...

.gitignore

@@ -0,0 +1,37 @@
+# Python temporary files
+__pycache__/
+*.py[cod]
+*$py.class
+*.pyc
+*.pyo
+*.pyd
+
+# Streamlit temporary and config files
+.streamlit/runtime/
+.streamlit/secrets.toml
+
+# Virtual environments
+.venv/
+venv/
+env/
+ENV/
+lib/
+bin/
+include/
+share/
+
+# Jupyter Notebook files and checkpoints
+*.ipynb
+.ipynb_checkpoints/
+
+# Generated HTML graph files (recreated at runtime)
+*.html
+
+# Backup files
+*.bak
+
+# Environment variables (Crucial for API keys security)
+.env
+.env.local
+.env.*
+!.env.example

app.py

@@ -12,6 +12,12 @@ st.set_page_config(
 
 st.title("Knowledge Graph From Text")
 
+# Initialize session state variables
+if 'graph_version' not in st.session_state:
+    st.session_state['graph_version'] = 0
+if 'qa_version' not in st.session_state:
+    st.session_state['qa_version'] = 0
+
 st.sidebar.title("Input document")
 input_method = st.sidebar.radio(
     "Choose an input method:",
@@ -32,15 +38,32 @@ if st.sidebar.button("1. Generate Knowledge Graph"):
         with st.spinner("Generating knowledge graph..."):
             net, graph_docs = generate_knowledge_graph(text)
             st.session_state['graph_docs'] = graph_docs
-            st.success("Knowledge graph generated successfully!")
-
+            
             output_file = "knowledge_graph.html"
             net.save_graph(output_file)
-            HtmlFile = open(output_file, 'r', encoding='utf-8')
-            components.html(HtmlFile.read(), height=600)
+            with open(output_file, 'r', encoding='utf-8') as f:
+                st.session_state['graph_html'] = f.read()
+            
+            # Increment version to force iframe refresh
+            st.session_state['graph_version'] += 1
+            
+            # Reset QA state for new graph
+            st.session_state.pop('qa_answer', None)
+            st.session_state.pop('qa_html', None)
+            
+            st.success("Knowledge graph generated successfully!")
     else:
         st.sidebar.error("Please provide some text to generate the graph.")
 
+# Display the main graph if it exists in session state
+if 'graph_html' in st.session_state:
+    st.subheader("Initial Knowledge Graph")
+    # Append version comment to force Streamlit to refresh the iframe when version changes
+    components.html(
+        st.session_state['graph_html'] + f"<!-- version {st.session_state['graph_version']} -->", 
+        height=600
+    )
+
 # QA Section
 if 'graph_docs' in st.session_state:
     st.markdown("---")
@@ -50,18 +73,29 @@ if 'graph_docs' in st.session_state:
     with col1:
         question = st.text_input("Your question :")
     with col2:
-        k_value = st.slider("Relationships to be analyzed (Top K)", min_value=1, max_value=20, value=7)
+        k_value = st.slider("Relationships to be analyzed (Top K)", min_value=1, max_value=30, value=15)
 
     if st.button("2. Analyze") and question:
         with st.spinner("Semantic search in the current graph..."):
-            answer, filtered_net = answer_question_with_graph(
-                question, 
+            answer, filtered_net = answer_question_with_graph(                question, 
                 st.session_state['graph_docs'],
                 k_relations=k_value
             )
+            
+            st.session_state['qa_answer'] = answer
+            
+            # Read and save the filtered graph HTML content
+            with open("filtered_graph.html", 'r', encoding='utf-8') as f:
+                st.session_state['qa_html'] = f.read()
+                
+            st.session_state['qa_version'] += 1
 
-            st.info(f"**Answer :** {answer}")
-
-            st.markdown("**Subgraph of the relationships used to answer the question :**")
-            HtmlFile = open("filtered_graph.html", 'r', encoding='utf-8')
-            components.html(HtmlFile.read(), height=450)
+    # Persist the QA results display
+    if 'qa_answer' in st.session_state and 'qa_html' in st.session_state:
+        st.info(f"**Answer :** {st.session_state['qa_answer']}")
+        st.markdown("**Subgraph of the relationships used to answer the question :**")
+        # Append version comment to force Streamlit to refresh the iframe when version changes
+        components.html(
+            st.session_state['qa_html'] + f"<!-- version {st.session_state['qa_version']} -->", 
+            height=450
+        )

generate_knowledge_graph.py

@@ -8,11 +8,12 @@ from pyvis.network import Network
 from dotenv import load_dotenv
 import os
 import asyncio
+import streamlit as st
 
 load_dotenv()
 api_key = os.getenv("OPENAI_API_KEY")
 
