This research is based on multiple implicit underlying assumptions, for example that all words in an article are equally important and that the assigned weights correctly represent the importance of a source. We recalculated all results changing one of the assumptions, but the final results do not change significantly regardless of the method. Out of top 100 most growing terms in the base methodology, 95 and 86 occur in top 250 most growing terms after first and second change respectively. For top 100 most growing terms, the values are 94 and 79. More weight has been given to title and the first paragraph in coefficient calculations (title weight: 5, first paragraph: weight 3). Terms which are more salient and important are assumed to be at the beginning of article. It may be misleading if title and the first paragraph are “clickbait” – there is risk of capturing divisive and eye-catching words, which are not emerging technologies. The changes may be desirable – e.g. "alleged" disappears from the growing list, mostly because vague terms are not used to catch attention, however, words like “cybersecurity” and “fake” disappear too, even though they may warrant further investigation. Terms which appear in the method with increased weight of title and the first paragraph seem random or irrelevant, which is why we consider this method inferior.

The weights of sources have been removed, so that all articles are equally important. In spite of Fastcompany and ZDNet being similarly popular according to Alexa.com rank, ZDNet has much more influence on final results, as ZDNet has over four times as many articles as Fastcompany. There have been more changes in trending keywords than in the first robustness check (for clarity we present here only terms which are outside top 250 in the alternative method), but the additional keywords do not seem relevant, while we have lost a great deal of terms related to politics. As the goal of this research is to discover trends in a wide variety of sources and find emerging technologies' importance for social issues, we have chosen the methodology of weighting by source.

As a further robustness check, we computed the growth rates between years. Unfortunately, this method is not well suited to technological terms and social movements, as they are likely not to occur at all in one year and then explode in popularity in another. Regardless of precise values, they both will have the same growth rate (infinity). Moreover, the method does not seem to be robust to changes in starting month. Table below shows the terms with the highest annual average growth rate. Annual growth rates could be useful to show the growth of multiple trending (by linear coefficient) terms, and identify the period when the term was most trending. Is it a robust technology with slowing growth rate (AI, blockchain), or an issue which grows in importance year after year (climate)? We show the results in the table below, but the charts in the main part of the article are sufficient for our purposes.

Exponential regressions may better suit the dynamics of term frequencies. However, the results show that the baseline method is better in identifying relevant trending terms. As logarithm is the inverse of the exponential function, running an OLS regression on logarithms is equivalent to exponential regression. Zero values pose a problem, as logarithm of 0 does not exist – they are discarded. In the regression, the explanatory variable is the number of the month (between 1 and 48, inclusive), and the response variable is the logarithm of the term frequency for the corresponding month. We get two parameter values from the regression – constant and β. The β parameter is highest for terms with very high values in the last analysed month, low in the second-to-last month, and zero in all previous months. Consequently, normalization is required – we multiply the β parameter by mean frequency to the power of a scaling parameter:
$$exp\mathrm{\_}coef\mathrm{\_}norm=\beta \ast mean\mathrm{\_}frequenc{y}^{scaling\mathrm{\_}parameter}$$
If the scaling parameter is 1, top results are similar to our results from OLS, only the order changes: 2018, 5g, huawei, facebook, and 2019 are five most trending terms (5th, 3rd, 1st, 2nd and 4th highest OLS coefficients respectively). The lower the scaling parameter, the more specific the results get. With the value of 2/3, pixel 4 enters the top 5, and is in the top 3 with scaling parameter 1/2. Some most trending terms from linear regression are still maintained in exponential regression with scaling parameter 1/2.