Entity-Level Sentiment

BRAD API v2.3.0 adds entitySentiments to most event responses. It complements the event-level sentiment score by showing how the text refers to each named entity.

See the BRAD API reference for endpoints, request parameters, and complete response schemas.

How it works

Sentiment is evaluated in the context of each named entity rather than across the event as a whole. When an entity appears more than once, BRAD aggregates the scores from all mentions into one sentiment profile.

Each profile includes likelihood scores for negative, neutral, and positive sentiment. Every score group contains these aggregates:

Metric	Range	Description
min	0–1	Lowest likelihood across the entity's mentions.
max	0–1	Highest likelihood across the entity's mentions.
avg	0–1	Average likelihood across the entity's mentions.

A value closer to 1 indicates stronger confidence in that sentiment. These values describe sentiment likelihood; they are not the same as the event-level sentiment score.

Response structure

entitySentiments is an array containing one object per evaluated entity.

Field	Type	Description
name	string	Entity name as identified in the event.
category	string	Entity category.
labels	string[]	Assigned labels: NEGATIVE, NEUTRAL, or POSITIVE.
negativeScores	object	Minimum, maximum, and average negative likelihood.
neutralScores	object	Minimum, maximum, and average neutral likelihood.
positiveScores	object	Minimum, maximum, and average positive likelihood.

Example

{
  "entitySentiments": [
    {
      "name": "Goldman Sachs",
      "category": "banking",
      "labels": ["NEUTRAL"],
      "negativeScores": {
        "min": 0.0008216507,
        "max": 0.0037292289,
        "avg": 0.00227544
      },
      "neutralScores": {
        "min": 0.99622947,
        "max": 0.99908435,
        "avg": 0.99765694
      },
      "positiveScores": {
        "min": 0.000041385654,
        "max": 0.00009400846,
        "avg": 0.00006769705
      }
    }
  ]
}

Interpretation

In this example, NEUTRAL is the assigned label and the neutral likelihood is consistently high across all mentions. The narrow gap between min and max also indicates that the sentiment is consistent across those mentions.