Topic classification is the process of automatically identifying the main themes or subject matter of a piece of text. It's another way on Pulsar to help you extract important information from large volumes of unstructured text, which can range from Tweets, Facebook and Instagram content, to Online News articles, Print articles, TV and Radio transcripts and even Podcasts.

Our classification uses a top-down taxonomy led approach, which is able to identify the main themes in a document, even if they aren’t explicitly mentioned, unlike with Keyword or Entities analysis, where that keyword or that person has to be explicitly referenced in the post.

An example of Topic classification applied to a post is the following: "Business Insider - XRP surges 10% on speculation the CFTC's Binance suit could impact the token developer's monumental SEC case". Topic classification on Pulsar would analyse the content of this post and assign it to some relevant categories based on its content. In this case, the most appropriate topics for this article might be: cryptocurrency, blockchain, financial regulation.

We surface topic classification on TRAC in the Content Insights section and provide you different ways to understand Topics.

The Treemap visualisation provides insights into the channels where certain topics are discussed most frequently. The size of the tile represents the prevalence of the topic being discussed on a particular channel. It also helps to identify if certain topics are over-indexed or under-indexed on some data sources. For instance, the screenshot below shows that drinks is a highly discussed topic across all channels. However, the Treemap highlights that the topic over-indexes by 2.02% on X compared to other data sources, which is an interesting insight.

When looking at topics displayed in a Sentiment Word Cloud, you can understand the most common topics in a search and the sentiment associated with those topics. The bigger the size of the topic, plus the more central it is in the graph, then the greater the number of posts and articles discussing that topic.

When looking at topics displayed in an Emotions Word Cloud, you can understand the most common topics in a search and the emotion associated with those topics. The bigger the size of the topic, plus the more central it is in the graph, then the greater the number of posts and articles discussing that topic.

When looking at topics displayed in a Segments or Network chart, you can start to identify groups of topics that tend to be associated or discussed together in the same conversation. This can be useful to help understand how specific topics are closely related. Related topics are grouped together into distinct segments, and the clustering algorithm we apply determines the relevance and importance of each topic within a given segment, helping you uncover the dominant themes in a dataset.

Analysing topics displayed in a Stream graph enables you to track how the discussion around particular themes evolves over time. This allows you to identify when a specific topic emerged, as well as when the conversation subsided and possibly resumed. By correlating this information with relevant media events, you can gain valuable insights into why certain topics were included in the discourse.

Sometimes known as a chord diagram, the topics Bundle chart is a graphical representation of the relationships between the different themes in a dataset. It's a great way to visualise the inter-relationships and flows between the topics as arcs or chords, that connect the topics. Each topic is represented as a segment around the perimeter of the circle, with the chords connecting the topics representing the degree of overlap, or connection between them. The bigger the segment around the perimeter of the circle is, then the more connections that topic has with other themes in the bundle chart.