You have been classified countless times at school! The most popular application of text classification in machine learning is sentiment analysis, where texts are given an emotional label such as ‘positive’ or ‘negative’. However, there are many other text classification applications that can be realized today with machine learning. In the following, these five applications of text classification will be discussed:

Sentiment Analysis

Sentiment analysis is the most popular application of text classification in machine learning. It ‘simply’ assigns texts an emotional label such as ‘positive’/’negative’ or ‘subjective’/’objective’. There are many texts that can be analyzed:

  • Comments on websites
  • Entries in user forums
  • Posts in social media platforms like Twitter or Facebook
  • Reviews of movies, music, books
  • Product reviews in online stores

Today, sentiment analysis programs classify up to 95% of texts correctly. And sentiment analysis has a big impact today. Twitter tweets for example can greatly influence the price of a stock today. A recent Quora Post states: “Similarly, Elon Musk tweets about a new product line, and Tesla’s shares jump 4%, adding $1.3 billion to Tesla’s market cap.”

Therefore sentiment analysis of tweets is increasingly being used as one of the criteria of stock analysis and company ratings.

Spam Filtering

In spam filtering, texts are classified as ‘spam’ or ‘not spam’. The main application area today is the classification of email. However, spam filtering can also be used to automatically detect spam in

  • Comments on websites
  • Posts on social media
  • Product reviews

The fields of application are similar, as in sentiment analysis.

Fake News Detection

Fake news detection is one of the relatively new but promising applications of text classification in machine learning. In fake news detection, texts are classified according to their truth value (‘true’/’false’ or ‘fake’). So, for example,

  • Fake News
  • Misinformation
  • Fraudulent tweets

can be recognized with an application.

Today it is (unfortunately) still difficult to check texts completely automatically for their truth value. That’s because there are far fewer fake news than true ones. However, systems that support people in the assessment of news messages are quite successful.

The Fake News Challenge deals for example with this. In its first round, it uses the so called stance detection approach. Here the title of a text is compared with the content (‘agrees’, ‘disagrees’, ‘discusses’, ‘unrelated’). On the basis of this comparison statements can be made about the truth of a news message.

Language Identification

Language Identification (also known as “Language Guessing”) analyzes the language in which a source text was written. Language classification is i.a. important in the

  • Preparation of text classifications (only if the source language is known, the text can be classified)
  • Preparation of text translations (only if the source language is known, the text can be translated)
  • Determining an unknown language or an unknown dialect (which language is this? Finnish, Norwegian, Italian, Portuguese?)

In a test, Google Translate passed 18 out of 20 tests. Error made the tool only with Belorussian and Tatar.

Genre Classification

The term ‘genre’ is often used in connection with music, literature or drama (blues, rock, jazz, fiction, mystery, drama). But here it has to be taken further. For example, texts can also be classified according to the following criteria:

  • News section (e.g. politics, culture, sport, science)
  • Topic (e.g. medicine, finance, science, restaurant review, product review)
  • Target audience (e.g. demographics, geographics, needs, lifestyle)
  • Style (e.g. serious, boulevard, academic, satirical)

The possibilities to classify texts are numerous and depend on the goals and requirements of the respective application. I hope we have given you a useful overview of the possible applications of text classification, and thus made it easier for you to get started with this powerful approach.

Signup for Our Newsletter

About the author: Thomas studied computer linguistics and philosophy and graduated with a PhD in computer science. He has worked as a consultant for natural language processing and application development for major Swiss banks. Thomas is founder of ipublia. He lives with his family in Zürich.

Leave a Reply