import pandas as pd import plotly.express as px from datasets import load_dataset import gradio as gr import yfinance as yf # --- LOAD DATASET --- df = pd.DataFrame(load_dataset("SelmaNajih001/NewsSentiment")["train"]) df = df[df["Company"].isin(["Tesla", "Microsoft", "Apple", "Facebook", "Google"])] # --- CONVERT DATE TO DATETIME --- df['Date'] = pd.to_datetime(df['Date'], errors='coerce') df['Year'] = df['Date'].dt.year df['Month'] = df['Date'].dt.to_period('M') df['Day'] = df['Date'].dt.date df = df[df['Year'] >= 2015] # --- TICKERS --- TICKERS = { "Tesla": "TSLA", "Microsoft": "MSFT", "Apple": "AAPL", "Facebook": "META", "Google": "GOOGL", "NASDAQ": "^IXIC" } # --- FETCH STOCK PRICES --- prices = {} for company, ticker in TICKERS.items(): start_date = "2015-01-01" end_date = pd.Timestamp.today() df_prices = yf.download(ticker, start=start_date, end=end_date) if isinstance(df_prices.columns, pd.MultiIndex): df_prices.columns = ['_'.join([str(c) for c in col]).strip() for col in df_prices.columns] df_prices = df_prices.reset_index()[['Date', f'Close_{ticker}']] if company == "NASDAQ": df_prices = df_prices.rename(columns={f'Close_{ticker}': 'Close_NASDAQ'}) prices[company] = df_prices # --- INTERPOLATE PRICES FOR DAY/MONTH --- def get_prices_for_agg(agg_col): df_prices_agg = {} for company, df_price in prices.items(): df_temp = df_price.copy() col = 'Close_NASDAQ' if company == 'NASDAQ' else f"Close_{TICKERS[company]}" df_temp = df_temp.rename(columns={df_temp.columns[1]: col}) if agg_col == "Day": df_temp = df_temp.set_index('Date').resample('D').mean().interpolate('linear').reset_index() elif agg_col == "Month": df_temp['Month'] = df_temp['Date'].dt.to_period('M').dt.to_timestamp() df_temp = df_temp.groupby('Month')[col].last().reset_index() elif agg_col == "Year": df_temp['Year'] = df_temp['Date'].dt.year df_temp = df_temp.groupby('Year')[col].last().reset_index() df_prices_agg[company] = df_temp return df_prices_agg # --- MERGE NEWS CON PREZZI --- df_merged = df.copy() for company in df['Company'].unique(): ticker_col = f"Close_{TICKERS[company]}" df_temp = prices[company][['Date', ticker_col]] df_merged = df_merged.merge(df_temp, on='Date', how='left') # NASDAQ per tutte le righe df_merged = df_merged.merge(prices['NASDAQ'][['Date', 'Close_NASDAQ']], on='Date', how='left') # --- GRADIO FUNCTION --- def show_sentiment(selected_companies=None, aggregation="Day", selected_year="All"): if not selected_companies: selected_companies = ["NASDAQ"] if isinstance(selected_companies, str): selected_companies = [selected_companies] df_filtered = df_merged.copy() if selected_year != "All" and selected_year is not None: df_filtered = df_filtered[df_filtered['Year'] == int(selected_year)] # colonna aggregazione group_col = aggregation if aggregation == "Month": df_filtered['Month'] = df_filtered['Month'].dt.to_timestamp() elif aggregation == "Day": df_filtered['Day'] = df_filtered['Date'] # prezzi interpolati prices_agg = get_prices_for_agg(aggregation) include_nasdaq = "NASDAQ" in selected_companies companies_to_plot = [c for c in selected_companies if c != "NASDAQ"] df_grouped_list = [] # aziende selezionate if companies_to_plot: df_sent = df_filtered[df_filtered['Company'].isin(companies_to_plot)] df_tmp = df_sent.groupby([group_col, 'Company']).agg({'Score':'sum'}).reset_index() for c in companies_to_plot: if c not in TICKERS: continue ticker_col = f"Close_{TICKERS[c]}" df_price_col = prices_agg[c][[group_col, ticker_col]] df_tmp = df_tmp.merge(df_price_col, on=group_col, how='left') df_grouped_list.append(df_tmp) # NASDAQ con sentiment generale if include_nasdaq: df_general = df_filtered.groupby(group_col).agg({'Score':'sum'}).reset_index() df_general['Company'] = 'General' df_general = df_general.merge(prices_agg['NASDAQ'].rename(columns={'Date':group_col}), on=group_col, how='left') df_grouped_list.append(df_general) # unisci tutto df_grouped = pd.concat(df_grouped_list, ignore_index=True, sort=False) df_grouped = df_grouped.sort_values([group_col, 'Company']) # --- Plot --- fig = px.line(df_grouped, x=group_col, y='Score', color='Company', title=f"Sentiment Trend ({aggregation} Aggregation)") for c in companies_to_plot: ticker_col = f"Close_{TICKERS[c]}" df_c = df_grouped[df_grouped['Company'] == c] if ticker_col in df_c.columns and df_c[ticker_col].notnull().any(): fig.add_scatter( x=df_c[group_col], y=df_c[ticker_col], mode='lines', name=f"{c} Price", yaxis="y2", line=dict(dash='dot') ) if include_nasdaq: df_c = df_grouped[df_grouped['Company'] == 'General'] if 'Close_NASDAQ' in df_c.columns and df_c['Close_NASDAQ'].notnull().any(): fig.add_scatter( x=df_c[group_col], y=df_c['Close_NASDAQ'], mode='lines', name="NASDAQ Price", yaxis="y2", line=dict(dash='dot') ) fig.update_layout( yaxis2=dict( title="Stock Price", overlaying="y", side="right" ) ) return df_grouped.tail(30), fig import gradio as gr # Markdown descrittivo adattato al tuo dashboard import gradio as gr # --- Markdown descrittivo --- description_text = """ ### Dynamic Sentiment Dashboard This dashboard allows you to explore the sentiment of news articles related to major tech companies (Apple, Tesla, Microsoft, Meta, Alphabet) and compare it with their stock prices. - **Dataset structure**: The dataset includes a company column; each row corresponds to a news item for a specific company. - **Sentiment aggregation**: Select a time aggregation level (Month or Year) to see how sentiment evolves over time. - **NASDAQ comparison**: Selecting "NASDAQ" shows the general market sentiment alongside the company-specific sentiment. - **Visual insights**: Top-left graph shows average sentiment score and closing price for the selected company. """ # --- Findings from thesis (specific companies and years) --- findings_text = """ ### Key Findings - Some news articles refer to multiple companies, e.g., the same article may mention Apple and Tesla. - Merging news with stock prices allows analyzing correlations between sentiment and stock movements for each company. - **Apple (2018, 2019, 2022):** Sentiment trends generally align with closing prices, showing similar monthly patterns. - **Tesla (2018, 2019, 2022):** More volatility observed; sentiment aligns with stock movement but is more sensitive to news on Elon Musk’s actions. - **Microsoft, Meta, Alphabet:** Across periods, sentiment trends follow stock prices with moderate correlation. - The custom sentiment model is more aligned with actual stock movements compared to FinBERT, which is more influenced by word positivity/negativity. - Aggregating sentiment by month or year helps identify broader trends while reducing noise from daily fluctuations. - Including “NASDAQ” as a general market reference allows comparison of individual companies’ sentiment with overall market sentiment. """ # --- Input options --- companies = sorted(df['Company'].unique().tolist()) + ["NASDAQ"] years = sorted(df['Year'].dropna().unique().tolist()) # --- Build Gradio Blocks --- with gr.Blocks() as demo: # Markdown in alto gr.Markdown("# Dynamic Sentiment Dashboard") gr.Markdown(description_text) # Input sotto il Markdown with gr.Row(): dropdown_companies = gr.Dropdown( choices=companies, value=None, multiselect=False, label="Select Companies" ) radio_aggregation = gr.Radio( choices=["Month", "Year"], value="Month", label="Aggregation Level" ) dropdown_year = gr.Dropdown( choices=["All"] + years, value="All", label="Select Year" ) # Bottone submit submit_btn = gr.Button("Submit") # Output data_table = gr.Dataframe(label="Sentiment Table", type="pandas") sentiment_plot = gr.Plot(label="Sentiment Trend") # Findings section gr.Markdown(findings_text) submit_btn.click( fn=show_sentiment, inputs=[dropdown_companies, radio_aggregation, dropdown_year], outputs=[data_table, sentiment_plot] ) demo.launch()