-llm = ChatOpenAI(temperature=0, model_name="gpt-4o")
+llm = ChatOpenAI(temperature=0, model_name="gpt-4o-mini")
 graph_transformer = LLMGraphTransformer(llm=llm)
 
 async def extract_graph_data(text):
@@ -22,14 +23,28 @@ async def extract_graph_data(text):
 
 def visualize_graph(graph_documents):
     net = Network(height="600px", width="100%", directed=True, notebook=False, bgcolor="#222222", font_color="white", filter_menu=True, cdn_resources='remote')
-    nodes = graph_documents[0].nodes
-    relationships = graph_documents[0].relationships
+    
+    # Collect all nodes and relationships across all graph documents in the list
+    all_nodes = []
+    all_relationships = []
+    for doc in graph_documents:
+        all_nodes.extend(doc.nodes)
+        all_relationships.extend(doc.relationships)
+
+    # Build node dictionary to deduplicate nodes by ID
+    node_dict = {node.id: node for node in all_nodes}
+    
+    # Dynamically add any nodes referenced in relationships but missing from nodes list
+    for rel in all_relationships:
+        if rel.source.id not in node_dict:
+            node_dict[rel.source.id] = rel.source
+        if rel.target.id not in node_dict:
+            node_dict[rel.target.id] = rel.target
 
-    node_dict = {node.id: node for node in nodes}
     valid_edges = []
     valid_node_ids = set()
 
-    for rel in relationships:
+    for rel in all_relationships:
         if rel.source.id in node_dict and rel.target.id in node_dict:
             valid_edges.append(rel)
             valid_node_ids.update([rel.source.id, rel.target.id])
@@ -41,17 +56,28 @@ def visualize_graph(graph_documents):
         except:
             continue
 
+    # Deduplicate edges to avoid rendering duplicates
+    edges_added = set()
     for rel in valid_edges:
-        try:
-            net.add_edge(rel.source.id, rel.target.id, label=rel.type.lower())
-        except:
-            continue
+        edge_key = (rel.source.id, rel.target.id, rel.type.lower())
+        if edge_key not in edges_added:
+            try:
+                net.add_edge(rel.source.id, rel.target.id, label=rel.type.lower())
+                edges_added.add(edge_key)
+            except:
+                continue
 
     net.set_options('{"physics": {"forceAtlas2Based": {"gravitationalConstant": -100, "centralGravity": 0.01, "springLength": 200, "springConstant": 0.08}, "minVelocity": 0.75, "solver": "forceAtlas2Based"}}')
     return net
 
+
+@st.cache_data
+def get_graph_data_cached(text):
+    return asyncio.run(extract_graph_data(text))
+
+
 def generate_knowledge_graph(text):
-    graph_documents = asyncio.run(extract_graph_data(text))
+    graph_documents = get_graph_data_cached(text)
     net = visualize_graph(graph_documents)
     return net, graph_documents
 
@@ -61,11 +87,11 @@ def answer_question_with_graph(question, graph_documents, k_relations=7):
         all_relationships.extend(doc.relationships)
 
     if not all_relationships:
-        return "Aucune relation trouvée dans le graphe.", visualize_graph(graph_documents)
+        return "No relationships found in the graph.", visualize_graph(graph_documents)
 
     rel_docs = []
     for i, rel in enumerate(all_relationships):
-        text_rep = f"L'entité '{rel.source.id}' a pour relation '{rel.type}' avec l'entité '{rel.target.id}'."
+        text_rep = f"The entity '{rel.source.id}' has a '{rel.type}' relationship with the entity '{rel.target.id}'."
         rel_docs.append(Document(page_content=text_rep, metadata={"rel_index": i}))
 
     embeddings = OpenAIEmbeddings(model="text-embedding-3-small")