感觉知识库嵌入和召回的能力好差，有什么改进办法

以往在毕昇，dify 中的 RAG 能力还相对好点，最近用上了 coze ，发现比较难匹配到 PDF 文档中的内容，用的默认的文档解析引擎。有没有什么改进办法呢，我发现默认的分段会很多，召回也比较差

No file, test, or entry point is named. Start by reproducing retrieval on the referenced PDF with the default parser, then trace how parsing, chunking, embedding, and recall are configured. Done would require an agreed improvement and evidence that matching and recall improve.

Written by the indexing model from the issue text.