[Bug]: Mishandling of section title when chunking a PDF with tables

[predoctech]

Is there an existing issue for the same bug?

• I have checked the existing issues.

RAGFlow workspace code commit ID

1160b58

RAGFlow image version

demo.ragflow.io

Other environment information

No response

Actual behavior

This PDF contains various section title followed by tables. When parsing these structures it seems like the titles will be adjoined to the texts before and after, thus distorting the meaning (or spelling) of these important title texts. Thus during a chat involving this PDF if you ask a question whether the document contains a section with this title, the reply is "NO", which is outright misleading.
Look for the title "MANAGEMENT DISCUSSION AND ANALYSIS" in the following screenshot:

Expected behavior

Section and table title should keep their semantic meaning and spelling intact during chunking.

Steps to reproduce

Upload the file, and carry out parsing using the "General" template.

Additional information

I suppose the embedding model chosen is irrelevant for this issue, but FYI embedding model used was Gemini.