--- layout: eye ---
Status: 28.08.2019
Data: Maël Kubli · Graphics: Lucien Baumgartner · Text: Clau Dermont / Lucien Baumgartner

For one year, the DigDemLab has mapped the Swiss media landscape in cooperation with the Swiss Media Database (SMD) and based on the work of Année Politique Suisse (APS). At the centre of our efforts is the possibility of classifying political news continuously and automatically in order to find out what Switzerland is currently occupied with. In our newsroom we present a weekly topic barometer that reveals the hot topics of the 2019 parliamentary election campaign. The different topic categories follow the Swiss election study (Selects) in order to be comparable with their surveys.

The Newsroom houses the current topic ranking according to the classification scheme of the APS, based on all German and (soon) French newspaper articles of the major national and regional newspapers. Of course, great importance is attached to the monitoring of changes in the thematic landscape over time. The results of our monitoring are commented here only very simply, in order to offer as objective a data platform as possible. For our more detailed comments and analyses of the weekly shifts in the ranking, please visit our Eye, where we regularly review and discuss the changes. All elements of our monitoring are updated every Tuesday.

Ranking

Our ranking illustrates which of the 15 topics are currently making headlines. It is based on the proportion of newspaper articles published last week on the topic in question. Of course, most newspaper articles contain more than just one topic, but one is usually more dominant than the others. Our classification of topics is probabilistic, i.e. each newspaper article is assigned a vector with the proababilities of belonging to the different topics. In order to enable a a more clear-cut analysis, for each text we only took the topic with the highest probability into account. The so-called topic prevalence, according to this scheme, is nothing but the percentage of texts with topic x as the most probable classification across the corpus of all texts of this week.

The following information can be found from left to right: Rank, difference in rank from the previous week, name of the topic, prevalence of topics in percent, and development of topic prevalence over time (inline graphics).

One of the top topics most of the time, Public Services & Infrastructure, has taken the lead again, overtaking two other candidates for 1st place: Environment & Energy and Education & Culture. With a prevalence of over 10%, all three are much more present in the public discourse than the other topics. But there is also a surprise in the midfield: International Relations & Army and Immigration & Asylum each lose 4 places.

Ranking over time

A ranking of discussion topics only makes sense if the temporal dimension is also taken into account. Since a complete presentation would quickly become confusing, we present a weekly selection of the three most interesting developments. You will find a more detailed overview for all topics here.

Although Public Services & Infrastructure currently leads the ranking, this topic will not be considered until further notice. The topic is so over-represented, because newspapers regularly publish or comment communal decisions for the construction or maintenance of regional infrastructure. This habitus shows hardly any temporal fluctuations and hardly reacts to events, unlike the topics discussed below.

The climate issue clearly dominates here, in accordance with the social attention for the climate crisis and the "Fridays for future". Meanwhile, the EU issue has lost traction since the summer. At the latest after the vote on arms law and thus the question of the continuation of the Schengen Treaty, the issue is less dominant. Another topic that determined the election campaign in 2015, may not attract attention this year either: Immigration & Asylum. The SVP, especially, might have problems in the election campaign because its two main topics do not draw attention anymore as they used to do. The parties that are interested in environmental issues (Greens, Green liberals) or have been dealing with the issues for some time (Social Democrats, BDP, Christian Democrats) are therefore likely to benefit most from the current situation. Last but not least, this also explains to a certain extent the green change of heart for the Liberals.

Timeline

The ranking itself is a discrete classification of the topic prevalence and thus an abstraction. For an insight into the actual proportions of topics in the overall discourse of the media you will find an overview below.

Topics

But what is actually behind these topics? In our models, topics are nothing more than so-called bag of words, or "word collections", i.e. sets of words or terms that occur frequently across all texts. We have created word clouds with the most dominant terms for each topic to make these abstract entities a little more tangible.

The word clouds only serve to illustrate the topics and have little statistical significance themselves. However, they are often used by researchers to validate the machine classification of texts. This happens in the training process for the algorithm, which currently classifies the latest texts according to this scheme on a weekly basis and is thus representative of the machine-algorithmic "understanding" of the corresponding topics.

