Spaces:
Running
Running
Create semantic_agent_interaction.py
Browse files
modules/semantic/semantic_agent_interaction.py
ADDED
@@ -0,0 +1,404 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# modules/semantic/semantic_agent_interaction.py
|
2 |
+
import os
|
3 |
+
import anthropic
|
4 |
+
import streamlit as st
|
5 |
+
import logging
|
6 |
+
import time
|
7 |
+
import json
|
8 |
+
from datetime import datetime, timezone
|
9 |
+
from io import BytesIO
|
10 |
+
import base64
|
11 |
+
|
12 |
+
# Local imports
|
13 |
+
from ..utils.widget_utils import generate_unique_key
|
14 |
+
from ..database.semantic_mongo_db import store_semantic_interaction
|
15 |
+
|
16 |
+
logger = logging.getLogger(__name__)
|
17 |
+
|
18 |
+
# Cache for conversation history to avoid redundant API calls
|
19 |
+
conversation_cache = {}
|
20 |
+
|
21 |
+
def get_conversation_cache_key(text, metrics, graph_data, lang_code):
|
22 |
+
"""
|
23 |
+
Generate a cache key for conversations based on analysis data.
|
24 |
+
"""
|
25 |
+
text_hash = hash(text[:1000]) # Only use first 1000 chars for hashing
|
26 |
+
metrics_hash = hash(json.dumps(metrics, sort_keys=True))
|
27 |
+
graph_hash = hash(graph_data[:100]) if graph_data else 0
|
28 |
+
return f"{text_hash}_{metrics_hash}_{graph_hash}_{lang_code}"
|
29 |
+
|
30 |
+
def format_semantic_context(text, metrics, graph_data, lang_code):
|
31 |
+
"""
|
32 |
+
Format the semantic analysis data for Claude's context.
|
33 |
+
"""
|
34 |
+
formatted_data = {
|
35 |
+
'text_sample': text[:2000], # Limit text sample
|
36 |
+
'key_concepts': metrics.get('key_concepts', []),
|
37 |
+
'concept_centrality': metrics.get('concept_centrality', {}),
|
38 |
+
'graph_description': "Network graph available" if graph_data else "No graph available",
|
39 |
+
'language': lang_code
|
40 |
+
}
|
41 |
+
|
42 |
+
return json.dumps(formatted_data, indent=2, ensure_ascii=False)
|
43 |
+
|
44 |
+
def initiate_semantic_conversation(text, metrics, graph_data, lang_code):
|
45 |
+
"""
|
46 |
+
Start a conversation with Claude about semantic analysis results.
|
47 |
+
"""
|
48 |
+
try:
|
49 |
+
api_key = os.environ.get("ANTHROPIC_API_KEY")
|
50 |
+
if not api_key:
|
51 |
+
logger.error("Claude API key not found in environment variables")
|
52 |
+
return get_fallback_response(lang_code)
|
53 |
+
|
54 |
+
# Check cache first
|
55 |
+
cache_key = get_conversation_cache_key(text, metrics, graph_data, lang_code)
|
56 |
+
if cache_key in conversation_cache:
|
57 |
+
logger.info("Using cached conversation starter")
|
58 |
+
return conversation_cache[cache_key]
|
59 |
+
|
60 |
+
# Format context for Claude
|
61 |
+
context = format_semantic_context(text, metrics, graph_data, lang_code)
|
62 |
+
|
63 |
+
# Determine language for prompt
|
64 |
+
if lang_code == 'es':
|
65 |
+
system_prompt = """Eres un asistente especializado en análisis semántico de textos.
|
66 |
+
El usuario ha analizado un texto y quiere discutir los resultados contigo.
|
67 |
+
Estos son los datos del análisis:
|
68 |
+
- Fragmento del texto analizado
|
69 |
+
- Lista de conceptos clave identificados
|
70 |
+
- Medidas de centralidad de los conceptos
|
71 |
+
- Un grafo de relaciones conceptuales (si está disponible)
|
72 |
+
|
73 |
+
Tu rol es:
|
74 |
+
1. Demostrar comprensión del análisis mostrado
|
75 |
+
2. Hacer preguntas relevantes sobre los resultados
|
76 |
+
3. Ayudar al usuario a interpretar los hallazgos
|
77 |
+
4. Sugerir posibles direcciones para profundizar el análisis
|
78 |
+
|
79 |
+
Usa un tono profesional pero accesible. Sé conciso pero claro.
