How to handle User Summarization Queries in Azure Cognitive Search

Question

I have implemented a system where users can upload files via a UI file upload screen. These files are then stored in Azure Blob Storage and indexed into Azure Cognitive Search. Later, users can query this indexed content through a UI chat screen.

The process involves the following steps:

1. File Upload: Files are uploaded and indexed.
2. Query Handling: When a user enters a question, Azure Cognitive Search is used to find related content.
3. Content Processing: The question and the related content are sent to OpenAI ChatGPT for further processing.

We encounter an issue when users ask for a summary or overview of the file content (e.g., "summarize this file," "provide an overview," or "give an abstract"). The problem is that specific terms like "summarize," "overview," or "abstract" might not be present in the indexed content. As a result, Azure Cognitive Search cannot generate any related content for these types of queries, leading to challenges in processing the user's request effectively.

I am seeking advice on how to address this issue given the current capabilities of Azure Cognitive Search and the integration with OpenAI ChatGPT.

Answer 1

Su Myat Hlaing , Apologies for the delayed response.

Just to clarify, Is the problem related to whether the user requests a summary or overview of the document they've recently uploaded, or of any document they're searching for?
You may take a look at some pre-filtering of the index

Vector query filters - Azure AI Search | Microsoft Learn

Based on my understanding of your scenario/issue description, GraphRAG is designed to  meet the use case you are describing, assuming large corpus of docs to summarize against - please checkout the doc: Welcome to GraphRAG (microsoft.github.io)

if the summarization question is always against the document context that was just loaded, then AI Language Summarization will perform best: What is summarization? - Azure AI services | Microsoft Learn, / If you are asking for document summaries of previously indexed content than GraphRAG will scale best. 

--- 

> please click Accept Answer - it will benefit the community/users to find the answer quickly.