Network analysis is an established method in quantitative social research. Back in 1961, James C. Coleman published a hand-drawn network of girls in a local school to illustrate the importance of networks and the individuals’ position within the network with regard to social behaviour. An experiment on anti-conflict intervention in 56 New Jersey public middle schools demonstrated, that interventions through “social referents”, i.e. students with a dense social network, are more effective than interventions addressed to all students irrespective of their role within the school’s social network.

This kind of network analysis is time-consuming and requires expensive data collection. Only with the advent of social media – or better social network platforms – it became possible to collect data on social networks and on the actual content of conversations between the “actors” at a large scale.

Indeed, combining social network analysis (SNA) with automated text analysis has the potential to understand how public opinion , marketing messages, brand preference or consumer expectations and attitudes spread throughout a target audience. SNA is particularly helpful to understand the role of influencers or opinion leaders for the adoption of new products or brand preference in specific markets.

Two case studies combining SNA with more traditional metrics of brand preference show that the heterogeneity or homogeneity of an individual’s network with regard to a specific product category indeed explains a large part of preference – both in B2B and B2C markets.

During the recent decade, developments in the field of GPU computing paved the way for the rise of Deep Learning (DL). Since then, this powerful Machine Learning (ML) technique has been proven to outperform humans as well as state-of-the-art algorithms in various fields of application. However, difficulties in acquiring suitable datasets for training or special reliability requirements can render the utilisation of DL methods challenging for researchers in certain application areas.

In this talk, I will introduce the general concept of DL and challenges arising thereof in the specific context of Biomedical Imaging. Starting off with the view of DL as a 'black box', I will contrast the paradigm of DL methods with classical algorithms in the field and point out the arising consequences for interpreting results. In particular, the definition of a so-called ‘ground-truth’, i.e. a reference for training and validation of a DL method, is often a difficult task. In view of that and the limited traceability of DL-based algorithms, the question of how such methods can potentially support doctors in daily clinical practice also comprises an ethical component.

All in all, this presentation should contribute to the discussions around the question: Are we able to fully comprehend a powerful tool like Deep Learning in order to use it for making crucial decisions.

Although gender equality is demanded in work-place across industries, the reality in many German companies is different. Women are clearly underrepresented in many occupations. Gender stereotypes are widespread and well documented in social psychological literature. In general, women are described as more social and connected than men, while men are more associated with leadership and activity. Numerous studies on the use of language by men and women have revealed significant differences between the both. For instance, women are said to use more social words, more emotional words or more words concerning relationships compared to men. These findings aroused our curiosity and we asked ourselves whether discriminatory gender effects could be found in job advertisements in the German job market.
To get to the bottom of this question, we collected over 32,000 job advertisements from online job boards and divided them into occupations with a high proportion of men or women. Finally, we examined the language of these job advertisements using the 100 Worte text analysis. The results were clear: job advertisements in male-dominated occupations contained more masculine words than those in female-dominated occupations. The same applies - even to a greater extent but in the opposite direction - to female-dominated occupations. We were thus able to replicate the findings of other researchers and, for the first time, prove them for the German region. Based on these findings, we come to the conclusion that there are structural inequalities in the language of job advertisements that perpetuate existing gender differences.