Sentiment analysis

InterTextueel provides a world-class Dutch and English sentiment analysis service, ready to use by your organization for big data analytics, or as a research tool for social scientists.

  • Detects expressions of primary emotions: anger, fear, happiness, surprise, disgust and sadness
  • Automatically detects language
  • Combination of Transformer-based (RoBERTa) classifiers with rule-based systems
  • Will recognize many different expressions, proverbs and slang
  • Grammar based analysis provides unrivaled precision
  • Easily accessible analysis from within a website (with just 3 lines of PHP or Python!)

All our Natural Language Processing services are accessible through API, allowing you to seamlessly integrate sentiment analysis into your existing website, dashboard or project. Our service can be used to process online and offline data.

Same dataset, more knowledge

All human texts (for example inside documents, books or websites) provide a rich source of subjectivity. Our analysis service helps you grasp the emotional expressions within a text. This will assist you in finding answers to questions such as:

  • What is the general mood of a text and how does it compare with others?
  • Which emotions are strongly connected to a particular topic, product or event being discussed in your dataset.
  • What emotion is dominant in a particular debate or topic on social media, and what changes do occur over time?

Let’s look at an example of the relative distribution of emotional expressions inside a corpus.

The anti-islamic Pegida movement held several demonstrations in Dutch cities in early 2016. These demonstrations received wide media attention and sparked debate. On the 29th of February, we autosearched the Dutch version of Google News for the keyword ‘Pegida’ and downloaded 308 articles, including news articles with user comments. We parsed all sentences and generated the following chart.

About 17.5% of all sentences contained emotional expressions, most of them negative. By proceeding to study the results per news source or platform, we are able to map how the demonstrations are perceived and discussed in different segments of society. Primary emotion recognition can be a powerful heuristic tool to start exploring your data and figure out which questions to ask.

Curious to query Google or Google News with keywords of your own design and see the analytical results? Use our contact form and we can run your query for you. The Google search engine also supports running queries concerning past events.

Social media ready

Remarkably, existing sentiment analysis tools function best with classical (web)texts, such as comments or reviews on a weblog, but do not correctly handle slangy, abbreviated internet-speak, or emoticons at all. The InterTextueel sentiment API maps hundreds of distinct emoji’s to one ore more corresponding emotions. It is also clever enough to discern the different intentions that may lie behind some emoticon-usage and handle them correctly.

Tweet with text 'Het is alweer droog buiten.' and sunny emoticon.

The emoji inside the tweet above is not matched to happiness, because it is used in the context of describing the weather.

Tweet with text: 'Mijn leraar is ziek. Ik heb vrij van school.' and sunny emoticon.

The emoji inside the tweet above is recognized as expressing happiness.

While we are showing examples, we may point out the previous tweet did not express disgust (according to our analysis service), but the next one does.

Tweet with text: 'Die gast is gewoon ziek.'Our API is simply ready to handle any social media data. Just insert a status update or a post and it will recognize emoji’s and medium-specific entities such as hashtags and mentions. Watch our demonstration of sentiment analysis on Dutch Twitter data.

Other use-cases

Because our sentiment analysis service can be easily integrated into an existing application, it is well-suited to become a component in artificially-intelligent software, such as:

  • Fingerprinting software (e.g. to find the similarities between texts) for the purposes of plagiarism detection, or recommendation systems.
  • Literary structure or narrative analysis tools.
  • Market analysis (e.g. track product discussions, automated product review analysis and more).
  • Machine translation software aid and human – machine interaction aid.
  • Text mining and text analytics tools, by clustering data around dominant emotions.

Please contact us with any questions relating to our emotion detection tool and perhaps we can provide pointers, thoughts and solutions.