File size: 46,306 Bytes
5c3c401 72f4cb5 5c3c401 72f4cb5 5c3c401 72f4cb5 5c3c401 72f4cb5 5c3c401 72f4cb5 5c3c401 72f4cb5 5c3c401 72f4cb5 5c3c401 b0e3d06 5c3c401 b0e3d06 5c3c401 72f4cb5 5c3c401 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 |
# pylint: disable=too-many-lines
"""
LDA controller for topic modeling and visualization.
"""
import re
import logging
from datetime import datetime, timedelta
from collections import Counter, defaultdict
import numpy as np
from nltk.tokenize import word_tokenize
from nltk.stem import WordNetLemmatizer
from nltk.corpus import stopwords
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from sklearn.metrics import silhouette_score
from gensim import corpora
from gensim.models import LdaModel
from gensim.models.coherencemodel import CoherenceModel
import requests
from langchain.schema import SystemMessage, HumanMessage
from models.llm import gpt # pylint: disable=import-error
from .keyword import FIN_KEYWORDS
from .keyword_analysis import get_time_range, get_previous_time_range
# Configure logger
logger = logging.getLogger(__name__)
class FinancialKeywordManager:
"""
Manages both static predefined financial keywords and dynamically fetched trending keywords.
This class handles the fetching and combining of financial keywords from different sources.
It provides mechanisms to fetch trending financial terms from external APIs and combine them
with a predefined list of static financial keywords.
Attributes:
static_keywords (set): Predefined set of financial keywords
trending_keywords (set): Dynamically fetched trending financial terms
"""
def __init__(self, static_keywords):
"""Initialize the FinancialKeywordManager with a set of static keywords."""
self.static_keywords = set(static_keywords)
self.trending_keywords = set()
def fetch_trending_keywords(self, days_back=30):
"""
Fetch trending keywords from external APIs
Args:
days_back (int): Number of days to look back for trending keywords
Returns:
None: Updates the trending_keywords set internally
"""
logger.info("Fetching trending keywords from last %d days...", days_back)
trending = set()
trending.update(self.fetch_from_eodhd(days_back))
self.trending_keywords = trending
def _is_valid_keyword(self, keyword: str) -> bool:
"""
Applies a series of rules to determine if a keyword is a valid, non-noise financial term.
This function is designed to filter out stock tickers, short codes, and common junk.
Args:
keyword (str): The keyword to validate.
Returns:
bool: True if the keyword is valid, False otherwise.
"""
# Rule 1: Basic length check
if not 3 <= len(keyword) <= 30:
return False
# Rule 2: Exclude keywords that are purely numeric (e.g., "2024")
if keyword.isdigit():
return False
# Rule 3: Exclude stock tickers (e.g., 'aapl', 'tsla', 'googl')
# Tickers are typically 1-5 characters, all lowercase or all uppercase.
if len(keyword) <= 5 and (keyword.islower() or keyword.isupper()) and keyword.isalpha():
# We check if it's NOT in our static keywords to avoid filtering out valid short words.
if keyword not in self.static_keywords:
return False
# Rule 4: Exclude keywords containing digits unless it's a known pattern (e.g., s&p 500)
# This removes terms like 'q1-2024', '10k'.
if any(char.isdigit() for char in keyword):
# Allow specific patterns that are known to be valid
if not re.search(r'(s&p 500|p/e|\d{2,4}q\d|\d{1,2}-k)', keyword):
return False
# Rule 5: Check against an expanded list of common junk words and stop words
expanded_junk_terms = {
'announces', 'reports', 'says', 'inc', 'corp', 'ltd', 'company', 'llc', 'plc',
'what', 'why', 'how', 'when', 'where', 'who', 'you', 'your', 'our', 'the',
'and', 'for', 'with', 'that', 'this', 'news', 'article', 'today', 'new', 'all',
'from', 'shares', 'stock', 'investor', 'market', 'business', 'daily', 'update',
'alert', 'breaking', 'story', 'post', 'view', 'click', 'here', 'details'
}
if keyword in expanded_junk_terms:
return False
# If all checks pass, the keyword is considered valid
return True
def fetch_from_eodhd(self, days_back=30): # pylint: disable=too-many-locals, too-many-branches
"""
Fetch trending financial keywords from EODHD API.
This method queries the EODHD news API to retrieve financial news articles,
extracts keywords from various metadata fields, and filters them based on
frequency and quality criteria.
