Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Enseignement superieur --- Education, Higher --- Recherche --- Research --- Belgium.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

In this thesis, we investigated the relationship between human capital and (subsequent) financial performance of academics in a research-intense university. In order to describe this relationship, we made use of publication data from a publicly available database (LIRIAS), and combined this with HR data from the university as well as financial metrics related to the academics. The features we used to describe the relationship between both domains can be broadly divided into six categories: i) time-related, ii) social network metrics, iii) publication-related, iv) organisational unit of the academics, iv) node role metrics and vi) access to research manager roles. For the node role metrics we used the recently released GraphWave algorithm, which encodes inter-node influences using a spectral method into a vector format. A first part of the research consisted of validating the suitability of using GraphWave embeddings to properly distinguish different node roles. This was done first on model systems and subsequently on more complex real networks constructed from publication data. A second part of the research consisted of training decision regression trees (selected for their high level of explainability). We compared the results of single decision trees, random forests and extreme gradient boosted models. We found that the latter 2 were significantly better to relate human capital related features to financial performance than the first one. We also observed that relative importances of the feature categories describing the relationship were very dependent on the time (Horizon) between human capital metrics and financial metrics. More specifically, we observed that the dominant features at low Horizon values were the time-related features and the betweenness centrality. These features decreased in relative importance as the Horizon got larger. The number of publications and number of assignments to different organisational units were quite important too, and became dominant at the large Horizons. Other social network related parameters (such as different types of degrees and centrality metrics) were relevant throughout the entire range of Horizon values. The access to research managers profile in the network showed a minor influence at short times. An interesting observation was related to the GraphWave features from which we selected 20 different network roles as representative for the entire collaborative network using clustering techniques. It was found that for the XGB model, these features - when taken jointly - were in fact the dominant features for all Horizon values.