This graph allows you to compare the similarity of your query term to other terms you are interested in.
Word models
The similarity scores are based on word embeddings. The algorithm to compute these embeddings depends on the corpus, and is described in the corpus documentation (click 'models' at the top of the page).
For each corpus that includes word embeddings, we have trained a model of the complete corpus, and separate models for consecutive time frames. Thus, we can compute the overall similarity between terms, and say how this similarity develops over time.
If words appear in the same context, their vectors are more alike, which is reflected in a higher cosine similarity. Words that appear in the same context are likely to have related meanings: you can think of synonyms like big and large, but also antonyms like black and white, or simply words that are associated with the same topic, like bank and financial.
The similarity graph
You can enter one or more terms to compare your query term.
For each term, the graph will show the cosine similarity of that term compared to your query term. For example, if you searched for netherlands and enter belgium and germany as your comparison terms, you will see the similarity scores of netherlands-belgium and netherlands-germany.
The graph will show how the similarity develops over time, using the model from each timeframe (see above).
Zooming
You can zoom in on the x-axis by dragging the mouse. Double click on the graph to reset the zoom level.