Args:
days_back (int): Number of days to look back for news articles
Returns:
set: Set of trending financial keywords
"""
API_TOKEN = ' 685e0d635924b8.73169691' # pylint: disable=invalid-name
base_url = 'https://eodhd.com/api/news'
end_date = datetime.now()
start_date = end_date - timedelta(days=days_back)
params = {
'api_token': API_TOKEN,
'from': start_date.strftime('%Y-%m-%d'),
'to': end_date.strftime('%Y-%m-%d'),
'limit': 1000,
'tag': ('finance,economy,market,business,banking,stocks,trading,investment,forex,'
'cryptocurrency,bonds,commodities,inflation,recession,gdp,fed,treasury,'
'interest,rates,earnings,ipo,merger,acquisition,economic-data,'
'central-bank,monetary-policy,fiscal-policy')
}
try: # pylint: disable=too-many-nested-blocks
response = requests.get(base_url, params=params, timeout=60)
if response.status_code == 200:
articles = response.json()
keyword_counts = {}
for article in articles:
metadata_fields = ['tags', 'keywords', 'topics', 'entities', 'concepts']
# Extract keywords from various metadata fields
for field in metadata_fields:
field_data = article.get(field, [])
if not field_data:
continue
keywords_to_process = []
if isinstance(field_data, list):
keywords_to_process = [kw for kw in field_data if isinstance(kw, str)]
elif isinstance(field_data, str):
keywords_to_process = [
kw.strip() for kw in field_data.split(',') if kw.strip()]
for kw in keywords_to_process:
clean_kw = kw.lower().strip()
if (3 <= len(clean_kw) <= 25 and
not (kw.isupper() and len(kw) <= 4) and
not clean_kw.startswith('http') and
(' ' not in clean_kw or len(clean_kw.split()) <= 3)):
keyword_counts[clean_kw] = keyword_counts.get(clean_kw, 0) + 1
title = article.get('title', '')
if title:
quoted_terms = re.findall(r'"([^"]{3,20})"', title)
for term in quoted_terms:
clean_term = term.lower().strip()
if 3 <= len(clean_term) <= 20:
keyword_counts[clean_term] = keyword_counts.get(clean_term, 0) + 1
company_pattern = (
r'\b([A-Z][a-z]+(?:\s+[A-Z][a-z]+)?)\s+'
r'(?:Bank|Corp|Financial|Capital|Fund|Securities|Group|Holdings)'
)
companies = re.findall(company_pattern, title)
for company in companies:
clean_company = company.strip().lower()
if len(clean_company) > 3:
keyword_counts[clean_company] = keyword_counts.get(
clean_company, 0) + 1
min_frequency = 2
filtered_keywords = {}
for kw, count in keyword_counts.items():
if count >= min_frequency and self._is_valid_keyword(kw):
filtered_keywords[kw] = count
trending_keywords = set(filtered_keywords.keys())
if len(trending_keywords) > 0:
# Get the top keywords from the filtered list
top_trending = sorted(
filtered_keywords.items(),
key=lambda x: x[1],
reverse=True)[:15]
sample_keywords = [f"{kw}({count})" for kw, count in top_trending]
logger.info("EODHD top trending examples: %s", sample_keywords)
return trending_keywords
except Exception as e: # pylint: disable=broad-exception-caught
logger.error("EODHD API error: %s", e)
return set()
class HeatedKeywordsAnalyzer: # pylint: disable=too-many-instance-attributes
"""
This class handles the extraction of financial terms from articles, calculates heating scores,
performs LDA topic clustering, and analyzes sentiment of financial terms.
Attributes:
finance_keywords (dict): Comprehensive dictionary of financial keywords by category
all_finance_keywords (set): Flattened set of all financial keywords
keyword_manager (FinancialKeywordManager): Manager for static and trending keywords
lemmatizer (WordNetLemmatizer): NLTK lemmatizer for word normalization
stop_words (set): Stop words for text processing
sentiment_tokenizer (AutoTokenizer): FinBERT tokenizer for sentiment analysis
sentiment_model (AutoModelForSequenceClassification): FinBERT model for sentiment analysis
vectorizer (CountVectorizer): Text vectorizer for LDA
lda_model (LatentDirichletAllocation): LDA model for topic clustering
feature_names (ndarray): Feature names from vectorizer
"""
def __init__(self, article_collection, keyword_collection):
logger.info("Initializing HeatedKeywordsAnalyzer...")
