Reviewing Extracted Entities

When you upload documents to Upstream, the system automatically extracts entities -- pain points, objections, feature requests, and more. Entities that are extracted with high confidence are added to your knowledge base automatically. Entities below 80% confidence are sent to the review queue for you to check before they go live.

The three sections

The review queue is organized into three sections, each addressing a different kind of attention your knowledge base needs.

Pending

These are newly extracted entities that need your approval. Each entry shows the extracted content, its confidence score, the entity type, and the source document it came from. Review each one and decide whether it belongs in your knowledge base.

Duplicates

Upstream identifies entities that look like they might be duplicates of something already in your knowledge base. For each potential duplicate, you will see the new entity alongside the existing match. You can merge them into a single entity (keeping the best information from both) or mark them as distinct if they are actually different.

Stale

Entities that have not been validated in a while appear here. Entities untouched for 30 days get a soft flag; those untouched for 90 days get a stronger one. This helps you keep your knowledge base current -- information that was true six months ago may not reflect today's reality.

Approving and rejecting entities

For each pending entity, you have two options:

Approve -- The entity is added to your knowledge base and becomes available in search, the graph, chat answers, and artifact generation.
Reject -- The entity is discarded. Use this when the extraction is incorrect, irrelevant, or too vague to be useful.

Take a moment to read the extracted content and check it against the source. If the content is mostly right but needs a small edit, you can approve it and then edit the entity details from the Knowledge Browser.

Batch operations

When you have many entities to review, use batch operations to speed things up. Select multiple entities using the checkboxes, then approve or reject them all at once. This is especially useful after a large document upload that generates dozens of extractions.

Why review matters

Your knowledge base is only as good as the information in it. Reviewing entities ensures that:

Accuracy stays high -- Bad extractions do not pollute your search results or AI-generated answers.
Duplicates stay merged -- You avoid seeing the same insight repeated across multiple entities.
Stale data gets refreshed -- Regular review keeps your knowledge base reflecting current reality, not outdated assumptions.

Think of the review queue as quality control for your knowledge base. A few minutes of review after each upload pays off every time you search, ask a question, or generate an artifact.

Using the Knowledge Browser -- Browse and explore your approved entities
Asking Questions with Chat -- Get AI-powered answers grounded in your reviewed knowledge
Creating Artifacts -- Generate deliverables from your verified entities

Still need help?

Email us at hello@getupstream.ai -- we read every message.