|
80 |
+
"""
|
81 |
+
user_prompt = f"""Aquí están los resultados del análisis semántico:
|
82 |
+
|
83 |
+
{context}
|
84 |
+
|
85 |
+
Por favor:
|
86 |
+
1. Haz un breve resumen de lo que notas en los resultados
|
87 |
+
2. Formula 2-3 preguntas interesantes que podríamos explorar sobre estos datos
|
88 |
+
3. Sugiere un aspecto del análisis que podría profundizarse
|
89 |
+
|
90 |
+
Mantén tu respuesta bajo 250 palabras."""
|
91 |
+
|
92 |
+
elif lang_code == 'fr':
|
93 |
+
system_prompt = """Vous êtes un assistant spécialisé dans l'analyse sémantique de textes.
|
94 |
+
L'utilisateur a analysé un texte et souhaite discuter des résultats avec vous.
|
95 |
+
Voici les données d'analyse:
|
96 |
+
- Extrait du texte analysé
|
97 |
+
- Liste des concepts clés identifiés
|
98 |
+
- Mesures de centralité des concepts
|
99 |
+
- Un graphique des relations conceptuelles (si disponible)
|
100 |
+
|
101 |
+
Votre rôle est:
|
102 |
+
1. Démontrer une compréhension de l'analyse présentée
|
103 |
+
2. Poser des questions pertinentes sur les résultats
|
104 |
+
3. Aider l'utilisateur à interpréter les résultats
|
105 |
+
4. Proposer des pistes pour approfondir l'analyse
|
106 |
+
|
107 |
+
Utilisez un ton professionnel mais accessible. Soyez concis mais clair.
|
108 |
+
"""
|
109 |
+
user_prompt = f"""Voici les résultats de l'analyse sémantique:
|
110 |
+
|
111 |
+
{context}
|
112 |
+
|
113 |
+
Veuillez:
|
114 |
+
1. Faire un bref résumé de ce que vous remarquez dans les résultats
|
115 |
+
2. Formuler 2-3 questions intéressantes que nous pourrions explorer
|
116 |
+
3. Suggérer un aspect de l'analyse qui pourrait être approfondi
|
117 |
+
|
118 |
+
Limitez votre réponse à 250 mots."""
|
119 |
+
|
120 |
+
elif lang_code == 'pt':
|
121 |
+
system_prompt = """Você é um assistente especializado em análise semântica de textos.
|
122 |
+
O usuário analisou um texto e quer discutir os resultados com você.
|
123 |
+
Aqui estão os dados da análise:
|
124 |
+
- Trecho do texto analisado
|
125 |
+
- Lista de conceitos-chave identificados
|
126 |
+
- Medidas de centralidade dos conceitos
|
127 |
+
- Um grafo de relações conceituais (se disponível)
|
128 |
+
|
129 |
+
Seu papel é:
|
130 |
+
1. Demonstrar compreensão da análise apresentada
|
131 |
+
2. Fazer perguntas relevantes sobre os resultados
|
132 |
+
3. Ajudar o usuário a interpretar os achados
|
133 |
+
4. Sugerir possíveis direções para aprofundar a análise
|
134 |
+
|
135 |
+
Use um tom profissional mas acessível. Seja conciso mas claro.
|
136 |
+
"""
|
137 |
+
user_prompt = f"""Aqui estão os resultados da análise semântica:
|
138 |
+
|
139 |
+
{context}
|
140 |
+
|
141 |
+
Por favor:
|
142 |
+
1. Faça um breve resumo do que você nota nos resultados
|
143 |
+
2. Formule 2-3 perguntas interessantes que poderíamos explorar
|
144 |
+
3. Sugira um aspecto da análise que poderia ser aprofundado
|
145 |
+
|
146 |
+
Mantenha sua resposta em até 250 palavras."""
