---
title: Sentiment & Emotions | Formbricks Hub
description: Hub can classify open-text feedback into a sentiment label and score, and into the emotions it expresses, and store them on the feedback record for analytics and filtering.
---

Hub can enrich open-text feedback with two related classifications: **sentiment** — the overall polarity of the text — and **emotions** — the specific emotions it expresses. Both run asynchronously and are stored on the same feedback record, so you can filter, segment, and aggregate qualitative feedback without reading every response.

They are independent enrichments: a deployment can run either, both, or neither, and each is separately gated per tenant.

Like translation, sentiment and emotion enrichment are additive and asynchronous: the original `value_text` is never changed, ingestion is never blocked, and the results appear shortly after a record is created or updated.

## The fields

Sentiment and emotions add three read-only fields to every [feedback record](/core-concepts/data-model/index.md). All are server-generated — you never set them when creating or updating a record — and are `null` until the record is enriched.

| Field             | Type            | Description                                                                                    |
| ----------------- | --------------- | ---------------------------------------------------------------------------------------------- |
| `sentiment`       | String          | Overall sentiment label of the text (for example `positive`, `neutral`, `negative`).           |
| `sentiment_score` | Number          | Signed polarity from `-1.0` (most negative) to `1.0` (most positive).                          |
| `emotions`        | Array of String | The emotions the text expresses; empty when none clearly apply (stored as `null`, never `[]`). |

See the [API Reference](/api/index.md) for the exact `sentiment` and `emotions` value sets.

## Enabling

Each enrichment runs only where the **deployment has a provider configured** — an operator sets `SENTIMENT_PROVIDER` + `SENTIMENT_MODEL` and/or `EMOTIONS_PROVIDER` + `EMOTIONS_MODEL` (plus provider credentials); see [Environment Variables](/reference/environment-variables#enrichment/index.md). With no provider, that enrichment is off for the whole deployment.

Where a provider is configured, each enrichment is **on by default per tenant** and can be turned off with the [`sentiment_enabled` / `emotions_enabled`](/core-concepts/tenant-settings/index.md) switches. Setting a switch to `false` stops Hub producing that enrichment for the tenant going forward; unset (or `null`) means enabled.

## How it works

When a text feedback record is created, or its `value_text` changes, Hub enqueues the enrichment jobs for whichever of sentiment and emotions are enabled. A background worker calls the provider and writes the result back to the record. The work happens off the ingestion path, so creating feedback is never slowed down or blocked.

- **Only open text is classified** — records whose `field_type` is `text` with a non-empty `value_text`. Numeric, boolean, date, and categorical answers are left untouched.
- **Edits re-classify.** Changing `value_text` clears the stored sentiment/emotions and re-runs the classification; emptying `value_text` clears them without re-running. During the brief window between the edit and the worker finishing, the fields read back as `null` (recomputing) rather than a stale value.
- **A record can express several emotions, or none.** `emotions` is a set (order is not significant); “no clear emotion” is `null`, not an empty array.

Sentiment and emotions are **eventually consistent** — populated shortly after ingestion, not synchronously. Read them defensively and treat `null` as “not classified yet”.

## Backfilling existing records

Enrichment runs automatically on new and edited records. To (re)classify an existing backlog — after first enabling an enrichment, or to recover records whose job was lost or finally failed — run the bundled backfill for the type, which enqueues jobs for records missing that classification:

Terminal window

```
# Run inside the Hub deployment, with the SENTIMENT_*/EMOTIONS_* provider and DATABASE_URL configured
./backfill-classify -type sentiment
./backfill-classify -type emotions
```

It is safe to re-run: records already classified are skipped. Re-enabling a tenant’s switch does not re-classify its backlog on its own — run the backfill to catch records edited while the enrichment was off.

## Relationship to translation and embeddings

Sentiment and emotions are independent of [translation](/core-concepts/translated-feedback/index.md) and of the embeddings used for [semantic search](/guides/hub-self-hosted-embeddings/index.md). All classify the original, source-language `value_text`; enabling or disabling one does not affect the others.

## Querying

Because the results live on the record, you can filter and aggregate straight from SQL or any connected BI tool:

```
SELECT sentiment, COUNT(*)
FROM feedback_records
WHERE field_type = 'text'
  AND tenant_id = 'org-123'
  AND sentiment IS NOT NULL
GROUP BY sentiment;
```

## Next Steps

- [Tenant Settings](/core-concepts/tenant-settings/index.md) - Toggle `sentiment_enabled` / `emotions_enabled` per tenant
- [Data Model](/core-concepts/data-model/index.md) - How feedback records are stored, including enrichment fields
- [Environment Variables](/reference/environment-variables#enrichment/index.md) - Configure the sentiment and emotion providers