Google Perspective

Analyze text for toxic content with the Google Perspective API

Overview

Google Perspective uses machine learning to analyze text and score it across multiple toxicity attributes: Toxicity, Severe Toxicity, Insult, Profanity, Identity Attack, Threat, Sexually Explicit, and Flirtation. Each attribute returns a probability score between 0 and 1. The API supports English, Spanish, French, German, Portuguese, Italian, and Russian text. Content type is always PLAIN_TEXT and scoreType is always “probability”.

Category: Analytics
Tool Name: google_perspective
Version: 1

Appearance: Icon: lucide-Shield | Color: #4285F4

Node Type

Action — processes input items and produces output

Input / Output

Direction	Port(s)
Input	Input
Output	Output, Error

Credentials

This tool requires Google Perspective OAuth2 API credentials. See the Credentials Guide for setup instructions.

Operations

Operation	Value	Description
Analyze Comment	analyzeComment	Analyze text for toxic content and return attribute probability scores

Parameters

Analyze Comment (analyzeComment)

Parameter	Type	Required	Default	Description
Text	string	Yes	—	The text to analyze. Content type is always PLAIN_TEXT.
Attributes to Analyze	fixedCollection	Yes	{}	One or more toxicity attributes to score. At least one is required.
— Attribute Name	options	No	toxicity	The toxicity attribute to score. See https://developers.perspectiveapi.com/s/about-the-api-attributes-and-languages for details.
				Options: flirtation, identity_attack, insult, profanity, severe_toxicity, sexually_explicit, threat, toxicity
— Score Threshold	number	No	0	Score above which to return results (0–1). At zero, all scores are returned.
Options	collection	No	{}	—
— Language	options	No	—	Language of the text input. If unspecified the API auto-detects the language.
				Options: "" (Auto-detect), en (English), es (Spanish), fr (French), de (German), pt (Portuguese), it (Italian), ru (Russian)

All Operations

Parameter	Type	Required	Default	Description
Authentication	options	No	oAuth2	Authentication method to use.
				Options: oAuth2 (OAuth2 (recommended)), serviceAccount
Google Account	credential	No	—	Connect or select your Google account. (shown when Authentication is oAuth2)
Service Account Email	string	Yes	—	The email address of the Google service account. (shown when Authentication is serviceAccount)
Private Key	string	Yes	—	The private key from the service account JSON key file. (shown when Authentication is serviceAccount)

Output Data

One output item per input item. The Perspective API response is merged onto the input item JSON — every field the item already carried is preserved — and binary data is forwarded unchanged.

The scores arrive under attributeScores, keyed by the attribute name in upper case (TOXICITY, SEVERE_TOXICITY, IDENTITY_ATTACK, INSULT, PROFANITY, THREAT, SEXUALLY_EXPLICIT, FLIRTATION), even though you pick them in lower case in the Attributes to Analyze list. Only the attributes you requested are present.

Every score is a probability between 0 and 1 — the modelled likelihood that a reader would perceive the comment as carrying that attribute. It is not a measure of how much of the attribute the text contains.

Reference a score downstream by expression, for example {{ $json.attributeScores.TOXICITY }}, and branch on it with an If or Switch node.

Usage Examples

• Check a comment for toxicity before publishing
• Analyze user-generated content for profanity
• Detect identity attack language in forum posts
• Score multiple toxicity dimensions with score thresholds
• Moderate content in English, Spanish, or French

Example Configuration

Score a comment for toxicity before publishing it:

{
  "type": "google_perspective",
  "parameters": {
    "operation": "analyzeComment",
    "text": "You are such an idiot!",
    "attributes": {
      "attributeValues": [
        {
          "attributeName": "toxicity",
          "scoreThreshold": 0.5
        }
      ]
    }
  }
}

Screen user-generated content across several attributes at once:

{
  "type": "google_perspective",
  "parameters": {
    "operation": "analyzeComment",
    "text": "This content might contain offensive language and threats.",
    "attributes": {
      "attributeValues": [
        {
          "attributeName": "toxicity",
          "scoreThreshold": 0.7
        },
        {
          "attributeName": "threat",
          "scoreThreshold": 0.5
        },
        {
          "attributeName": "insult",
          "scoreThreshold": 0.6
        }
      ]
    }
  }
}

Analyse Spanish text, returning every score regardless of how low it is:

{
  "type": "google_perspective",
  "parameters": {
    "operation": "analyzeComment",
    "text": "Este comentario podría ser tóxico",
    "attributes": {
      "attributeValues": [
        {
          "attributeName": "toxicity",
          "scoreThreshold": 0
        }
      ]
    },
    "options": {
      "languages": "es"
    }
  }
}

Error Handling

Mode	Behavior
stop	Halts workflow on first error
continue	Skips failed items, passes successful ones through
errorPort	Routes failed items to Error output port

Tips

Analyze text for toxic content using the Google Perspective API across attributes like toxicity, insult, profanity, and threat. Use when you need to score user-generated content for moderation or safety filtering. Returns probability scores between 0 and 1 for each requested attribute.

Behavior notes

• Content type is always plain text. Markup is not stripped for you — strip HTML upstream if your source is rich text, or the tags themselves become part of the analysed comment.
• Score thresholds filter the response, they do not change the model. A threshold of 0 returns every requested score; a higher threshold suppresses scores below it, so a downstream expression may find the attribute missing rather than low.
• Pick attributes deliberately. Each one you request is scored, and requesting attributes you never read only makes the call slower.
• Supported languages are English, Spanish, French, German, Portuguese, Italian and Russian. Leave Language on auto-detect unless you already know the language of the text.