Google Cloud Natural Language

Extract meaning and structure from unstructured text with Google Cloud Natural Language: entities, sentiment, entity sentiment, syntax, and content classification.

7 actions available

New product review posted to the feedback store

Agent runs entity and entity-sentiment analysis on the review

Agent identifies which product features drew positive or negative sentiment

Agent tags the review by feature and sentiment

Product team gets a feature-level sentiment rollup

A review with strongly negative sentiment on a safety topic routes to a manager

What This Integration Enables

Cloud Natural Language reads text for structure and meaning. It finds the entities being discussed, scores overall and entity-level sentiment, returns syntax, and classifies content into categories. Entity-level sentiment is the standout: instead of scoring a whole review, it tells you which specific thing the writer liked or disliked. That precision only pays off if the flow acts on it. Knowing a review is negative about a specific feature is a routing decision waiting to happen. An orchestration layer takes the entity-sentiment signal, tags and files the review, and escalates the ones that cross a line, and FlowRunner is built for that layer.

Without FlowRunner

Feedback read manually Someone reads every review to find the themes

Sentiment too coarse A whole review scored, with no idea which feature drove it

Text unstructured Comments sit as free text nobody can slice

With FlowRunner

Themes surfaced automatically Entities and topics pulled from every review

Feature-level sentiment Sentiment tied to the specific feature it is about

Text made analyzable Reviews become structured, filterable records

Use Case Scenarios

Feature-Level Feedback Analysis

Product reviews come in continuously. The agent runs entity and entity-sentiment analysis on each one to find which features drew which sentiment, then tags the review accordingly. The product team gets a rollup showing exactly which feature is generating complaints, not just an overall score. The theme analysis that used to require a person reading the whole stream is automatic.

Content Classification and Routing

Inbound messages need to reach the right team. The agent classifies each message's content and routes it by category. The sorting step disappears, and each team sees a queue that is already relevant to them.

Brand Monitoring

Mentions of the company arrive from several sources. The agent extracts the entities and scores sentiment on each mention. Neutral and positive mentions are logged. A mention with strongly negative sentiment about a sensitive topic is flagged and routed to a person, so the ones that could become an issue get a human read quickly.

Human-in-Loop Highlight

Sentiment scoring is good at measuring tone and blind to context. When Natural Language scores text as strongly negative on a sensitive topic, or flags an entity that needs attention, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, assembles the text and its entity-level scores, and sends it to a manager via Slack. A person decides whether it is routine or needs a response. The model measures the signal; a human judges what it means.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via Slack

Human decides

Agent resumes with decision

Agent Capabilities

7 actions

Text Analysis

Analyze Entities Identifies known entities (people, organizations, locations, events, works of art, consumer goods, numbers, dates, and more) in the supplied text and returns their type, salience, mentions with character offsets, and any associated metadata (such as Wikipedia URLs and Knowledge Graph MIDs).
Analyze Sentiment Determines the overall emotional attitude of the supplied text, returning a document-level sentiment score (-1.
Analyze Entity Sentiment Combines entity extraction with sentiment analysis, returning each detected entity along with the aggregate sentiment expressed toward it across the document and the sentiment of each individual mention.
Analyze Syntax Performs syntactic analysis of the supplied text, breaking it into sentences and tokens and returning each token's part of speech, lemma (base form), and dependency-tree relationship to other tokens.
Classify Text Classifies the supplied text into one or more content categories (such as "/Computers & Electronics" or "/Finance/Investing"), each with a confidence score.
Moderate Text Scans the supplied text for potentially harmful or sensitive content and returns a list of safety moderation categories (such as Toxic, Violent, Sexual, Insult, or Profanity), each with a confidence score between 0 and 1.
Annotate Text Runs multiple analyses on the supplied text in a single request. Enable any combination of entity extraction, document sentiment, text classification, and content moderation via the feature toggles; the response contains only the sections for the enabled features.

Start building with Google Cloud Natural Language

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us