Network Embeddings DeepWalk 2020-2022

This dataset contains network embeddings of all individuals who are registered in the Basisregistratie Personen (BRP) on January 1st YYYY and part of at least one population network file, i.e., BURENNETWERKYYYYTAB, COLLEGANETWERKYYYYTAB, FAMILIENETWERKYYYYTAB, HUISGENOTENNETWERKYYYYTAB, KLASGENOTENNETWERKYYYYTAB. All network datasets from the same year were concatenated and duplicate relations between every pair of individuals removed. The network was made symmetric by adding missing reciprocal relations. The resulting undirected and unweighted network was used to create the embeddings. Network embeddings are numerical representations with a fixed number of dimensions that encode the position of an individual in the network. The embeddings in this dataset were created using the <a href="https://doi.org/10.1145/2623330.2623732" target="blank"> DeepWalk</a> algorithm and have 32, 64, or 128 dimensions. The DeepWalk algorithm samples node sequences (random walks) from a network and creates embeddings by training a <a href="https://proceedings.neurips.cc/paper/2013/hash/9aa42b31882ec039965f3c4923ce901b-Abstract.html" target="blank">Skip-gram</a> model to predict nodes in each sequence from a context window. The embeddings in this dataset were created using 100 sequences of length 20 starting at each node in the network. The size of the context window was 5. The SkipGram model was trained for 2 epochs, that is, each sequence was iterated over twice. The dataset files follow the schema NETEMBEDDEEPWALKYYYYDIMXXX where DIMXXX refers to the number of dimensions.