Are there any plans to implement "Sentence Window Retrieval" and/or "Auto-merging Retrieval" similar to what is available in llamaindex library?

[peterlitvak]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

n/a

Description

There are two very interesting retrieval algorithms that are implemented in llamaindex library, the "Sentence Window Retrieval" and "Auto-merging Retrieval". They appear to improve relevance quite a bit.
Are there any plans to implement these in LangChain?

System Info

n/a

[dosubot[bot]]

🤖

Hey there @peterlitvak! Great to see you back with another insightful question. Hope you're doing awesome!

Based on the current version of LangChain (v0.1.7), it appears that the framework does not support retrieval algorithms similar to "Sentence Window Retrieval" and "Auto-merging Retrieval" as found in the llamaindex library. The retrieval algorithms currently implemented in LangChain include:

1. GoogleVertexAISearchRetriever
2. GoogleVertexAIMultiTurnSearchRetriever
3. GoogleCloudEnterpriseSearchRetriever
4. TFIDFRetriever

The TFIDFRetriever is a term frequency-inverse document frequency (TF-IDF) based retriever. You can find more details about these retrievers in the google_vertex_ai_search.py and tfidf.py files in the repository.

As for the implementation of "Sentence Window Retrieval" and "Auto-merging Retrieval" in LangChain, I'm unable to provide a definitive answer as I don't have access to the roadmap or future plans for the LangChain repository. It would be best to reach out to the repository maintainers or contributors for more accurate information.

If you have more details about the specific functionalities of the "Sentence Window Retrieval" and "Auto-merging Retrieval" algorithms, it would be helpful to understand how they could potentially be integrated into LangChain.

Sources

• libs/langchain/langchain/retrievers/google_vertex_ai_search.py
• libs/langchain/langchain/retrievers/tfidf.py

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[minhkids]

As far as I can see, the Sentence Window Retrieval is a pretty good retriever and I'm trying to rewrite the code from the llama-index to make it fit LangChain (hopefully Langchain will extend its Retrieval). If you want to find 1 retriever similar to the sentence window, then I suggest using ContextualCompressionRetriever. ✌️

[ayseozgun]

is there any update here?

[minhkids]

I figured out that the code in Llama-index is quite simple the code will be like this:

def get_window_context(sentences, sentence_idx):
     window_size = 3
     start_idx = max(0, sentence_idx - window_size)
     end_idx = min(sentence_idx + 1 + window_size, len(sentences))
     left_padding = max(0, window_size - sentence_idx)
     right_padding = max(0, sentence_idx + window_size + 1 - len(sentences))
     left_context = ["-1"] * left_padding + sentences[start_idx:sentence_idx]
     right_context = sentences[sentence_idx + 1:end_idx] + ["-1"] * right_padding
     context = left_context + ["0"] + right_context
     context = [item for item in context if item != "-1"]  
     context = " ".join(context)
     return context.strip()

I don't know if I did it right or wrong but I got similar output with sentence retrieval from llama-index.( I use metadata for the index )

[rsk2327]

@ayseozgun I have created a PR to implement Sentence window retrieval. Its only implemented for FAISS vector store yet. But the core code should be the same for all vector databased.

Can check the PR here : #20392

[AliHaider20]

I see that this wasn't merged, so is there a new PR?