Sentiment analysis is a machine learning technique to identify and extract subjective information from unstructured and mostly text-based information. HolonIQ's proprietary sentiment analysis engine scores textual data on its sentiment by determining a score between -1 and +1, representing the spectrum of sentiment from highly negative to highly positive. This engine is being trained overtime on specific contexts, such as education, finance, geopolitical, or consumer.

Common Use Cases

  • Track positive and negative developments on competitors, partners, key topics

  • Monitor brand, reputation and product sentiment.

  • Public opinion and perception towards topics and brands.

  • Customer feedback and understanding customer needs.

  • Due Diligence on organizations and people of interest

  • Crisis and incident detection

  • Quantitative models using a time series of sentiment scores and polarity or threshold flags

How it works

HolonIQ is building hundreds of thousands of expert sentiment-scored text in different forms and in different contexts. These examples are used to train our sentiment engine and evaluate the accuracy of the engine as it further develops.

Context is king. For example, the sentiment score of a press release headline about an education-related event will use different language to convey a different sentiment than say a journalist's headline in the financial press about the same issue.

Alternatively, a long-form opinion piece on the future of sustainability will also use language in different ways than a reporter might convey the key developments of an important sequence of real-world events.

In this regard, sentiment analysis is challenging and our approach is to build a context-sensitive sentiment engine for superior accuracy as we develop the engine over time.

Scores and Polarity

The sentiment engine will produce a score and polarity for each piece of text it evaluates. For signals, this includes both the headline and the body of the signal which is combined with the title weighted higher as the body can often move through a wide range of sentiment as part of an overall argument or position.

Scores capture a numeric representation of the sentiment with -1 representing highly negative and +1 representing high positive.

  • Scores above +0.1 are labeled as 'positive'

  • Scores below -0.1 are labeled as 'negative'

  • Scores between -0.1 and +0.1 are labeled as 'neutral'.

All of these variables are available in the Studio for further analysis.


Did this answer your question?