|
147 |
+
|
148 |
+
else: # Default to English
|
149 |
+
system_prompt = """You are an assistant specialized in semantic text analysis.
|
150 |
+
The user has analyzed a text and wants to discuss the results with you.
|
151 |
+
Here is the analysis data:
|
152 |
+
- Sample of the analyzed text
|
153 |
+
- List of identified key concepts
|
154 |
+
- Concept centrality measures
|
155 |
+
- A concept relationship graph (if available)
|
156 |
+
|
157 |
+
Your role is to:
|
158 |
+
1. Demonstrate understanding of the shown analysis
|
159 |
+
2. Ask relevant questions about the results
|
160 |
+
3. Help the user interpret the findings
|
161 |
+
4. Suggest possible directions to deepen the analysis
|
162 |
+
|
163 |
+
Use a professional but accessible tone. Be concise but clear.
|
164 |
+
"""
|
165 |
+
user_prompt = f"""Here are the semantic analysis results:
|
166 |
+
|
167 |
+
{context}
|
168 |
+
|
169 |
+
Please:
|
170 |
+
1. Give a brief summary of what you notice in the results
|
171 |
+
2. Formulate 2-3 interesting questions we could explore
|
172 |
+
3. Suggest one aspect of the analysis that could be deepened
|
173 |
+
|
174 |
+
Keep your response under 250 words."""
|
175 |
+
|
176 |
+
# Initialize Claude client
|
177 |
+
client = anthropic.Anthropic(api_key=api_key)
|
178 |
+
|
179 |
+
# Call Claude API
|
180 |
+
start_time = time.time()
|
181 |
+
response = client.messages.create(
|
182 |
+
model="claude-3-sonnet-20240229",
|
183 |
+
max_tokens=1024,
|
184 |
+
temperature=0.7,
|
185 |
+
system=system_prompt,
|
186 |
+
messages=[
|
187 |
+
{"role": "user", "content": user_prompt}
|
188 |
+
]
|
189 |
+
)
|
190 |
+
logger.info(f"Claude API call completed in {time.time() - start_time:.2f} seconds")
|
191 |
+
|
192 |
+
# Extract response
|
193 |
+
initial_response = response.content[0].text
|
194 |
+
|
195 |
+
# Cache the result
|
196 |
+
conversation_cache[cache_key] = initial_response
|
197 |
+
|
198 |
+
return initial_response
|
199 |
+
|
200 |
+
except Exception as e:
|
201 |
+
logger.error(f"Error initiating semantic conversation: {str(e)}")
|
202 |
+
return get_fallback_response(lang_code)
|
203 |
+
|
204 |
+
def continue_conversation(conversation_history, new_message, lang_code):
|
205 |
+
"""
|
206 |
+
Continue an existing conversation about semantic analysis.
|
207 |
+
"""
|
208 |
+
try:
|
209 |
+
api_key = os.environ.get("ANTHROPIC_API_KEY")
|
210 |
+
if not api_key:
|
211 |
+
logger.error("Claude API key not found in environment variables")
|
212 |
+
return get_fallback_response(lang_code)
|
213 |
+
|
214 |
+
# Prepare conversation history for Claude
|
215 |
+
messages = []
|
216 |
+
for msg in conversation_history:
|
217 |
+
messages.append({
|
218 |
+
"role": "user" if msg["sender"] == "user" else "assistant",
|
219 |
+
"content": msg["message"]
|
220 |
+
})
|
221 |
+
|
222 |
+
# Add the new message
|
223 |
+
messages.append({"role": "user", "content": new_message})
|
224 |
+
|
225 |
+
# System prompt based on language
|
226 |
+
if lang_code == 'es':
|
227 |
+
system_prompt = """Continúa la conversación sobre el análisis semántico.
|
228 |
+
Sé conciso pero útil. Responde en español."""
|
229 |
+
elif lang_code == 'fr':
|
230 |
+
system_prompt = """Continuez la conversation sur l'analyse sémantique.
|
231 |
+
Soyez concis mais utile. Répondez en français."""
|
232 |
+
elif lang_code == 'pt':
|
233 |
+
system_prompt = """Continue a conversa sobre a análise semântica.
|
234 |
+
Seja conciso mas útil. Responda em português."""
