On Radarly, thematic content analysis uncovers what the most conversational topics are within a defined time frame.
Our technology has classified these themes according to three levels:
Categories and Sub-categories: are based on the ontology of the news which means that they refer to categories that you will find in the media
Concept : referring to our internal knowledge base to our tool
Note: The publications are classified in a Category and in a Sub-category only when they are relevant and the different Concepts will be subsequently extracted from the same articles, posts.
This article will cover:
Topic Wheel
The Topic Wheel is a graphic in the form of a circle showing you the three levels of themes (namely Categories, Sub-categories and Concepts). By default, the Topic Wheel will select and display the most relevant topics of the articles.
The Topic Wheel is based on three levels :
The inner level of the wheel contains the Categories. (Note that the size of each level depends on occurrence's numbers of the given Category).
The middle level of the wheel contains the subcategories. The size of each level depends on occurrence's numbers of the given Sub-category.
The outer level of the wheel contains the most significant concepts from documents classified under a theme. These concepts are sorted in alphabetical order and the height of the bar is proportional to the number of occurrences of the concept.
Note: You can obtain more information on the themes by hovering over the Topic Wheel with your mouse.
Topic Stream Graph
The Topic Stream Graph allows you to visualize the evolution of the volume of subjects over a defined period. Only Categories and Sub-categories are displayed on this graph in order to obtain a synthetic view of the distribution of subjects over time.
Accessing These Features on Radarly
From the Main View on Posts & Analytics, click on the tab at the top centre of your screen to switch views and access the Topics custom view.
Going Further on the Extraction of Topics on Radarly
Method 1: Recognition of named entities:
Named entities recognition is the process of detecting semantically interesting words or groups of words (such as businesses, places, people, etc.) that takes place through machine learning.
For example, the following words in bold will be considered relevant and therefore extracted : Brigitte stayed with her partner Emmanuel Macron during French presidential election.
Note: Since this is a learning process, there will always be a small margin of error.
Method 2 : Named entities links:
It is the process of mapping a named entity to a well-defined entity in a knowledge base. At Linkfluence, we use Wikidata as a knowledge base.
If we take the same example as before, we can understand that Brigitte name is linked to Brigitte Macron, the wife of french president Emmanuel Macron and that the presidential election is a french one and not american.
Note: Wikidata is an open source project that integrates Wikipedia in a way that makes it easier to analyze and calculate relevant information.
Method 3 : Classification thematic:
Once the topics have been extracted and linked together, our algorithm will classify the topics. We perform this classification with a hierarchical classifier (Machine Learning again!).
Note: There is a unique classifier for each language since the documents are no longer represented by words but by sequences of identifiers.
💡 Tip
Need more help? Feel free to reach out to us via Live Chat or check out our Customer Community.
Find answers and get help from Meltwater Support and Community Experts.