Sentiment analysis trained on tweets

Machine learning models in Rosette® are trained on tweets and reviews to detect strong positive and negative sentiment in a document overall and toward specific entities. What if a product review mentions several products? Which products are more positively received than others? Here, Rosette applies entity extraction to identify the products and determines the sentiment for each one by relating the sentiment in the review to each entity (product).

Sentiment analysis filters torrents of data into useful segments to fully leverage the available data. The benefits are:

• Public relations teams get early alerts of developing negative issues
• Marketers track the effectiveness of a campaign or competitor activities
• Human resources departments predict employee churn from negative signals
• Political operatives keep a finger on the pulse of public opinion on candidates
• Innovators detect strong emotional content to find potentially unreliable clickbait articles and possible fake news.

Custom domains and sentiment analysis in 30+ languages

If you need sentiment analysis in more than the six prebuilt language models that ship with Rosette, the Rosette Classification Field Training Kit has you covered. From a set of hand-picked documents that best represent each sentiment, you can teach Rosette new domains (such as hospitality, beverages, and foods) or languages. The training kit is built on the linguistic foundations of Rosette Base Linguistics, so the necessary language-specific processing is already in place to easily train Rosette to detect sentiment in 30+ languages.

Product highlights

• Six supported languages
• 18 detected entity types
• Document and entity-level analysis
• A classification field training kit to add custom domain models or support for 30+ languages
• Cloud and enterprise deployments
• Fast and scalable
• Industrial-strength support
• Constantly stress-tested and improved

Supported Languages

Entity Types

/sentiment endpoint

{
  "document": {
    "label": "pos",
    "confidence": 0.7962072011038756
  },
  "entities": [
    {
      "type": "PRODUCT",
      "mention": "Ghostbusters",
      "normalized": "Ghostbusters",
      "count": 3,
      "entityId": "Q108745",
      "sentiment": {
        "label": "pos",
        "confidence": 0.8111117726721541
      }
    },
    {
      "type": "PERSON",
      "mention": "Dan Aykroyd",
      "normalized": "Dan Aykroyd",
      "count": 2,
      "entityId": "Q105221",
      "sentiment": {
        "label": "pos",
        "confidence": 0.6385089278441162
      }
    },
    {
      "type": "ORGANIZATION",
      "mention": "The Hollywood Reporter",
      "normalized": "The Hollywood Reporter",
      "count": 1,
      "entityId": "Q61503",
      "sentiment": {
        "label": "pos",
        "confidence": 0.5338094035254866
      }
    }
  ]
}