self.article_collection = article_collection
self.keyword_collection = keyword_collection
self.finance_keywords = FIN_KEYWORDS
self.all_finance_keywords = set()
for category_keywords in self.finance_keywords.values():
self.all_finance_keywords.update(category_keywords)
self.keyword_manager = FinancialKeywordManager(self.all_finance_keywords)
self.lemmatizer = WordNetLemmatizer()
self.stop_words = set(stopwords.words('english'))
self.stop_words.update(['said', 'says', 'according', 'report', 'news', 'article'])
self.sentiment_tokenizer = None
self.sentiment_model = None
self._sentiment_initialized = False
self.lda_model = None
self.id2word = None
self.topics_info = {}
self.feature_names = None
self.lda_model = None
self.feature_names = None
self.llm_model = gpt
self.geographic_context = None # Will store the detected geographic context
def _initialize_sentiment_model(self):
"""Lazy initialization of sentiment model."""
if not self._sentiment_initialized:
logger.info("Initializing sentiment analysis model...")
try:
self.sentiment_tokenizer = AutoTokenizer.from_pretrained("yiyanghkust/finbert-tone")
self.sentiment_model = AutoModelForSequenceClassification.from_pretrained(
"yiyanghkust/finbert-tone")
self._sentiment_initialized = True
logger.info("Sentiment analysis model initialized.")
except Exception as e: # pylint: disable=broad-exception-caught
logger.error("Failed to initialize sentiment models: %s", e)
self._sentiment_initialized = True # Prevent retry
def fetch_articles(self, start_date, end_date, sample_size=None):
"""
Fetch articles from MongoDB within a specified date range.
Args:
start_date (datetime): Start date for article retrieval
end_date (datetime): End date for article retrieval
Returns:
list: List of document dictionaries with 'text', 'title', 'content', and 'date' fields
"""
logger.info("Fetching articles from %s to %s", start_date.date(), end_date.date())
if sample_size:
pipeline = [
{"$match": {
"publishDate": {
"$gte": start_date.strftime("%Y-%m-%d"),
"$lte": end_date.strftime("%Y-%m-%d")
}
}},
{"$sample": {"size": sample_size}}
]
cursor = self.article_collection.aggregate(pipeline)
else:
cursor = self.article_collection.find({
"publishDate": {
"$gte": start_date.strftime("%Y-%m-%d"),
"$lte": end_date.strftime("%Y-%m-%d")
}
})
articles_text = [
(f"Title: {doc.get('title', '')}\n"
f"Content: {doc.get('content', '')}\n"
f"sentimentScore: {doc.get('sentimentScore', '')}") for doc in cursor]
documents = []
for article_text in articles_text:
lines = article_text.split('\n')
title = ""
content = ""
sentiment_score = 0
for line in lines:
if line.startswith('Title: '):
title = line[7:]
elif line.startswith('Content: '):
content = line[9:]
elif line.startswith('sentimentScore: '):
sentiment_score = float(line[16:])
if title or content:
full_text = f"{title} {content}".strip()
if len(full_text) > 50:
documents.append({
'text': full_text,
'title': title,
'content': content,
'sentimentScore': sentiment_score,
'date': start_date.strftime("%Y-%m-%d")
})
return documents
def extract_financial_terms(self, text): # pylint: disable=too-many-locals, too-many-branches, too-many-statements
"""
Extract financial terms from text with advanced matching techniques.
Args:
text (str): Text to extract financial terms from
Returns:
list: List of tuples (term, category) of extracted financial terms
"""
text_lower = text.lower()
found_terms = []
used_spans = []
# Build a map of term variations to canonical terms and categories
term_variations_map = {}
for category, terms in self.finance_keywords.items():
for canonical_term in terms:
variations = set()
# Handle multi-word terms
if ' ' in canonical_term:
term_words = canonical_term.split()
variations.add(canonical_term)
# Add lemmatized form
lemmatized_words = [self.lemmatizer.lemmatize(word) for word in term_words]
lemmatized_term = ' '.join(lemmatized_words)
variations.add(lemmatized_term)
# Add plural form if not already plural
if not canonical_term.endswith('s'):
plural_words = term_words[:-1] + [term_words[-1] + 's']
plural_term = ' '.join(plural_words)
variations.add(plural_term)
# Add singular form if plural
if canonical_term.endswith('s') and len(term_words[-1]) > 1:
singular_words = term_words[:-1] + [term_words[-1][:-1]]
singular_term = ' '.join(singular_words)
variations.add(singular_term)
# Handle single-word terms
else:
variations.add(canonical_term)
lemmatized_term = self.lemmatizer.lemmatize(canonical_term)
variations.add(lemmatized_term)
if not canonical_term.endswith('s'):
