fix: QUALITY_THRESHOLD 0.5→0.3, prompt scoring más generoso
Build & Deploy ResearchOwl / build-and-push (push) Successful in 5s
Build & Deploy ResearchOwl / build-and-push (push) Successful in 5s
This commit is contained in:
+1
-1
@@ -31,7 +31,7 @@ class Settings(BaseSettings):
|
||||
# Processing
|
||||
chunk_size: int = Field(800, env="CHUNK_SIZE") # tokens per chunk
|
||||
chunk_overlap: int = Field(100, env="CHUNK_OVERLAP")
|
||||
quality_threshold: float = Field(0.5, env="QUALITY_THRESHOLD") # 0-1, chunks below discarded
|
||||
quality_threshold: float = Field(0.3, env="QUALITY_THRESHOLD") # 0-1, chunks below discarded
|
||||
|
||||
# App
|
||||
log_level: str = Field("INFO", env="LOG_LEVEL")
|
||||
|
||||
@@ -227,6 +227,8 @@ class ContentProcessor:
|
||||
import anthropic
|
||||
prompt = (
|
||||
f'Rate 0-10 how relevant this text is to the topic "{topic}". '
|
||||
f'Be generous — if the text is tangentially related, score 4+. '
|
||||
f'Only score below 3 if completely unrelated. '
|
||||
f'Reply with only a number.\n\nText:\n{chunk[:500]}'
|
||||
)
|
||||
try:
|
||||
@@ -251,7 +253,7 @@ class ContentProcessor:
|
||||
normalized = min(1.0, score / 10.0)
|
||||
logger.debug("Claude relevance score", raw=score, normalized=round(normalized, 2))
|
||||
return normalized
|
||||
return 0.6
|
||||
return 0.5
|
||||
except Exception as e:
|
||||
logger.warning("Claude scoring failed, falling back to Ollama", error=str(e))
|
||||
return await self._score_with_ollama(chunk, topic)
|
||||
|
||||
Reference in New Issue
Block a user