Skip to content

fix: Prevent keyword matcher from applying all labels #733

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
stranske merged 2 commits into from
Jan 10, 2026
Merged

fix: Prevent keyword matcher from applying all labels #733

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3407399
fix: Prevent keyword matcher from applying all labels
stranske Jan 10, 2026
c3e0699
Merge branch 'main' into fix/label-matcher-keyword-false-positives
stranske Jan 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
109 changes: 102 additions & 7 deletions scripts/langchain/label_matcher.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,95 @@ class LabelMatch:
KEYWORD_FEATURE_SCORE = 0.9
KEYWORD_DOCS_SCORE = 0.9
_IGNORED_LABEL_TOKENS = {"type", "kind"}
# Common words that appear in label descriptions but shouldn't trigger keyword matching
_COMMON_STOPWORDS = {
"this",
"that",
"the",
"a",
"an",
"is",
"are",
"or",
"and",
"for",
"to",
"of",
"in",
"on",
"with",
"be",
"it",
"not",
"if",
"by",
"as",
"at",
"from",
"has",
"have",
"can",
"will",
"would",
"should",
"may",
"might",
"must",
"need",
"issue",
"issues",
"pull",
"request",
"requests",
"information",
"further",
"already",
"exists",
"changes",
"additions",
"improvements",
"triggers",
"analysis",
"suggestions",
"formatted",
"template",
"format",
"optimize",
"optimization",
"new",
"code",
"clean",
"only",
"create",
"follow",
"up",
"verification",
"acceptance",
"criteria",
"checkbox",
"evaluate",
"evaluation",
"compare",
"comparison",
"multiple",
"models",
"providers",
"decompose",
"break",
"down",
"large",
"tasks",
"smaller",
"maintainable",
"requires",
"human",
"intervention",
"attention",
"help",
"wanted",
"good",
"first",
}
_BUG_KEYWORDS = {
"bug",
"bugs",
Expand Down Expand Up @@ -246,19 +335,25 @@ def _keyword_match_score(label: LabelRecord, query: str) -> float | None:
return None

query_lower = query.lower()
label_text = " ".join(part for part in (label.name, label.description) if part)
label_tokens = _tokenize(label_text) - _IGNORED_LABEL_TOKENS
if label_tokens and label_tokens.intersection(tokens):
# Only match on significant tokens, excluding common stopwords
significant_tokens = tokens - _COMMON_STOPWORDS

# Require label NAME to appear in query for high-confidence keyword match
# (not just overlapping description tokens)
label_name_tokens = _tokenize(label.name) - _IGNORED_LABEL_TOKENS
if label_name_tokens and label_name_tokens.intersection(significant_tokens):
Comment thread
stranske marked this conversation as resolved.
return 0.95

normalized = _normalize_label(label_text)
# Use label NAME only (not description) for category matching to avoid false positives
# e.g., "duplicate" description contains "request" but shouldn't match feature keywords
label_name_normalized = _normalize_label(label.name)
score = 0.0

if "bug" in normalized and any(
if "bug" in label_name_normalized and any(
_token_matches_keyword(token, keyword) for token in tokens for keyword in _BUG_KEYWORDS
):
score = max(score, KEYWORD_BUG_SCORE)
if any(tag in normalized for tag in ("feature", "enhancement", "request")) and (
if any(tag in label_name_normalized for tag in ("feature", "enhancement", "request")) and (
any(
_token_matches_keyword(token, keyword)
for token in tokens
Expand All @@ -267,7 +362,7 @@ def _keyword_match_score(label: LabelRecord, query: str) -> float | None:
or any(phrase in query_lower for phrase in _FEATURE_PHRASES)
):
score = max(score, KEYWORD_FEATURE_SCORE)
if "doc" in normalized and any(
if "doc" in label_name_normalized and any(
_token_matches_keyword(token, keyword) for token in tokens for keyword in _DOCS_KEYWORDS
):
score = max(score, KEYWORD_DOCS_SCORE)
Expand Down
7 changes: 4 additions & 3 deletions tests/scripts/test_label_matcher.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -249,9 +249,10 @@ def test_find_similar_labels_keyword_multicategory_match():


def test_find_similar_labels_keyword_docs_description_match():
"""Test that labels with 'doc' in their NAME match docs keywords."""
labels = [
label_matcher.LabelRecord(name="type:bug"),
label_matcher.LabelRecord(name="quality", description="Documentation updates"),
label_matcher.LabelRecord(name="type:documentation", description="Documentation updates"),
]
vector_store = label_matcher.LabelVectorStore(
store=object(), provider="unit-test", model="unit-test-model", labels=labels
Expand All @@ -261,8 +262,8 @@ def test_find_similar_labels_keyword_docs_description_match():

names = {match.label.name for match in matches}
assert "type:bug" in names
assert "quality" in names
doc_match = next(match for match in matches if match.label.name == "quality")
assert "type:documentation" in names
doc_match = next(match for match in matches if match.label.name == "type:documentation")
assert doc_match.score >= label_matcher.KEYWORD_DOCS_SCORE


Expand Down
Loading