  • EU / Europe
    Since the EEA referendum in 1992, Switzerland's relations with the European Union have been a constant political issue, with the Bilateral Agreements and in particular the free movement of persons regularly providing the subject for discussion. The current debate is dominated by the framework agreement with the EU.
  • Environment & Energy
    Environment & Energy is probably the trend topic of this year, especially when it comes to renewable energies. The debate about renewable energies has not only been viral since the climate strike, but has a longer history, shaped by the votes on nuclear energy and the energy strategy.
  • Education & Culture
    In addition to Economy and Environment & Energy, Education & Culture is one of the top topics. The majority of the topic is determined by discussions on educational policy. Surprisingly, however, the discussion on the position of the church in society also defines this topic. In cultural terms, art exhibitions in particular seem to be a lively topic of discussion, while concerts and theatre performances seem to be less so.
  • Economy
    Economy is one of the dominant topics in the 2019 election year. Unlike periodically growing topics such as agriculture or the EU/Europe, the economy remains unbrokenly at the forefront all the time. However, the business vocabulary is very widespread and the top terms shown on the right occur only slightly more frequently than the standard vocabulary of economic analyses in the media.
  • Internationale Relations & Army
    This topic reads like an annual review of international politics for 2018.
  • Immigration & Asylum
    The vocabulary in the asylum debate has hardly changed since the refugee crisis of 2014, as the adjacent word cloud shows. Buzzwords and legalistic vocabulary are reaching out and the focus is increasingly on the latest asylum statistics and not on integration issues. Only the region of origin has shifted (back) from Syria to the Mediterranean. The topic is also far less present than during the wave of refugees or the last elections in 2015.
  • Agriculture
    The agricultural debate in the media is dominated by the discussion on the use of genetic engineering. This is joined periodically by discussions on frost insurance in the winter months or the hunting season in autumn. The topic is generally not very prevalent over time, but can dominate unusually for a limited time due to a vote or a mediatized event.
  • Social Security & Welfare State
    In the area of social insurance, old-age provision and the failed reform of it dominate in particular. In addition, there is the debate on social assistance in various cantons and cities.
  • Gender Issues & Discrimination
    Compared to the other topics, this topic is hardly present in the media any more. It was only in June 2019 that the topic was discussed for a few weeks. In general, the topic is defined by gender equality issues, especially with regard to equal opportunities and employment for women and men in the labour market. But also the sexism debate from 2017/2018, inspired by the #metoo movement, is still reflected in the media discourse.
  • Labour Market
    Debates about the labour market are practically absent from the media in 2019. As expected, "unemployment" is the buzzword. The remaining top terms for the topic mainly describe the actors in the labour market. Except in times of economic crises, however, the topic rarely makes the headlines. Topics like this, however, are essential as negative controls for the validation of the classification. If, contrary to expectations, the topic were viral, the validity of the results would be doubtful.
  • Law & Order
    This topic contains mainly the most frequently appearing actors. What is interesting, however, is how strongly the police use of bodycams was discussed. The topic is very susceptible to the mediatization of important processes and judgements, which is reflected in the high fluctuation of topic prevalence.
  • Finance & Taxes
    Fiscal policy determines how much money is available for political solutions and is therefore a regularly discussed topic. Since the crisis, there has also been a discussion of the tax system for companies and wealthier individuals. In addition, it is evident that this topic is often discussed in the media in a community context.
  • Public Health
    The health care system is a relatively heterogeneous topic with strong fluctuations. The media mainly focus on costs, especially in connection with health insurance premiums. But medical topics such as new treatment methods or studies on new drugs or procedures are also increasingly making the headlines.
  • Public Services & Infrastructure
    Welcome to the unbeaten top topic for 2019. At first glance, this result is astonishing. But above all regional media publish community resolutions for the construction or maintenance of regional infrastructure at frequent intervals, which catapults the topic forward. In addition, there are discussions about the expansion of the bicycle road network in Swiss cities, as well as reports on noise complaints.
  • Regions & National Cohesion
    The relationship between regions and languages is important in the issue of national cohesion. "Swiss" idiosyncrasies such as the militia system, history or the myth surrounding the founding of the Swiss Confederation are also regularly featured in the media - but especially around the 1st of August.

On the Method

The evaluations shown here are based on machine learning models for the classification of texts. As a basis we take the newspaper articles of all (German-speaking) daily and weekly newspapers in Switzerland. In a two-stage procedure, the texts are first classified according to whether they express an opinion on Swiss politics (e.g. in contrast to European politics, which has nothing to do with Switzerland, e.g. many Brexit articles). The texts are then assigned to 16 categories. Fifteen of these categories have content, one category represents the overarching theme of the political system and its processes. The texts can also belong to several categories, e.g. to a topic and the political system in the case of a vote.

The newspaper articles come from the Swiss Media Database (SMD). The machine classification is based on newspaper articles which have been categorized by people in political topics. This categorisation as a basis for the models comes from the Année Politique Suisse (APS). The models combine different approaches for machine learning, using neural networks, gradient boosting and random forest - but what looks straightforward and produces simple indicators in the end, is based on complex, computation-intensive models, which would not have been feasible with normal computing capacities a few years ago.

The current evaluation is based on training data from 2012 to 2018 and is applied to texts from the year 2019. An extension to French-language newspaper articles is planned, but this is more difficult due to the smaller data base.