Spaces:
Running
Running
File size: 11,712 Bytes
f1fce37 c29e353 f1fce37 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 |
import spacy
from transformers import pipeline
from dateutil.parser import parse
import re
import pandas as pd
from difflib import SequenceMatcher
class TextClassifier:
def __init__(self):
# Use a larger model for better NER (optional)
self.nlp = spacy.load("en_core_web_lg") # "en_core_web_lg"
try:
# Use a smaller, PyTorch-compatible model for zero-shot classification
self.classifier = pipeline("zero-shot-classification", model="typeform/distilbert-base-uncased-mnli")
self.model_available = True
print("Successfully loaded zero-shot classification model.")
except Exception as e:
print(f"Failed to load zero-shot classification model: {e}. Falling back to keyword-based classification.")
self.classifier = None
self.model_available = False
self.intents = [
"get_net_income",
"get_revenue",
"get_stock_price",
"get_profit_margin",
"get_company_profile",
"get_market_cap",
"get_historical_stock_price",
"get_dividend_info",
"get_balance_sheet",
"get_cash_flow",
"get_financial_ratios",
"get_earnings_per_share",
"get_interest",
"get_research_info",
"get_cost_info",
"get_income_tax"
]
# Mapping of company names to ticker symbols (case-insensitive)
self.company_to_ticker = {
"apple": "AAPL",
"microsoft corporation": "MSFT",
"microsoft": "MSFT",
"nvidia corporation": "NVDA",
"nvidia": "NVDA",
"amazon": "AMZN",
"alphabet inc": "GOOGL",
"google": "GOOGL",
"meta platforms": "META",
"meta": "META",
"facebook": "META",
"tesla": "TSLA",
"walmart inc": "WMT",
"walmart": "WMT",
"visa inc": "V",
"visa": "V",
"coca cola": "KO"
}
# Mapping of keywords to intents (case-insensitive)
self.intent_to_keywords = {
"get_net_income": ["net income", "income", "earnings"],
"get_revenue": ["revenue", "sales", "turnover", "gross income"],
"get_stock_price": ["stock price", "stock", "price", "share price", "current price", "price now", "stock value"],
"get_profit_margin": ["profit margin", "margin", "profit percentage", "net margin", "profit"],
"get_company_profile": ["who is", "company profile", "about company", "company info"],
"get_market_cap": ["market cap", "market capitalization", "company value", "valuation"],
"get_historical_stock_price": ["historical stock price", "stock price on", "past stock price", "stock price in", "price on"],
"get_dividend_info": ["dividend info", "dividend payout", "payout ratio", "dividend yield", "dividend"],
"get_balance_sheet": ["balance sheet", "sheet", "financial position", "assets and liabilities", "balance"],
"get_cash_flow": ["cash", "flow", "cash flow", "cashflow", "cash from operations", "operating cash"],
"get_financial_ratios": ["financial ratios", "ratios", "current ratio", "liquidity ratio", "debt ratio"],
"get_earnings_per_share": ["earnings per share", "eps", "per share earnings"],
}
def classify_by_keywords(self, text):
"""
Classify the intent based on keyword mapping.
Args:
text (str): The input text to classify.
Returns:
str: The predicted intent, or None if no match is found.
"""
text_lower = text.lower()
for intent, keywords in self.intent_to_keywords.items():
if any(keyword in text_lower for keyword in keywords):
print(f"Classified intent: {intent} based on keywords: {keywords}")
return intent
print("No intent matched based on keywords.")
return None # Fallback if no keywords match
def classify_with_llm(self, text):
if not self.model_available:
print("Zero-shot classifier not available. Using keyword-based classification.")
return self.classify_by_keywords(text)
try:
hypothesis_template = "This text is requesting {} information."
result = self.classifier(text, candidate_labels=self.intents, hypothesis_template=hypothesis_template, multi_label=False)
predicted_intent = result["labels"][0]
print(f"Predicted intent: {predicted_intent} with scores: {dict(zip(result['labels'], result['scores']))}")
return predicted_intent
except Exception as e:
print(f"Error classifying intent with model: {e}. Falling back to keyword-based classification.")
return self.classify_by_keywords(text)
def extract_entities(self, text):
doc = self.nlp(text)
entities = {"ticker": None, "metric": None, "year": None, "date": None}
# Step 1: Extract entities using spaCy NER
for ent in doc.ents:
if ent.label_ == "ORG":
org_name = ent.text.lower()
ticker = self.company_to_ticker.get(org_name)
if ticker:
entities["ticker"] = ticker
else:
# If not found in the mapping, search in the CSV file
try:
# Load the CSV file (adjust the path as needed)
csv_path = "financial data sp500 companies.csv" # Same path as used in Retriever
df = pd.read_csv(csv_path)
# Ensure the required columns exist
if "firm" not in df.columns or "Ticker" not in df.columns:
print("Required columns 'firm' or 'Ticker' not found in CSV. Using fallback ticker.")