|
235 |
+
else:
|
236 |
+
system_prompt = """Continue the conversation about semantic analysis.
|
237 |
+
Be concise but helpful. Respond in English."""
|
238 |
+
|
239 |
+
# Initialize Claude client
|
240 |
+
client = anthropic.Anthropic(api_key=api_key)
|
241 |
+
|
242 |
+
# Call Claude API
|
243 |
+
response = client.messages.create(
|
244 |
+
model="claude-3-sonnet-20240229",
|
245 |
+
max_tokens=1024,
|
246 |
+
temperature=0.7,
|
247 |
+
system=system_prompt,
|
248 |
+
messages=messages
|
249 |
+
)
|
250 |
+
|
251 |
+
return response.content[0].text
|
252 |
+
|
253 |
+
except Exception as e:
|
254 |
+
logger.error(f"Error continuing semantic conversation: {str(e)}")
|
255 |
+
return get_fallback_response(lang_code)
|
256 |
+
|
257 |
+
def get_fallback_response(lang_code):
|
258 |
+
"""
|
259 |
+
Return fallback response if Claude API fails.
|
260 |
+
"""
|
261 |
+
if lang_code == 'es':
|
262 |
+
return """Parece que hay un problema técnico. Por favor intenta de nuevo más tarde.
|
263 |
+
|
264 |
+
Mientras tanto, aquí hay algunas preguntas que podrías considerar sobre tu análisis:
|
265 |
+
1. ¿Qué conceptos tienen la mayor centralidad y por qué podría ser?
|
266 |
+
2. ¿Hay conexiones inesperadas entre conceptos en tu grafo?
|
267 |
+
3. ¿Cómo podrías profundizar en las relaciones entre los conceptos clave?"""
|
268 |
+
|
269 |
+
elif lang_code == 'fr':
|
270 |
+
return """Il semble y avoir un problème technique. Veuillez réessayer plus tard.
|
271 |
+
|
272 |
+
En attendant, voici quelques questions que vous pourriez considérer:
|
273 |
+
1. Quels concepts ont la plus grande centralité et pourquoi?
|
274 |
+
2. Y a-t-il des connexions inattendues entre les concepts?
|
275 |
+
3. Comment pourriez-vous approfondir les relations entre les concepts clés?"""
|
276 |
+
|
277 |
+
elif lang_code == 'pt':
|
278 |
+
return """Parece haver um problema técnico. Por favor, tente novamente mais tarde.
|
279 |
+
|
280 |
+
Enquanto isso, aqui estão algumas perguntas que você poderia considerar:
|
281 |
+
1. Quais conceitos têm maior centralidade e por que isso pode ocorrer?
|
282 |
+
2. Há conexões inesperadas entre conceitos no seu grafo?
|
283 |
+
3. Como você poderia aprofundar as relações entre os conceitos-chave?"""
|
284 |
+
|
285 |
+
else:
|
286 |
+
return """There seems to be a technical issue. Please try again later.
|
287 |
+
|
288 |
+
Meanwhile, here are some questions you might consider about your analysis:
|
289 |
+
1. Which concepts have the highest centrality and why might that be?
|
290 |
+
2. Are there unexpected connections between concepts in your graph?
|
291 |
+
3. How could you explore the relationships between key concepts further?"""
|
292 |
+
|
293 |
+
def store_conversation(username, text, metrics, graph_data, conversation):
|
294 |
+
"""
|
295 |
+
Store the conversation in the database.
|
296 |
+
"""
|
297 |
+
try:
|
298 |
+
result = store_semantic_interaction(
|
299 |
+
username=username,
|
300 |
+
text=text,
|
301 |
+
metrics=metrics,
|
302 |
+
graph_data=graph_data,
|
303 |
+
conversation=conversation
|
304 |
+
)
|
305 |
+
|
306 |
+
logger.info(f"Conversation stored successfully: {result}")
|
307 |
+
return result
|
308 |
+
except Exception as e:
|
309 |
+
logger.error(f"Error storing conversation: {str(e)}")
|
310 |
+
return False
|
311 |
+
|
312 |
+
def display_semantic_chat(text, metrics, graph_data, lang_code, t):
|
313 |
+
"""
|
314 |
+
Display the chat interface for semantic analysis discussion.
