown("---")
|
228 |
+
st.sidebar.caption("©2025 Etikos AI")
|