app.py

import streamlit as st
import networkx as nx
import matplotlib.pyplot as plt
import pandas as pd
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.cluster import KMeans


def main():
 st.title("Financial Graph App")
 st.write("Enter a financial sentence and see its similarity to predefined keywords.")

 # User input
 financial_sentence = st.text_area("Enter the financial sentence", value="")

 # Check if the user entered a sentence
 if financial_sentence.strip() != "":
 # Predefined keywords
 keywords = [
 "Finance",
 "Fiscal",
 "Quarterly results",
 "Revenue",
 "Profit",
 "Loss",
 "Net income",
 "Gross margin",
 "Operating expenses",
 "Cash flow",
 "Shareholders",
 "Guidance",
 "Forecast",
 "Market performance",
 "Stock price",
 "Capital expenditures",
 "Acquisitions",
 "Mergers",
 "Debt",
 "Financial ratios",
 "Growth",
 "Outlook",
 "Visibility",
 "Revenue growth",
 "Market trends",
 "Profitability",
 "Efficiency",
 "Projections",
 "Performance"
 ]

 # Load the pre-trained Sentence-Transformers model
 model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2")
 st.set_option('deprecation.showPyplotGlobalUse', False)
 # Generate word embeddings for the financial sentence and keywords
 sentence_embedding = model.encode([financial_sentence])
 keyword_embeddings = model.encode(keywords)

 # Calculate cosine similarity between the sentence embedding and keyword embeddings
 similarity_scores = cosine_similarity(sentence_embedding, keyword_embeddings)[0]

 # Create a graph
 G = nx.Graph()

 # Add the sentence embedding as a node to the graph
 G.add_node(financial_sentence, embedding=sentence_embedding[0])

 # Add the keyword embeddings as nodes to the graph
 for keyword, embedding, similarity in zip(keywords, keyword_embeddings, similarity_scores):
 G.add_node(keyword, embedding=embedding, similarity=similarity)

 # Add edges between the sentence and keywords with their similarity scores as weights
 for keyword, similarity in zip(keywords, similarity_scores):
 G.add_edge(financial_sentence, keyword, weight=similarity)

 # Perform KNN clustering on the keyword embeddings
 kmeans = KMeans(n_clusters=3)
 cluster_labels = kmeans.fit_predict(keyword_embeddings)

 # Add cluster labels as node attributes
 for node, cluster_label in zip(G.nodes, cluster_labels):
 G.nodes[node]["cluster"] = cluster_label

 # Set node positions using spring layout
 pos = nx.spring_layout(G)

 # Get unique cluster labels
 unique_clusters = set(cluster_labels)

 # Assign colors to clusters
 cluster_colors = ["lightblue", "lightgreen", "lightyellow"]

 # Draw nodes with cluster colors
 nx.draw_networkx_nodes(
 G,
 pos,
 node_color=[cluster_colors[G.nodes[node].get("cluster", 0)] for node in G.nodes],
 node_size=800,
 )

 # Draw edges
 nx.draw_networkx_edges(G, pos, edge_color="gray", width=1, alpha=0.7)

 # Draw labels
 nx.draw_networkx_labels(G, pos, font_size=10, font_weight="bold")

 # Draw edge labels (cosine similarity scores)
 edge_labels = nx.get_edge_attributes(G, "weight")
 nx.draw_networkx_edge_labels(G, pos, edge_labels=edge_labels, font_size=8)

 # Set plot attributes
 plt.title("Financial Context and Keywords")
 plt.axis("off")

 # Save the graph as an image
 plt.savefig("financial_graph.png")

 # Show the graph
 st.pyplot()

 # Save the similarity scores in a CSV file
 df = pd.DataFrame({"Keyword": keywords, "Cosine Similarity": similarity_scores})
 st.write("Similarity Scores:")
 st.dataframe(df)


if __name__ == "__main__":
 main()