highperfocused/danswer
1
0
Fork 0
mirror of https://github.com/danswer-ai/danswer.git synced 2025-09-20 04:37:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

More permissive quote matching (#295)

Browse Source
This commit is contained in:
Yuhong Sun
2023-08-14 15:03:21 -07:00
committed by GitHub
parent 59db40cf36
commit 848e5653a9
1 changed files with 11 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
backend/danswer/utils/text_processing.py
View File

@@ -1,3 +1,5 @@
import re
from bs4 import BeautifulSoup
from danswer.configs.constants import HTML_SEPARATOR
@@ -15,23 +17,17 @@ def clean_model_quote(quote: str, trim_length: int) -> str:
def shared_precompare_cleanup(text: str) -> str:
"""LLMs models sometime restructure whitespaces or edits special characters to fit a more likely
distribution of characters found in its training data, but this hurts exact quote matching
"""
text = text.lower()
# GPT models like to return cleaner spacing, not good for quote matching
text = "".join(text.split())
# GPT models sometimes like to clean up bulletpoints represented by *
text = text.replace("*", "")
# GPT models sometimes like to edit the quoting, ie "Title: Contents" becomes Title: "Contents"
text = text.replace('\\"', "")
text = text.replace('"', "")
# GPT models often change up punctuations to make the text flow better.
text = text.replace(".", "")
text = text.replace(":", "")
text = text.replace(",", "")
text = text.replace("-", "")
# \s: matches any whitespace character (spaces, tabs, newlines, etc.)
# |: acts as an OR.
# \*: matches the asterisk character.
# \\": matches the \" sequence.
# [.,:`"#-]: matches any character inside the square brackets.
text = re.sub(r'\s|\*|\\"|[.,:`"#-]', "", text)
return text