The making of the incredibly differentiated labor market

Project teamFüllsack, Manfred (Uni Graz); Kaiser, Max (ÖNB); Kleinert, Jörn (Uni Graz); Vogeler, Georg (Uni Graz) InstitutionsUniversity of Graz, Austrian National Library (ONB) We describe the emergence of and the continuous change in the labor market in Austria over a time span of about 100 years from the mid of 19th to the mid of 20th century through the lens of job ads in newspapers. A large pool of job ads generated from the ANNO newspaper database provided by the ONB is the basis of our analysis. From this pool we use the advertisements pages of the 35 highest-circulation German-language newspapers from Austria-Hungary and isolate the individual job advertisements. As the almost 8 million newspaper pages of the 35 newspapers cannot be analysed manually, we need an algorithm to help and translate the job offers into machine-readable language. . To assess these short, differentiated texts, we rely on background knowledge about the early Austrian labor market and on theoretical insights from the two-sided matching process. Natural language processing techniques (sentiment analysis, topic modeling) are harnessed to assess the huge amount of qualitative information. We aim for three goals: First, we intend to generate a unique data source for empirical analyses in various fields of humanities and economics, assuming that job advertisements in newspapers are a great source of individuals wills, wishes and offers, which are socially embedded but widely unfiltered by others. Second, we want to illustrate the strong change in labor relationships in many dimensions such as extent, regional reach, sector focus, skill requirements, job characteristics, self- perceptions of employers and employees and expected or pretended relationships. Job ads hold many qualitative information about the vacancy and the candidate which can be assessed using modern natural language processing tools. We document the emergence, development and differentiation of the Austrian labor market and job ads are the means to overcome incomplete information in this market. Third, we want to test and further develop digital methods for text mining and text analyses on a database of millions of advertisements with fairly heterogeneous content and structure. Standard algorithms will not suffice in our project. We will rely on adjusted cleaning and correction procedures as a first study based on a small sample has shown. The efforts are nevertheless worth doing it: we will provide a database of millions of advertisements from a period of 100 years which hold qualitative raw data. We are confident that this data is interesting for several fields of social science also beyond our interest in heterogeneity in the labor market.