Is it possible to create an Extractive Q&A pipeline, where the answer is highlighted directly in the source?

[asmith26]

Hi, is it possible to create an Extractive Question & Answering pipeline, where the answer is highlighted directly in the source/context (like in the image)? Thanks!

(image from: https://github.com/deepset-ai/haystack-demos/tree/main/healthcare)

[dosubot[bot]]

Hey @asmith26! Great to see you back with another intriguing question 🌟. How's the quest for knowledge treating you? 🚀

Yes, it is possible to create an Extractive Question & Answering pipeline in LlamaIndex where the answer is directly highlighted in the source or context. The key components involved in creating such a pipeline include:

1. IngestionPipeline: For ingesting and indexing documents. You can find more details about its structure and functionality in the llama-index-core/llama_index/core/ingestion/pipeline.py file.

2. QueryPipeline: For processing queries and retrieving relevant documents or passages.

3. CondenseQuestionChatEngine: For generating answers based on the queries and the context provided by the retrieved documents.

To adapt this pipeline for extractive Q&A where answers are highlighted in the source, you would need to modify the response synthesizer component (e.g., CompactAndRefine) to include functionality for highlighting the answer in the context. This might involve adjusting the response generation logic to return the start and end indices of the answer within the context or directly modifying the context to include highlight markers around the answer.

Also, ensure the query processing and document retrieval steps are fine-tuned to support extractive Q&A. This might involve using embeddings or models that are specifically trained for extractive tasks.

The existing code structure supports the integration of these components and modifications. However, specific implementation details for highlighting answers within the source context are not provided in the snippet. You would need to implement or extend the response synthesizer logic to include highlighting functionality based on your requirements.

You can refer to the code snippet from llama_index/legacy/command_line/rag.py in the LlamaIndex repository for a basic structure of setting up a Q&A pipeline using LlamaIndex's legacy components. This pipeline involves the ingestion of documents, querying, and response generation, which can be adapted to highlight answers directly in the source.

I hope this helps! If you have any more questions or need further clarification, feel free to ask.

Sources

• llama-index-legacy/llama_index/legacy/command_line/rag.py
• llama-index-core/llama_index/core/ingestion/pipeline.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.

[asmith26]

Ok, I think this feature is more about using a extractive question answering model, e.g. https://huggingface.co/tasks/question-answering