Docs
Question Answering

Question Answering

The Question and Answer (Q&A) module of Unbody is designed to extract answers from your data, transforming it into a resource for direct query responses.

The Q&A module introduces an ask {} operator available in GraphQL Get queries. It returns the answer with the highest certainty level, ensuring reliable and accurate responses.

Default Q&A Module

The default module operates on the bert-large-uncased-whole-word-masking-finetuned-squad (uncased) model, providing a balance between performance and accuracy. It follows a two-step process as given below.

  1. Semantic Search: It locates documents likely to contain the answer.
  2. Answer Extraction: It uses BERT-style extraction on text and string properties of the document.

Outcomes:

  • No answer was found.
  • Answer found but below the specified minimum certainty.
  • An answer meeting the desired certainty is returned.

For further details, you can refer to the Weaviate documentation (opens in a new tab).

OpenAI Q&A Module

💡

OpenAI’s Q&A is best suited for complex query contexts or when you require the advanced capabilities of OpenAI's models. It provides nuanced and context-aware responses, making it a powerful tool for intricate query needs.

Leveraging OpenAI’s completions endpoint, this module also adopts a two-step process:

  1. Semantic Search: Semantic search identifies potential documents.
  2. Answer Generation: It generates a prompt for OpenAI, which then extracts the answer.

The outcomes are similar to the default module as described above.

For a more detailed explanation, you can visit the Weaviate documentation (opens in a new tab).

Conclusion

Unbody’s Q&A module transforms your data into a knowledge base that can be queried, providing precise answers and enhancing user interaction. Choose the module that best suits your needs to start leveraging this capability today.

Happy developing!