Issues Using Excel Files with Bedrock Knowledge Base – Any Samples?

Question

I’ve successfully used Bedrock Knowledge Base with input data from PDF, TXT, and Word files stored in S3, and the results have been quite good. Now, I’m exploring the use of Excel files (.xlsx) as input data for the Knowledge Base. My use case is quite simple: I’m prompting with basic questions like “Show me all the students whose marks are above 90” but the responses are mostly incorrect.

I’ve experimented with: Multiple Excel files with multiple sheets, A single Excel file with multiple sheets, A single Excel file with only one sheet.  None of these yielded accurate results. The answers returned seem to misinterpret or ignore the actual data.

**My question to the community:**
Are there any sample Excel files and step-by-step instructions for using Excel data effectively with Bedrock Knowledge Base? Any guidance would be greatly appreciated!

svna · Answer

Thank you all for the support. After further attempts and additional research, I resolved the issue by switching to the Structured data store instead of the Vector store.

Initially, the following Amazon Q response led me to proceed with the vector store: "Amazon Bedrock Knowledge Base supports a variety of file types as input data when using a vector store. Here's a comprehensive list of the supported file formats: Plain text (.txt), Markdown (.md), HTML (.html), Microsoft Word (.doc/.docx), CSV (.csv), Microsoft Excel (.xls/.xlsx), PDF (.pdf), JPEG (.jpeg), PNG (.png)"

re:Post Agent · Answer

Based on your experience with Excel files in Amazon Bedrock Knowledge Base, I understand you're facing challenges getting accurate responses to queries about tabular data.

While Excel files (.xls/.xlsx) are indeed supported formats for Bedrock Knowledge Base, there are some best practices that might help improve your results:

1. Consider converting your Excel data to CSV format. CSV files are specifically mentioned as a best practice for including metadata in a data source, especially for large datasets. Using columns to designate content fields and metadata fields can improve data management and retrieval accuracy.

2. Add metadata filtering to your documents. This can significantly improve the relevancy of responses. Your Excel or CSV files should include document metadata attributes or fields that can be used for filtering.

3. For structured/tabular data like student marks, you might want to explore Bedrock Knowledge Base's structured data retrieval capabilities, as this type of query ("Show me all students whose marks are above 90") is more like a database query than a semantic search.

4. Make sure your files don't exceed the 50 MB size limit per file.

5. If you're using multiple sheets in Excel, consider whether the knowledge base is properly processing all sheets or if it might be more effective to split them into separate files.

For your specific use case with student marks, adding metadata for filtering (such as score ranges) might help the system better understand and retrieve the relevant information when queried about score thresholds.  
**Sources**  
[Prerequisites for your Amazon Bedrock knowledge base data - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/knowledge-base-ds.html)  
[Improve search results and accurate in Amazon Bedrock | AWS re:Post](https://repost.aws/knowledge-center/bedrock-knowledge-bases-accurate-results-filters)  
[Amazon Bedrock Knowledge Base Chunks Retrieval Issue | AWS re:Post](https://repost.aws/questions/QUKqtze4wGS5uvsiqc5hGRVg/amazon-bedrock-knowledge-base-chunks-retrieval-issue)

Issues Using Excel Files with Bedrock Knowledge Base – Any Samples?

Add your answer

Relevant content