Spaces:
Sleeping
Sleeping
File size: 8,021 Bytes
99b658c |
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 |
import gradio as gr
import networkx as nx
import spacy
import plotly.graph_objects as go
from sources import RSS_FEEDS
from fetch import fetch_articles
# Imports for the LLM knowledge graph transformer
from langchain.schema import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import ChatOpenAI
from langchain_experimental.graph_transformers import LLMGraphTransformer
def build_interactive_knowledge_graph(feed_items):
"""
Build an interactive knowledge graph from aggregated RSS feed text using an LLM.
Steps:
1. Combine the title and summary of all feed items.
2. Create a Document and split it into chunks.
3. Use ChatOpenAI and LLMGraphTransformer to get graph information.
4. Merge nodes and relationships into a directed NetworkX graph.
5. Compute a spring layout and convert the graph to a Plotly figure.
6. Compute node hover text showing all outgoing/incoming connections.
7. Re-add arrow annotations to indicate direction (with no extra text).
8. Return the Plotly figure.
"""
# 1. Combine all feed items into one aggregated text.
combined_text = "\n\n".join([f"{item['title']}. {item['summary']}" for item in feed_items])
# 2. Create a Document and split it.
doc = Document(page_content=combined_text)
text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
docs = text_splitter.split_documents([doc])
# 3. Initialize the LLM and transformer.
llm = ChatOpenAI(temperature=0, model="gpt-4o")
llm_transformer = LLMGraphTransformer(llm=llm)
graph_documents = llm_transformer.convert_to_graph_documents(docs)
# 4. Build a directed NetworkX graph.
G = nx.DiGraph()
for graph_doc in graph_documents:
# Convert the Pydantic model to a dictionary.
gdoc = graph_doc.model_dump()
nodes = gdoc.get("nodes", [])
# In these documents, relationships are stored under "relationships".
relationships = gdoc.get("relationships", [])
# Add nodes.
for node in nodes:
node_id = node.get("id") or node.get("name")
if node_id:
G.add_node(node_id)
# Add relationships as directed edges.
for rel in relationships:
source_obj = rel.get("source", {})
target_obj = rel.get("target", {})
source = source_obj.get("id")
target = target_obj.get("id")
rel_type = rel.get("type", "")
if source and target:
if G.has_edge(source, target):
if "relation_types" in G[source][target]:
if rel_type not in G[source][target]["relation_types"]:
G[source][target]["relation_types"].append(rel_type)
else:
G[source][target]["relation_types"] = [rel_type]
G[source][target]["weight"] += 1
else:
G.add_edge(source, target, weight=1, relation_types=[rel_type])
# 5. Compute positions using a spring layout.
pos = nx.spring_layout(G, k=1.2)
#pos = nx.kamada_kawai_layout(G)
# 6. Prepare node hover text.
# For each node, list all outgoing and incoming connection details.
node_hover = {}
for node in G.nodes():
outgoing = []
for u, v, data in G.out_edges(node, data=True):
rels = ", ".join(data.get("relation_types", []))
outgoing.append(f"Out: {node} - {rels} -> {v}")
incoming = []
for u, v, data in G.in_edges(node, data=True):
rels = ", ".join(data.get("relation_types", []))
incoming.append(f"In: {u} - {rels} -> {node}")
details = outgoing + incoming
if details:
node_hover[node] = "<br>".join(details)
else:
node_hover[node] = node # Fallback if there are no connections.
# 7. Create node trace using calculated positions and hover text.
node_x = []
node_y = []
node_text = [] # Displayed text is just the node name.
node_hover_list = [] # Custom hover info with connection details.
for node in G.nodes():
x, y = pos[node]
node_x.append(x)
node_y.append(y)
node_text.append(node)
node_hover_list.append(node_hover.get(node, node))
node_trace = go.Scatter(
x=node_x,
y=node_y,
mode='markers+text',
text=node_text,
textposition="top center",
hoverinfo='text',
hovertext=node_hover_list,
marker=dict(
size=10,
color='#1f78b4'
)
)
# 8. Create edge traces: one trace per edge.
edge_traces = []
for edge in G.edges(data=True):
x0, y0 = pos[edge[0]]
x1, y1 = pos[edge[1]]
edge_trace = go.Scatter(
x=[x0, x1],
y=[y0, y1],
mode='lines',
line=dict(width=1, color='#888'),
hoverinfo='none'
)
edge_traces.append(edge_trace)
# 9. Build the interactive Plotly figure.
fig = go.Figure(
data=edge_traces + [node_trace],
layout=go.Layout(
title='<br>Interactive Knowledge Graph (LLM-derived)',
showlegend=False,
hovermode='closest',
margin=dict(b=20, l=5, r=5, t=40),
xaxis=dict(showgrid=False, zeroline=False, showticklabels=False),
yaxis=dict(showgrid=False, zeroline=False, showticklabels=False),
width=1200, # wider figure
height=800, # taller figure
dragmode='pan'
)
)
# 10. Re-add arrow annotations for each edge (without hover text).
for edge in G.edges(data=True):
x0, y0 = pos[edge[0]]
x1, y1 = pos[edge[1]]
fig.add_annotation(
x=x1,
y=y1,
ax=x0,
ay=y0,
xref='x',
yref='y',
axref='x',
ayref='y',
showarrow=True,
arrowhead=3,
arrowcolor='#888',
arrowwidth=2,
text="", # No text; rely on node hover for details.
)
return fig
def get_combined_feed(source_choice, selected_news_sites):
"""
Create an aggregated feed from selected RSS sources
and build an interactive Plotly knowledge graph.
"""
feed_items = []
# Fetch articles from selected news sites.
if "News" in source_choice and selected_news_sites:
selected_feeds = {name: url for name, url in RSS_FEEDS.items() if name in selected_news_sites}
feed_items += fetch_articles(selected_feeds, limit=6)
# Aggregate feed text.
feed_text = "\n\n".join([f"🔹 {item['title']} ({item['published']})\n{item['link']}" for item in feed_items])
# Build an interactive knowledge graph using Plotly.
graph_fig = build_interactive_knowledge_graph(feed_items)
return feed_text, graph_fig
# Define the Gradio interface with a button to trigger processing.
with gr.Blocks() as demo:
with gr.Row():
with gr.Column():
source_selector = gr.CheckboxGroup(
["News"], value=["News"], label="Select Sources"
)
news_site_selector = gr.CheckboxGroup(
list(RSS_FEEDS.keys()), value=["BBC", "Wired"], label="News Sites"
)
with gr.Column():
feed_output = gr.Textbox(label="Aggregated Feed", lines=20)
with gr.Row():
with gr.Column():
graph_output = gr.Plot(label="Interactive Knowledge Graph")
# Button to trigger graph generation.
generate_button = gr.Button("Generate Graph")
generate_button.click(
fn=get_combined_feed,
inputs=[source_selector, news_site_selector],
outputs=[feed_output, graph_output]
)
demo.launch()
|