Search Steps

You can have one or multiple search steps. If you have more than one you need to use a classification step to select the right one.

If you only have one or don't have a classification step, the name does not matter as the system will select the first one.

Select New Search Step to create an additional step.
Step Name - Enter the name of the step. If you are using query classification this must match the classification models label.
Search Type - Select the search type from the dropdown.
- Standard - Uses the standard BM25 Keyword Search.
- Cosine Similarity - Uses a dense vector and cosine similarity.
- Rank Feature - Uses a rank feature (sparse vector) search.
AI Model Service - Select the model service that will provide the vectorisation from the dropdown.
- This is not needed for standard searches.
AI Model Provider - Select the provider that matches the selected search type.
- Cosine Similarity -> Huggingface Sentence Transformers
- Rank Feature -> Huggingface Sparse Vector
Model - Select the model for the vectorisation from the drop down.
- If using a dense vector the dimensions must match.

Vector - For Cosine Similarity select the right vector that was populated during enrichment.
Bucket Size - If you want to use cosine similarity over larger result sets, configure this between 50 and 500. Some testing will need to be completed to determine the best value.
Rank Feature - For Rank Features select the right rank feature that was populated during enrichment.

Select Add new term matches.
Field - Select the field to match from the dropdown.
Term Matches - Start typing the terms to match it with. You can add multiple terms here.
Select Add

Last updated 6 months ago