[Question]: What slicing strategies are available in ragflow?

[Monkeycrop]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

Describe your problem

I'd like to know what slicing strategies are available in Ragflow?

Because when parsing documents, I often encounter situations where a complete sentence is split into two fragments, resulting in the loss of semantic information.How can I avoid this situation? @dosu

[dosubot]

Ragflow supports several slicing (chunking) strategies, including naive, manual, qa, general, paper, book, and law, which you can configure at the dataset or document level using parameters like chunk_token_num and delimiter. For best results in preserving sentence or semantic integrity, use the SentenceSplitter or SemanticSplitter methods—these are specifically designed to avoid splitting sentences or paragraphs and are preferable to general or length-based chunking, which can break up sentences and reduce coherence. You can set these options when creating or updating a dataset or document before parsing; if needed, update the document config after upload to switch strategies.

Chunking parameters like chunk_token_num (soft upper bound for chunk size), delimiter (for custom splitting), and overlapped_percent (for overlapping chunks) let you fine-tune how documents are sliced. If you want strict control, adjust the delimiter and chunk size, but note that the slicing method logic may override these settings. For best semantic preservation, always choose SentenceSplitter or SemanticSplitter as your chunking method at the dataset or document level before parsing.

If you need to preserve context across chunks, consider using overlapping chunks or post-processing retrieval results to fetch adjacent chunks, since Ragflow treats all chunks independently by default.

References:

• Semantic slicing methods and configuration
• Chunking strategies and parameters
• Delimiter and chunk size tuning
• Overlapping chunks and context preservation

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Monkeycrop]

However, I don't see the SentenceSplitter or SemanticSplitter splitting methods in the ragflow UI. @dosu

[dosubot]

You’re right—there are no “SentenceSplitter” or “SemanticSplitter” options in the Ragflow UI. Ragflow’s chunking methods (like General, Q&A, Resume, Manual, Paper, etc.) are all based on configurable parameters such as chunk size and delimiters, but there aren’t dedicated sentence- or semantic-aware splitters exposed in the UI or backend under those names.

If you want to avoid splitting sentences, set the delimiter to sentence-ending punctuation (like . or 。；！？) and adjust the chunk size to a value that fits your needs. This way, Ragflow will try to split at those delimiters, which helps preserve sentence integrity. For more semantic chunking (like by section or heading), use modes like “Paper” or “Book,” which leverage document structure, or tweak the delimiter and chunk size in “Manual” or “General” mode.

There’s no built-in, selectable “semantic” or “sentence” splitter, so the best approach is to carefully configure delimiters and chunk size to match your document’s structure and language. If you need more advanced or custom chunking, you’d need to wait for future updates that allow custom chunking logic in the data ingestion pipeline.

Let me know if you want tips on delimiter settings for your specific document type!

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[KevinHuSh]

Could you share the file with us and specify which part is not proper enough?
kai.hu@infiniflow.org