plural_term = canonical_term + 's'
variations.add(plural_term)
if canonical_term.endswith('s') and len(canonical_term) > 1:
singular_term = canonical_term[:-1]
variations.add(singular_term)
for variation in variations:
if variation and len(variation.strip()) > 0:
term_variations_map[variation] = (canonical_term, category)
# First process multi-word terms (to avoid partial matches)
multi_word_variations = {k: v for k, v in term_variations_map.items() if ' ' in k}
sorted_multi_word = sorted(multi_word_variations.keys(), key=len, reverse=True)
for variation in sorted_multi_word:
canonical_term, category = multi_word_variations[variation]
start = 0
while True:
idx = text_lower.find(variation, start)
if idx == -1:
break
# Check for word boundaries
if ((idx == 0 or not text_lower[idx-1].isalpha()) and
(idx + len(variation) == len(text_lower) or
not text_lower[idx + len(variation)].isalpha())):
# Check if this span overlaps with previously found terms
span_covered = any(
s <= idx < e or s < idx + len(variation) <= e for s, e in used_spans)
if not span_covered:
found_terms.append((canonical_term, category))
used_spans.append((idx, idx + len(variation)))
start = idx + len(variation)
# Then process single-word terms
single_word_variations = {k: v for k, v in term_variations_map.items() if ' ' not in k}
tokens = word_tokenize(text_lower)
offset = 0
for token in tokens:
token_start = text_lower.find(token, offset)
token_end = token_start + len(token)
# Check if this token overlaps with previously found terms
if not any(s <= token_start < e for s, e in used_spans):
if token in single_word_variations:
canonical_term, category = single_word_variations[token]
found_terms.append((canonical_term, category))
offset = token_end
return found_terms
def analyze_sentiment(self, text):
"""
Analyze sentiment of text using FinBERT model.
Args:
text (str): Text to analyze sentiment for
Returns:
tuple: (sentiment_label, confidence_score)
Note:
Sentiment label is one of 'positive', 'neutral', or 'negative'
Confidence score is a float between 0.0 and 1.0
"""
try:
if not self._sentiment_initialized:
self._initialize_sentiment_model()
inputs = self.sentiment_tokenizer(
text[:512],
return_tensors="pt",
truncation=True,
max_length=512
)
with torch.no_grad():
outputs = self.sentiment_model(**inputs)
predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
labels = ['negative', 'neutral', 'positive']
scores = predictions[0].numpy()
predicted_label = labels[np.argmax(scores)]
confidence = float(np.max(scores))
return predicted_label, confidence
except Exception as e: # pylint: disable=broad-exception-caught
logger.error("Sentiment analysis error: %s", e)
return 'neutral', 0.5
def calculate_heating_scores(self, current_docs, previous_docs): # pylint: disable=too-many-locals
"""
Calculate heating scores and perform LDA clustering.
Args:
current_docs (list): List of document dictionaries for current period
previous_docs (list): List of document dictionaries for previous period
Returns:
tuple: (term_frequencies, heating_scores, lda_results)
- term_frequencies: Counter of term frequencies
- heating_scores: Dict mapping terms to heating scores
- lda_results: Dict with LDA clustering results
"""
# Extract terms from current documents
current_terms = []
for doc in current_docs:
doc_terms = [term for term, _ in self.extract_financial_terms(doc['text'])]
current_terms.extend(doc_terms)
current_freq = Counter(current_terms)
heating_scores = {}
# Calculate heating scores if previous documents exist
if previous_docs and len(previous_docs) > 0:
previous_terms = []
for doc in previous_docs:
doc_terms = [term for term, _ in self.extract_financial_terms(doc['text'])]
previous_terms.extend(doc_terms)
previous_freq = Counter(previous_terms)
total_current = sum(current_freq.values())
total_previous = sum(previous_freq.values())
if total_current > 0 and total_previous > 0:
for term in current_freq:
current_pct = (current_freq[term] / total_current) * 100
previous_pct = (previous_freq.get(term, 0) / total_previous) * 100
if previous_pct > 0:
heating_score = ((current_pct - previous_pct) / previous_pct) * 100
else:
heating_score = 100.0 if current_pct > 0 else 0.0
heating_scores[term] = heating_score
else:
for term in current_freq:
heating_scores[term] = 0.0
else:
for term in current_freq:
heating_scores[term] = 0.0
lda_results = self.perform_lda_clustering(current_docs)
return current_freq, heating_scores, lda_results
def filter_topics_by_financial_keywords(self, topics_info):
"""
Filters and enhances topic keywords.
