3 Advanced vector retrieval strategies

In the previous chapter of this book, you have learned about the basics of text embeddings and vector similarity search. By converting text into numerical vectors, you have seen how machines can understand the semantic meaning of content. Combining text embedding and vector similarity search techniques allows for an optimized and accurate retrieval of relevant unstructured text from vast amounts of documents, enabling more accurate and up-to-date answers in RAG applications. Suppose you have implemented and deployed a RAG application as described in Chapter 1. After some testing, you and the users of the RAG applications noticed that the accuracy of the generated answers is lacking due to incomplete or irrelevant information in the retrieved documents. Consequently, you have been assigned the task of enhancing the retrieval system to improve the accuracy of the generated answers.

3.1 Parent document retriever

3.2 Complete RAG pipeline

3.3 Summary

3.4 References