NGI FORWARD

Filtering process

• Most relevant, trending words and bigrams related to social issues are selected for analysis
• The co-occurrence is calculated for word pairs consisting of these terms and the most trending 15000 terms
• For each social issue, a separate filtering has been carried out
• The terms are sorted by the index described in the methodology section, and most relevant are picked from the top 100 words by normalized (by root of frequency) co-occurrence

Issue mapping

Articles have been assigned to the European group if they have been written for a European source (The Guardian, The Register, The Conversation, Techforge, Euractiv or Politico.eu) or contain a noun or adjective representing European Union member states or their capitals. Otherwise, they are marked as American. It is a simplification, “Rest of the World” may have been more precise in some cases, but a vast majority of issues is related to American companies, products and politics.

94 social words have been chosen and their mean inverse frequency by month checked. Words which are social, but common have a lower inverse frequency (like ethnic: 705) , and less frequent words’ inverse frequency is higher (e.g. green_new: 3751). If the sum of inverse frequencies of social words occurring in an article exceeds 2500, it is a social article; otherwise it is categorized as technological. It does not matter how many times a word occurs in an article, so longer articles do not have a significantly increased probability of being classified as social.

After the assignments, top trending terms have been ranked according to their frequency in European and social articles. Because of bias in the choice of sources – American sources included in the analysis are specifically technological, while some European ones (like the Guardian) are rather mixed or even clearly social (Politico.eu), the result was an apparent linear dependence – the more frequently a word occurs in social articles, the more European it is. To find the terms which are more European than expected, we performed OLS regression on the ranks, to modify the EU/US axis to residuals from the regression. Even if a word is more often used in the American sources and is technological it may be shown on the European half, because it is more European than one could expect given the choice of sources.

Sentiment analysis

The sentiment analysis has been prepared using VADER*, an open-source lexicon and rule-based sentiment analysis tool. VADER is specifically designed for social media analysis, but can be also applied for other text sources. The sentiment lexicon was compiled using various sources (other sentiment data sets, Twitter etc.) and was validated by human input. The advantage of VADER is that the rule-based engine includes word-order sensitive relations and degree modifiers.

As VADER is more robust in the case of shorter social media texts, the analysed articles have been divided into paragraphs. The analysis has been carried out for the social issues presented in the co-occurrence exercise.

The process included the following main steps:

• The 100 most frequently co-occurring terms are identified for every term (using the co-occurrence methodology)
• The articles containing the given social issue and co-occurring term are identified
• The identified articles are divided into paragraphs
• Social issue and co-occurring words are removed from the paragraph
• The VADER sentiment analysis is carried out for every identified and modified paragraph
• The average for the given word pair is calculated for the final result

The procedure has been repeated for all identified issues.

The sentiment analysis resulted in a compound score for every paragraph. The score is calculated from the sum of the valence scores of each word in the paragraph, and normalised between the values -1 (most extreme negative) and +1 (most extreme positive). Finally, the average is calculated from the paragraph results. Removal of terms is meant to exclude sentiment of the co-occurring word itself, because the word may be misleading, e.g. when some technologies or companies attempt to solve a negative issue. The neighbourhood's scores would be positive, but the negative term would bring the paragraph's score down.

The presented tables include the most extreme co-occurring terms for the analysed social issue. The examples are chosen from the list of words with 30 most positive and 30 most negative sentiment. The presented graphs show the evolution of sentiments for social issues. The analysed paragraphs are selected the following way:

• The articles containing the given social issue are identified
• The paragraphs containing the social issue are selected for sentiment analysis

Charts

This section compares the trends of terms across different sources and provides interesting insights. Tech journalists and scientists have different focus in their analyses. More general words have a robust trend in popular news sources, with much less significant increase in academic sources. On the other hand, scientists do research on smaller sub-fields in their fields.

The charts also show the difference in social science (SSRN) and computer science (arXiv) research fields.

Trend robustness

• 100 most trending terms for the full analysis period
• Additional analysis of last 3, 6 and 12 months
• The table shows that trends are not linear
• Verification of methodology: news articles are covering future events and predictions