"""
all_financial_keywords = self.all_finance_keywords
for _, topic_info in topics_info.items():
llm_name = topic_info.get('llm_name', 'Unnamed Topic')
words = topic_info['words']
weights = topic_info['weights']
financial_terms = []
financial_weights = []
for word, weight in zip(words, weights):
word_clean = word.replace('_', ' ').lower()
is_financial = False
if word_clean in all_financial_keywords:
is_financial = True
else:
for financial_term in all_financial_keywords:
if ' ' in financial_term and word_clean in financial_term.split():
is_financial = True
break
if word_clean in financial_term or financial_term in word_clean:
is_financial = True
break
if is_financial:
financial_terms.append(word.replace('_', ' '))
financial_weights.append(weight)
if financial_terms:
sorted_pairs = sorted(
zip(financial_terms, financial_weights),
key=lambda x: x[1],
reverse=True
)
topic_info['words'] = [term for term, _ in sorted_pairs[:10]]
topic_info['weights'] = [weight for _, weight in sorted_pairs[:10]]
topic_info['word_weight_pairs'] = sorted_pairs[:10]
topic_info['financial_terms_count'] = len(financial_terms)
else:
topic_info['words'] = [word.replace('_', ' ') for word in words[:5]]
topic_info['weights'] = weights[:5]
topic_info['word_weight_pairs'] = list(zip(
[word.replace('_', ' ') for word in words[:5]],
weights[:5]
))
topic_info['financial_terms_count'] = 0
topic_info['llm_name'] = llm_name
return topics_info
def preprocess_documents(self, documents):
"""
Preprocess documents for LDA analysis while preserving financial terms.
This method performs text preprocessing specifically designed for financial
text analysis, including preserving multi-word financial terms by replacing
spaces with underscores.
Args:
documents (list): List of document dictionaries
Returns:
list: List of preprocessed text strings
"""
processed_texts = []
for doc in documents:
text = doc['text'].lower()
protected_text = text
# Get all financial terms
all_financial_terms = []
for category_terms in self.finance_keywords.values():
all_financial_terms.extend(category_terms)
# Sort terms by length (descending) to prioritize longer matches
sorted_terms = sorted(all_financial_terms, key=len, reverse=True)
# Replace spaces in multi-word terms with underscores to preserve them
for term in sorted_terms:
if ' ' in term:
term_pattern = term.replace(' ', r'\s+')
underscore_term = term.replace(' ', '_')
protected_text = re.sub(
r'\b' + term_pattern + r'\b',
underscore_term,
protected_text
)
# Remove non-alphabetic characters except underscores
text = re.sub(r'[^a-zA-Z\s_]', '', protected_text)
text = ' '.join(text.split())
processed_texts.append(text)
return processed_texts
def _gensim_to_dense_matrix(self, gensim_corpus, num_topics):
"""
Helper to convert gensim's sparse output to a dense numpy array for silhouette scoring.
"""
num_docs = len(gensim_corpus)
dense_matrix = np.zeros((num_docs, num_topics))
for i, doc_topics in enumerate(gensim_corpus):
for topic_id, prob in doc_topics:
dense_matrix[i, topic_id] = prob
return dense_matrix
def find_optimal_topics_gensim(self, corpus, id2word, tokenized_texts, documents_count): # pylint: disable=too-many-locals
"""
Dynamically determines the optimal number of topics for a gensim model.
