As people engage with online platforms by the millions, they generate massive, unstructured digital traces of their behavior. While these records contain rich, fine-grained, and voluminous information about how people interact, it remains challenging to interpret and draw meaning from the macroscale structure that emerges from them. In this project, we introduce and develop behavioral embeddings, a neural embedding methodology for extracting insight from large-scale, unstructured social data. In behavioral embeddings, entities such as communities, tags, or articles are embedded in a high-dimensional space such that two entities are close in the space if and only if similar users engage with them. We demonstrate that this notion of social similarity provides an extremely useful framework for understanding the macroscale structure of large social datasets.

Links

Publications

Community embeddings reveal large-scale cultural organization of online platforms
Isaac Waller and Ashton Anderson. Working paper.
BibTeX
@article{waller2020community,
  title={Community embeddings reveal large-scale cultural organization of online platforms},
  author={Waller, Isaac and Anderson, Ashton},
  eprint={2010.00590},
  archivePrefix={arXiv},
  year={2020}
}
Generalists and Specialists: Using Community Embeddings to Quantify Activity Diversity in Online Platforms
Isaac Waller and Ashton Anderson. WWW 2019.
BibTeX
@inproceedings{waller2019generalists,
  title={Generalists and specialists: Using community embeddings to quantify activity diversity in online platforms},
  author={Waller, Isaac and Anderson, Ashton},
  booktitle={The World Wide Web Conference},
  pages={1954--1964},
  year={2019}
}