Visualizing Semantic Similarity in Political Discourse: A Comparative Analysis of Word Embeddings in Trump’s 2016 and 2024 Campaign Speeches

Language is more than just a way to exchange information; it is the primary tool through which political leaders construct reality, define who belongs, and frame the battles of a nation. When a politician speaks, they are not merely listing policies; they are weaving a tapestry of meaning where words connect to one another in specific, often invisible ways. For decades, scholars have studied these speeches by counting words or reading for tone, but a newer field of study uses computers to map the hidden architecture of these connections. This approach treats language like a vast, three-dimensional landscape where words that often appear together sit close to one another, while words that rarely meet drift far apart. By charting these distances, researchers can see not just what a leader is saying, but how the very structure of their thinking has shifted over time. This is the territory explored in a recent study that looks at the evolution of Donald Trump's political rhetoric, comparing the speeches from his 2016 campaign to those from his 2024 run to understand how the map of his political world has been redrawn.

The researchers began with a simple but ambitious question: how has the way Donald Trump connects ideas changed over an eight-year period? To answer this, they gathered two massive collections of text: 166,297 words from his 2016 campaign speeches and 218,892 words from his 2024 campaign speeches. These were not random snippets but carefully selected transcripts from rallies and official events, ensuring a fair comparison of the same type of political communication. The team then fed this text into a computer model designed to learn the meaning of words based on their company. Imagine a model that reads thousands of sentences and learns that "king" and "queen" are close because they often appear in similar contexts, while "king" and "sand" are far apart. This model, known as Word2Vec, converts every word in the speeches into a mathematical point in space. The distance between these points reveals how similar the words are in meaning within that specific era.

Once the computer had mapped the words, the researchers used a method to group them into clusters, much like sorting a mixed bag of marbles by color and size to see the dominant patterns. They then used a technique to flatten these complex, multi-layered maps into two-dimensional images that could be seen and analyzed by human eyes. What they found was a story of both continuity and profound change. The core themes of Trump's message remained remarkably stable across the two periods. In both 2016 and 2024, the speeches were built on a foundation of populism, pitting "the people" against a corrupt "elite," and promising national strength and renewal. The computer confirmed that these central pillars of his rhetoric did not disappear; they were still the heaviest, most prominent features of the landscape.

However, the way these themes were arranged around each other had undergone a dramatic transformation. In the 2016 speeches, the map was dense and tightly woven. Economic ideas like jobs, trade, and national development were clustered closely together with themes of security and identity. The language suggested a unified vision where economic prosperity and national strength were inextricably linked parts of a single, cohesive story. The computer visualizations showed these ideas huddled together, indicating that in 2016, the political narrative was a single, integrated fabric where one topic naturally led to another.

By 2024, that fabric had frayed. The same core themes of populism and strength were still present, but they were no longer tightly bound to the economic issues that once defined them. Instead, the 2024 map appeared scattered and fragmented. The computer analysis revealed that the themes had drifted apart, forming distinct, isolated islands of meaning. The discourse had become more divided, with new clusters emerging around political legitimacy, media narratives, and identity-based conflicts. The tight connection between economic policy and national identity had loosened, replaced by a structure where different topics existed in separate, often conflicting, realms. This suggests that the political message had shifted from a broad, policy-oriented vision to a more fractured narrative driven by ideological battles and the defense of political truth.

The study also looked closely at specific words to see how their meanings had shifted in the space between the two campaigns. Words like "China," "fight," "election," and "truth" showed the most dramatic movement. In 2016, "China" was a neighbor to words about trade, manufacturing, and jobs; it was framed primarily as an economic competitor. By 2024, the word had moved far away from those economic neighbors and settled next to terms related to national security, threats, and geopolitical rivalry. Similarly, the word "fight" had shifted its location. In the earlier speeches, it was associated with military strength and external threats. In the later speeches, it had migrated to the center of domestic political struggles, clustering with words about elections, media censorship, and internal conflict.

Perhaps the most telling shift occurred with the word "truth." In 2016, this word was on the periphery, not central to the main clusters of meaning. In 2024, it had become a focal point, forming its own isolated cluster surrounded by words about media narratives, censorship, and political identity. This indicates a move away from discussing objective facts or policy details toward a struggle over the very nature of reality and who gets to define it. The word "election" underwent a similar journey, moving from a context of victory and popular support in 2016 to a context of fraud, legitimacy disputes, and contested outcomes in 2024.

The researchers concluded that these changes were not just about swapping one topic for another, but about a fundamental restructuring of how political meaning is built. The 2016 discourse was characterized by a cohesive, economically oriented structure where ideas reinforced one another. The 2024 discourse was more fragmented, ideologically driven, and focused on identity and conflict. The study suggests that as political polarization deepens, the language used to describe the world becomes less about shared policy goals and more about defending separate, often incompatible, realities. The computer models did not just count words; they revealed that the very architecture of political speech has changed, moving from a unified narrative to a collection of disconnected, ideologically charged fragments. This finding offers a new way to understand how political leaders adapt their language not just to new events, but to a changing landscape of public belief and conflict.