"""
if documents_count < 100:
topic_range = range(6, min(12, documents_count // 8))
elif documents_count< 300:
topic_range = range(12, min(20, documents_count // 10))
else:
topic_range = range(20, min(35, documents_count // 25))
if len(topic_range) < 2:
return max(2, min(3, documents_count))
logger.info("Testing %d topic configurations with gensim...", len(topic_range))
perplexities, coherence_scores, silhouette_scores = [], [], []
for n_topics in topic_range:
try:
lda_model = LdaModel(corpus=corpus, id2word=id2word, num_topics=n_topics,
random_state=42, passes=10, alpha='auto', eta='auto')
# Metric 1: Perplexity
perplexities.append(lda_model.log_perplexity(corpus))
# Metric 2: Coherence
coherence_model = CoherenceModel(model=lda_model, texts=tokenized_texts,
dictionary=id2word, coherence='c_v')
coherence_scores.append(coherence_model.get_coherence())
# Metric 3: Silhouette
doc_topic_matrix_sparse = lda_model.get_document_topics(
corpus, minimum_probability=0)
doc_topic_matrix_dense = self._gensim_to_dense_matrix(
doc_topic_matrix_sparse, n_topics)
if len(set(np.argmax(doc_topic_matrix_dense, axis=1))) > 1:
silhouette_scores.append(silhouette_score(
doc_topic_matrix_dense,
np.argmax(doc_topic_matrix_dense, axis=1)))
else:
silhouette_scores.append(0.0)
except Exception as e: # pylint: disable=broad-exception-caught
logger.warning("Error evaluating %d topics: %s", n_topics, e)
perplexities.append(float('inf'))
coherence_scores.append(0.0)
silhouette_scores.append(0.0)
if not all([perplexities, coherence_scores, silhouette_scores]):
return max(2, min(5, documents_count // 10))
norm_perp = [(max(perplexities) - p) / (max(perplexities) - min(perplexities))
if max(perplexities) != min(perplexities) else 0.5 for p in perplexities]
norm_coh = [(c - min(coherence_scores)) / (max(coherence_scores) - min(coherence_scores))
if max(coherence_scores) != min(coherence_scores)
else 0.5 for c in coherence_scores]
norm_sil = [(s - min(silhouette_scores)) / (max(silhouette_scores) - min(silhouette_scores))
if max(silhouette_scores) != min(silhouette_scores)
else 0.5 for s in silhouette_scores]
combined_scores = [0.4 * nc + 0.3 * np + 0.3 * ns
for np, nc, ns in zip(norm_perp, norm_coh, norm_sil)]
optimal_topics = list(topic_range)[np.argmax(combined_scores)]
logger.info("Optimal number of topics found: %d", optimal_topics)
return optimal_topics
def perform_lda_clustering(self, documents, n_topics=None):
"""
Perform LDA clustering using gensim, with a sophisticated method for finding
the optimal number of topics.
"""
try:
if len(documents) < 5:
logger.warning("Too few documents for meaningful LDA clustering.")
return None
self.geographic_context = self.detect_geographic_context(documents)
logger.info("Detected geographic context: %s", self.geographic_context)
processed_texts_str = self.preprocess_documents(documents)
tokenized_texts = [text.split() for text in processed_texts_str]
self.id2word = corpora.Dictionary(tokenized_texts)
self.id2word.filter_extremes(no_below=max(2, len(documents) // 20), no_above=0.85)
corpus = [self.id2word.doc2bow(text) for text in tokenized_texts]
if not corpus or not any(corpus):
logger.warning("Corpus is empty after preprocessing.")
return None
if n_topics is None:
n_topics = self.find_optimal_topics_gensim(
corpus, self.id2word, tokenized_texts, len(documents))
logger.info("Fitting FINAL gensim LDA model with %d topics...", n_topics)
self.lda_model = LdaModel(corpus=corpus, id2word=self.id2word, num_topics=n_topics,
random_state=42, passes=15, alpha='auto', eta='auto')
topic_assignments = []
for doc_bow in corpus:
topics = self.lda_model.get_document_topics(doc_bow, minimum_probability=0.3)
best_topic = max(topics, key=lambda item: item[1]) if topics else None
topic_assignments.append(best_topic[0] if best_topic else -1) # pylint: disable=unsubscriptable-object
# Extract topic words with geographic context
self.topics_info = self.extract_topic_words(n_topics)
return {
'topic_assignments': topic_assignments,
'topics_info': self.topics_info,
'n_topics': n_topics,
'total_documents': len(documents),
'corpus': corpus,
'id2word': self.id2word,
'geographic_context': self.geographic_context # Include context in results
}
except Exception as e: # pylint: disable=broad-exception-caught
logger.error("Gensim LDA clustering error: %s", e)
return None
def detect_geographic_context(self, documents):
"""
Detect the primary geographic context from a collection of documents.
Args:
documents (list): List of document dictionaries
Returns:
str: Primary geographic region (e.g., 'China', 'Global', 'US', etc.)
"""
geographic_indicators = {
'china': ['china', 'chinese', 'beijing', 'shanghai', 'yuan', 'renminbi', 'pboc', 'ccp'],
'us': ['united states', 'america', 'american', 'federal reserve',
'fed', 'dollar', 'usd', 'wall street'],
'europe': ['europe', 'european', 'euro', 'ecb', 'brexit', 'eu'],
'japan': ['japan', 'japanese', 'yen', 'tokyo', 'boj'],
'global': ['global', 'worldwide', 'international', 'world']
}
region_scores = {region: 0 for region in geographic_indicators.keys()} # pylint: disable=consider-iterating-dictionary
total_docs = len(documents)
for doc in documents:
text_lower = doc['text'].lower()
for region, indicators in geographic_indicators.items():
for indicator in indicators:
if indicator in text_lower:
region_scores[region] += 1
# Normalize scores by document count
for region in region_scores:
region_scores[region] = region_scores[region] / total_docs if total_docs > 0 else 0
# Determine primary region
primary_region = max(region_scores, key=region_scores.get)
# If no clear geographic context, default to 'Global'
if region_scores[primary_region] < 0.1:
primary_region = 'global'
return primary_region.title()
def name_topic_with_llm(self, topic_keywords, geographic_context=None):
"""
Generates a concise, human-readable name for a topic using an LLM with geographic context.
