“The only constant in life is change.”
– Heraclitus, Greek philosopher
The modern world is changing at unprecedented speed and intensity, creating a great deal of uncertainty for some – and opportunities for others. As a business, how do you become part the “opportunity” camp? The first step is to get a deep understanding of the changes that are happening. Based on this, you can match external changes to your internal strategy and competences and come up with a successful strategy for integration.
Trends are an important component of change. In a broad sense, trends exist at multiple levels. They can be new topics dominating the discussion, changing attitudes or narratives as well as the ubiquitous megatrends – global tectonic shifts that affect all levels of humankind. While trusted networks such as the own workforce or industry experts can provide valuable insights on trends, public media plays a growing role in shaping and promoting trends.
In this article, we will see how trending topics in media can be extracted and analyzed using Natural Language Processing in three steps:
- Step 1: We look at prominence timeseries patterns to spot trending topics.
- Step 2: We analyze the meaning of a topic by constructing and inspecting its semantic context.
- Step 3: We represent changes over time in the semantic context of a topic.
This article is part of a series of articles on trend analysis from text data and the operationalization of these trends for business purposes.
Detecting trending topics based on prominence
Most topics can be represented by nouns or noun phrases. They can be old existing concepts that gain a new meaning in a specific context. Think of the topic “vaccination” – while being a relatively unspectacular topic before the pandemic, it evolved into a central social and medical issue spanning the levels of politics, economy, ethics, human rights etc. Topics can also be new terms or “buzzwords”, such as LGBT and Cancel Culture. Finally, in technology, many topics are waiting in the shadows for years or decades until the technological state-of-the-art matures for their implementation.
As an example, let’s consider the topic Metaverse. It has been around since 1992, when it was coined in the science fiction novel Snow Crash (Neal Stephenson) as a portmanteau of "meta" and "universe. Being reserved to science fiction and gaming for almost three decades, it led a marginal existence until 2021. The following chart shows the evolution of the share-of-voice of the term “metaverse” since the beginning of 2020:
Figure 1: Share-of-voice for “metaverse”
This time series shows the hockeystick-like development of the topic over the past year, taking off in July 2021 when Mark Zuckerberg announced that Facebook is dedicating a new team to the metaverse. One doesn’t need to be an expert to spot that something important is going on there – and companies in related industries should pay attention.
Capturing the semantic context of topics
So, the metaverse has tremendously gained in popularity since July 2021 – but what does it actually stand for? According to Satyan Nadella, CEO of Microsoft, the metaverse is really “just games”. This is a helpful – though simplifying – classification, but there is obviously more to it. As any other topic, the metaverse is embedded into a larger semantic context that should also be considered. Let’s use word embeddings to construct this semantic context. The following graph shows which terms are closely associated with “metaverse” in the media:
Figure 2: Semantic context of “metaverse”
We can spot different types of associated topics to further inform strategic analysis and planning:
- Companies such as Facebook, HTC and OpenAI are relevant for competitive intelligence.
- Technologies such as Blockchain and NFT immediately point us to the adjacent technologies to explore, be it internally or externally.
- General topics such as Avatar, Immersion and Futurism provide further clues which can also be used for efficient communication.
How semantic contexts evolve over time
Finally, the meaning of a topic changes over time, thus modifying the related business priorities. To understand this dynamics, we take the top closest terms as shown in Figure 2 and show how strongly they are associated with the metaverse in different time periods. For this, we train a word embedding model for each month and consider the similarity scores between “metaverse” and each of the associated topics:
Figure 3: Evolution of the semantic context of “metaverse” over time
Looking at this map from some distance, we intuitively spot that the left half is dark, while the right half is rather light. Once more, this reflects the growing popularity of metaverse since Facebook’s commitment in July 2021 and shows how it is getting embedded more tightly into a larger context of the related technologies and concepts. Further, we can see that the four dominant companies – Facebook, OpenAI, HTC and Microsoft – shape the metaverse discourse in specific periods. However, they don’t end up being highly relevant in the recent months. This shows that the metaverse market might well be open for a diverse set of players.
In this article, we have considered NLP techniques for trend analysis from public media data. Quantitative metrics, such as frequency and share-of-voice, can be used to understand the prominence of a topic. Word embeddings can be used to construct the wider semantic context of a topic. Finally, word embeddings can further be decomposed in time to see how the semantic field of the topic is changing over time.
While this article has focused on the example of the metaverse, the presented methodology is highly scalable and can be reused for any other topic. The methodology can be further refined by analyzing the influence scope of a topic – especially the industries, domains and geographies it is spanning. Finally, the presented method looks at trends in retrospective – in a future article, we will see how major trends can be predicted based on timeseries patterns.
This article was originally published on The Yuan.
References and further readings
 Data basis: over 500k articles from 451 major English-language online media sources for the timespan January 2020 – February 2022.
 Mark Hachmann (2022)., Microsoft CEO: The metaverse is just games, really, accessed 15 March 2022, <https://www.pcworld.com/article/611678/microsoft-ceo-the-metaverse-is-just-games.html>.
 Janna Lipenkova (2021). Word embeddings: your secret weapon for instant intelligence, accessed 15 March 2022 <https://jannalipenkova.com/tpost/7ln0d1jhin-word-embeddings-your-secret-weapon-for-i>.
 Henrik Veilgaard (2012). Anatomy of a Trend, paperback.
 Matthew Mockridge (2011). Your Next Big Thing, paperback.
 Binling Nie and Shouqian Sun (2017). Using Text Mining Techniques to Identify Research Trends: A Case Study of Design, Applied Sciences 2017, 7(4).
 Hyosun An and Minjung Park (2017). Approaching fashion design trend applications using text mining and semantic network analysis, Fashion and Textiles 2017, volume 7.