File size: 3,159 Bytes
34b531b | 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 | from __future__ import annotations
import re
import unicodedata
from app.config import (
CONTEXT_COMPRESSION_ENABLED,
CONTEXT_MAX_CHARS_PER_CHUNK,
CONTEXT_MAX_SENTENCES_PER_CHUNK,
CONTEXT_MIN_SENTENCE_CHARS,
)
from app.schemas import RetrievedChunk
FINANCE_TERMS = {
"doanh",
"thu",
"loi",
"nhuan",
"lnst",
"eps",
"roe",
"roa",
"bien",
"lai",
"no",
"vay",
"tang",
"giam",
"gia",
"muc",
"tieu",
"khuyen",
"nghi",
"co",
"phieu",
"rsi",
"macd",
"volume",
"khoi",
"luong",
"thanh",
"khoan",
}
STOPWORDS = {
"anh",
"bao",
"cho",
"co",
"cua",
"khong",
"hay",
"la",
"mot",
"nhung",
"the",
"thi",
"trong",
"va",
"ve",
"voi",
}
def normalize_text(text: str) -> str:
decomposed = unicodedata.normalize("NFD", str(text).lower())
without_accents = "".join(char for char in decomposed if unicodedata.category(char) != "Mn")
return re.sub(r"\s+", " ", without_accents).strip()
def tokens(text: str) -> set[str]:
return {
token
for token in re.findall(r"[\w]+", normalize_text(text), flags=re.UNICODE)
if len(token) > 2 and token not in STOPWORDS
}
def sentence_candidates(text: str) -> list[str]:
normalized = re.sub(r"\s+", " ", str(text)).strip()
if not normalized:
return []
pieces = re.split(r"(?<=[.!?。！？])\s+|\n+|(?<=;)\s+", normalized)
return [piece.strip(" -:\t") for piece in pieces if len(piece.strip()) >= CONTEXT_MIN_SENTENCE_CHARS]
def compact_text(text: str, limit: int) -> str:
compact = " ".join(str(text).split())
if len(compact) <= limit:
return compact
return compact[: limit - 3].rstrip() + "..."
def sentence_score(sentence: str, query_tokens: set[str], ticker: str) -> float:
sentence_tokens = tokens(sentence)
if not sentence_tokens:
return 0.0
overlap = len(sentence_tokens & query_tokens)
finance_overlap = len(sentence_tokens & FINANCE_TERMS)
score = overlap * 2.0 + finance_overlap * 0.35
if ticker and ticker.lower() in sentence.lower():
score += 1.0
if re.search(r"\d", sentence):
score += 0.5
return score
def compress_chunk_text(query: str, chunk: RetrievedChunk) -> str:
if not CONTEXT_COMPRESSION_ENABLED:
return chunk.text
sentences = sentence_candidates(chunk.text)
if not sentences:
return compact_text(chunk.text, CONTEXT_MAX_CHARS_PER_CHUNK)
query_tokens = tokens(query)
ranked = sorted(
enumerate(sentences),
key=lambda item: sentence_score(item[1], query_tokens, chunk.ticker),
reverse=True,
)
selected_indexes = sorted(
index for index, sentence in ranked[:CONTEXT_MAX_SENTENCES_PER_CHUNK] if sentence_score(sentence, query_tokens, chunk.ticker) > 0
)
if not selected_indexes:
return compact_text(chunk.text, CONTEXT_MAX_CHARS_PER_CHUNK)
compressed = " ".join(sentences[index] for index in selected_indexes)
return compact_text(compressed, CONTEXT_MAX_CHARS_PER_CHUNK)
|