Args:
topic_keywords (list): List of keywords for the topic
geographic_context (str): Geographic context of the articles
(e.g., 'China', 'US', 'Global')
"""
# Fallback name in case of LLM failure
fallback_name = f"Topic: {', '.join(topic_keywords[:3])}"
if not self.llm_model:
logger.warning("LLM model not available. Using fallback topic name.")
return fallback_name
# Prepare the keywords for the prompt
keywords_str = ", ".join(topic_keywords)
# Construct geographic context string
geo_context_str = ""
if geographic_context and geographic_context.lower() != 'global':
geo_context_str = f" in the {geographic_context} market"
# Construct the prompt with geographic context
system_message = SystemMessage(
content=(
"You are an expert financial analyst. Your task is to analyze a list of keywords "
"from a financial news topic and create a concise, descriptive title for it. "
"Consider the geographic context when creating the topic name."
)
)
human_message = HumanMessage(
content=(
f"Keywords: {keywords_str}\n"
f"Geographic Context: {geographic_context or 'Global'}\n\n"
f"Based on these keywords from financial news{geo_context_str}, "
"what is the core financial theme? "
"Please provide a topic name that is 3-5 words long."
" The name should be clear and professional. "
"If the geographic context is significant to understanding the topic,"
" incorporate it naturally. "
"Do NOT use quotes or any other formatting in your response. Just the topic name."
)
)
try:
response = self.llm_model.invoke([system_message, human_message])
topic_name = response.content.strip().replace('"', '').replace("'", "")
return topic_name.title()
except Exception as e: # pylint: disable=broad-exception-caught
logger.warning("LLM naming failed: %s. Using fallback name.", e)
return fallback_name
def extract_topic_words(self, n_topics, n_top_words=15):
"""
Extracts top words for each topic and adds an LLM-generated descriptive
name with geographic context.
"""
topics_info = {}
if not self.lda_model:
logger.error("LDA model is not available for extracting topic words.")
return topics_info
for topic_idx in range(n_topics):
# Correctly get top words and their weights using the Gensim API
topic_terms = self.lda_model.get_topic_terms(topic_idx, topn=n_top_words)
# The id2word dictionary maps word IDs to the actual words
top_words = [self.id2word[word_id] for word_id, _ in topic_terms]
top_weights = [weight for _, weight in topic_terms]
# Clean up underscores from multi-word terms before processing
cleaned_words = [word.replace('_', ' ') for word in top_words]
llm_generated_name = self.name_topic_with_llm(
cleaned_words[:10],
self.geographic_context
)
topics_info[f'Topic_{topic_idx}'] = {
'llm_name': llm_generated_name,
'words': cleaned_words,
'weights': top_weights,
'word_weight_pairs': list(zip(cleaned_words, top_weights))
}
filtered_topics = self.filter_topics_by_financial_keywords(topics_info)
return filtered_topics
def calculate_topic_sentiments_from_documents(self, lda_results, documents):
"""
Calculate topic sentiments using document-level sentiment scores.
Args:
lda_results (dict): LDA clustering results
documents (list): List of document dictionaries with sentimentScore field
Returns:
dict: {topic_id: sentiment_score} where sentiment_score is -1.0 to 1.0
"""
topic_sentiments = {}
topic_assignments = lda_results.get('topic_assignments', [])
# Group documents by their assigned topic
topic_documents = {}
for i, topic_id in enumerate(topic_assignments):
if topic_id != -1: # Skip unassigned documents
topic_key = f'Topic_{topic_id}'
if topic_key not in topic_documents:
topic_documents[topic_key] = []
topic_documents[topic_key].append(documents[i])
# Calculate average sentiment for each topic
for topic_id, topic_docs in topic_documents.items():
if topic_docs:
# Extract sentiment scores from documents
sentiment_scores = []
for doc in topic_docs:
sentiment_score = doc.get('sentimentScore', 0)
sentiment_scores.append(sentiment_score)
# Calculate average sentiment
if sentiment_scores:
avg_sentiment = sum(sentiment_scores) / len(sentiment_scores)
topic_sentiments[topic_id] = avg_sentiment
else:
topic_sentiments[topic_id] = 0.0
else:
topic_sentiments[topic_id] = 0.0
return topic_sentiments
def get_lda_results(self, lda_results):
"""
Get LDA topic names, document counts and sentiment scores.