entities["ticker"] = ent.text.upper()
else:
# Calculate similarity scores between org_name and each firm name
df["similarity"] = df["firm"].apply(
lambda x: SequenceMatcher(None, org_name, str(x).lower()).ratio()
)
# Find rows with similarity >= 80%
matches = df[df["similarity"] >= 0.5]
if not matches.empty:
# Take the first match (highest similarity)
best_match = matches.sort_values(by="similarity", ascending=False).iloc[0]
ticker = best_match["Ticker"]
print(f"Found ticker {ticker} for {org_name} with similarity {best_match['similarity']:.2f}")
entities["ticker"] = ticker
else:
print(f"No match found for {org_name} with >= 50% similarity. Using fallback ticker.")
entities["ticker"] = ent.text.upper()
except Exception as e:
print(f"Error searching CSV for ticker: {e}. Using fallback ticker.")
entities["ticker"] = ent.text.upper()
elif ent.label_ == "DATE":
date_text = ent.text.lower()
try:
parsed_date = parse(date_text, fuzzy=True, default=parse("2025-01-01"))
# If the date is a year (e.g., "2023", "this year") or parsed as January 1
if "year" in date_text or date_text.isdigit() or (parsed_date.day == 1 and parsed_date.month == 1):
entities["year"] = parsed_date.strftime("%Y")
else:
# Otherwise, treat it as a specific date (e.g., "Jan 5")
entities["date"] = parsed_date.strftime("%Y-%m-%d")
except ValueError:
# Fallback if parsing fails
if "year" in date_text or date_text.isdigit():
entities["year"] = date_text
else:
entities["date"] = date_text
# Step 2: Fallback ticker extraction if spaCy fails to identify ORG
if not entities["ticker"]:
text_lower = text.lower()
for company_name, ticker in self.company_to_ticker.items():
if company_name in text_lower:
entities["ticker"] = ticker
break
# Step 3: Extract metric using keyword matching with synonyms
text_lower = text.lower()
if any(keyword in text_lower for keyword in ["net income", "net", "income"]):
entities["metric"] = "netIncome"
elif "revenue" in text_lower:
entities["metric"] = "revenue"
elif any(keyword in text_lower for keyword in ["profit margin", "profit", "margin"]):
entities["metric"] = "netProfitMargin"
elif any(keyword in text_lower for keyword in ["market cap", "market capitalization", "market"]):
entities["metric"] = "mktCap"
elif any(keyword in text_lower for keyword in ["payout ratio", "dividend payout"]):
entities["metric"] = "payoutRatio"
elif any(keyword in text_lower for keyword in ["current ratio", "liquidity ratio"]):
entities["metric"] = "currentRatio"
elif any(keyword in text_lower for keyword in ["eps", "earnings per share", "earnings"]):
entities["metric"] = "eps"
elif any(keyword in text_lower for keyword in ["stock", "stock price", "current price", "valuation", "price"]):
entities["metric"] = "price"
elif any(keyword in text_lower for keyword in ["company info", "about company", "who is"]):
entities["metric"] = "ceo"
elif any(keyword in text_lower for keyword in ["balance sheet", "sheet", "assets"]):
entities["metric"] = "Assets&Liabilities"
elif any(keyword in text_lower for keyword in ["historical", "earnings per share", "earnings"]):
entities["metric"] = "historical"
elif any(keyword in text_lower for keyword in ["cash", "flow", "cash flow"]):
entities["metric"] = "cashFlowFromOperatingActivities"
elif any(keyword in text_lower for keyword in ["tax"]):
entities["metric"] = "IncomeTax"
elif any(keyword in text_lower for keyword in ["interest", "interest expense", "expense"]):
entities["metric"] = "InterestExpense"
elif any(keyword in text_lower for keyword in ["research", "research development", "development"]):
entities["metric"] = "Research"
elif any(keyword in text_lower for keyword in ["cost", "total cost"]):
entities["metric"] = "TotalCost"
# Step 4: Normalize year (handle "this year", "last year", etc.)
if entities["year"]:
year_text = entities["year"].lower()
current_year = 2025 # Based on the current date (April 16, 2025)
if "this year" in year_text:
entities["year"] = str(current_year)
elif "last year" in year_text:
entities["year"] = str(current_year - 1)
elif re.match(r"^\d{4}$", year_text):
entities["year"] = year_text
else:
# If year is not a valid format, unset it
entities["year"] = None
return entities |