|
315 |
+
"""
|
316 |
+
try:
|
317 |
+
# Initialize session state for conversation if not exists
|
318 |
+
if 'semantic_chat' not in st.session_state:
|
319 |
+
st.session_state.semantic_chat = {
|
320 |
+
'history': [],
|
321 |
+
'initialized': False
|
322 |
+
}
|
323 |
+
|
324 |
+
# Container for chat display
|
325 |
+
chat_container = st.container()
|
326 |
+
|
327 |
+
# Initialize conversation if not done yet
|
328 |
+
if not st.session_state.semantic_chat['initialized']:
|
329 |
+
with st.spinner(t.get('initializing_chat', 'Initializing conversation...')):
|
330 |
+
initial_response = initiate_semantic_conversation(
|
331 |
+
text, metrics, graph_data, lang_code
|
332 |
+
)
|
333 |
+
|
334 |
+
st.session_state.semantic_chat['history'].append({
|
335 |
+
"sender": "assistant",
|
336 |
+
"message": initial_response
|
337 |
+
})
|
338 |
+
st.session_state.semantic_chat['initialized'] = True
|
339 |
+
|
340 |
+
# Store initial conversation
|
341 |
+
if 'username' in st.session_state:
|
342 |
+
store_conversation(
|
343 |
+
st.session_state.username,
|
344 |
+
text,
|
345 |
+
metrics,
|
346 |
+
graph_data,
|
347 |
+
st.session_state.semantic_chat['history']
|
348 |
+
)
|
349 |
+
|
350 |
+
# Display chat history
|
351 |
+
with chat_container:
|
352 |
+
st.markdown("### 💬 " + t.get('semantic_discussion', 'Semantic Analysis Discussion'))
|
353 |
+
|
354 |
+
for msg in st.session_state.semantic_chat['history']:
|
355 |
+
if msg["sender"] == "user":
|
356 |
+
st.chat_message("user").write(msg["message"])
|
357 |
+
else:
|
358 |
+
st.chat_message("assistant").write(msg["message"])
|
359 |
+
|
360 |
+
# Input for new message
|
361 |
+
user_input = st.chat_input(
|
362 |
+
t.get('chat_input_placeholder', 'Ask about your semantic analysis...')
|
363 |
+
)
|
364 |
+
|
365 |
+
if user_input:
|
366 |
+
# Add user message to history
|
367 |
+
st.session_state.semantic_chat['history'].append({
|
368 |
+
"sender": "user",
|
369 |
+
"message": user_input
|
370 |
+
})
|
371 |
+
|
372 |
+
# Display user message immediately
|
373 |
+
with chat_container:
|
374 |
+
st.chat_message("user").write(user_input)
|
375 |
+
with st.spinner(t.get('assistant_thinking', 'Assistant is thinking...')):
|
376 |
+
# Get assistant response
|
377 |
+
assistant_response = continue_conversation(
|
378 |
+
st.session_state.semantic_chat['history'],
|
379 |
+
user_input,
|
380 |
+
lang_code
|
381 |
+
)
|
382 |
+
|
383 |
+
# Add assistant response to history
|
384 |
+
st.session_state.semantic_chat['history'].append({
|
385 |
+
"sender": "assistant",
|
386 |
+
"message": assistant_response
|
387 |
+
})
|
388 |
+
|
389 |
+
# Display assistant response
|
390 |
+
st.chat_message("assistant").write(assistant_response)
|
391 |
+
|
392 |
+
# Store updated conversation
|
393 |
+
if 'username' in st.session_state:
|
394 |
+
store_conversation(
|
395 |
+
st.session_state.username,
|
396 |
+
text,
|
397 |
+
metrics,
|
398 |
+
graph_data,
|
399 |
+
st.session_state.semantic_chat['history']
|
400 |
+
)
|
401 |
+
|
402 |
+
except Exception as e:
|
403 |
+
logger.error(f"Error displaying semantic chat: {str(e)}")
|
404 |
+
st.error(t.get('chat_error', 'Error in chat interface. Please try again.'))
|