"""
topics_info = lda_results.get('topics_info', {})
topic_sentiments = lda_results.get('topic_sentiments', {})
topic_assignments = lda_results.get('topic_assignments', [])
if not topics_info:
logger.warning("No topic information available")
return None
# Calculate document counts for each topic
doc_counts = Counter(t for t in topic_assignments if t != -1)
topic_data = []
# Prepare data in a dictionary
for topic_id, topic_info in topics_info.items():
topic_num = int(topic_id.split('_')[-1])
topic_name = topic_info.get('llm_name', f'Topic {topic_num + 1}')
doc_count = doc_counts.get(topic_num, 0)
sentiment_score = topic_sentiments.get(topic_id, 0.0)
if doc_count > 0:
topic_data.append({
'topic_name': topic_name,
'doc_count': doc_count,
'sentiment_score': sentiment_score
})
return topic_data
def analyze_heated_keywords(self, filter_type, analyzer=None): # pylint: disable=too-many-locals
"""
Analyzes heated keywords for a specific time period.
Fetches articles for the current and previous time periods, calculates heating scores,
performs LDA topic clustering, and analyzes sentiment for financial keywords.
Args:
filter_type (str): Time filter type ('today', 'week', or 'month')
analyzer (HeatedKeywordsAnalyzer, optional): Analyzer instance
Returns:
dict: Results containing term frequencies, heating scores, sentiment scores,
LDA results, categorized terms, term weights, and current documents
"""
if analyzer is None:
analyzer = self
current_start, current_end = get_time_range(filter_type)
prev_start, prev_end = get_previous_time_range(filter_type, current_start)
current_docs = analyzer.fetch_articles(current_start, current_end)
previous_docs = analyzer.fetch_articles(prev_start, prev_end)
if len(current_docs) < 1:
logger.warning("Insufficient documents (%d) for %s analysis",
len(current_docs), filter_type)
return None
term_frequencies, heating_scores, lda_results = analyzer.calculate_heating_scores(
current_docs, previous_docs)
if not term_frequencies:
logger.warning("No financial terms found for %s", filter_type)
return None
sentiment_scores = self.calculate_keyword_sentiments(
current_docs, term_frequencies, analyzer)
total_mentions = sum(term_frequencies.values())
term_weights = {
term: (freq / total_mentions) * 100 for term, freq in term_frequencies.items()}
results = {
'term_frequencies': term_frequencies,
'heating_scores': heating_scores,
'sentiment_scores': sentiment_scores,
'lda_results': lda_results,
'categorized_terms': analyzer.categorize_terms_thematically(term_frequencies),
'term_weights': term_weights,
'current_docs': current_docs
}
return results
def calculate_keyword_sentiments(self, documents, term_frequencies, analyzer):
"""
Calculate sentiment scores for top keywords.
"""
sentiment_scores = {}
top_terms = term_frequencies.most_common(30)
for term, _ in top_terms:
relevant_docs = [doc for doc in documents if term.lower() in doc['text'].lower()]
if relevant_docs:
relevant_sentences = []
for doc in relevant_docs[:5]:
sentences = doc['text'].split('.')
for sentence in sentences:
if term.lower() in sentence.lower():
relevant_sentences.append(sentence.strip())
if relevant_sentences:
combined_text = ' '.join(relevant_sentences[:3])
sentiment, confidence = analyzer.analyze_sentiment(combined_text)
sentiment_scores[term] = (sentiment, confidence)
else:
sentiment_scores[term] = ('neutral', 0.5)
else:
sentiment_scores[term] = ('neutral', 0.5)
return sentiment_scores
def categorize_terms_thematically(self, term_frequencies):
"""
Categorize terms by financial themes.
Groups financial terms into predefined categories based on the finance_keywords dictionary.
Args:
term_frequencies (Counter): Counter object with term frequencies
Returns:
dict: Dictionary with terms grouped by financial categories
"""
categorized = defaultdict(list)
for term, frequency in term_frequencies.items():
found = False
for category, terms in self.finance_keywords.items():
if term in terms:
categorized[category].append({'term': term, 'frequency': frequency})
found = True
break
if not found:
categorized['general'].append({'term': term, 'frequency': frequency})
return dict(categorized)
|