A

Graph Mining and Network Reconstruction from Incomplete Information: From Missing Statistics to Fragmentary Evidence

ACM International Conference on Information and Knowledge Management (CIKM), pp. 9314–9318

Abstract

Many information-management and data-mining pipelines assume that the graph is already observed. In practice, relational structure may itself need to be constructed from aggregate constraints, node covariates, public signals, and partial records. My research studies graph mining and network reconstruction under incomplete information, with emphasis on how different information sources should be represented, combined, and evaluated when the adjacency matrix is hidden. Economic and financial networks serve as demanding testbeds because dyadic relations are often confidential, delayed, or only partially observable; they are not the primary object of the thesis. The dissertation is organized around four connected components: reconstruction with missing sufficient statistics, directed graph inference from public time series, information-based design of partial observations, and inference from passively observed fragmentary records. The fourth component addresses settings where researchers cannot choose what is observed: complete outgoing records may exist for some nodes and only known subsets of possible dyads for others. Together, the four components form an information-management and data-mining agenda for constructing and evaluating graph representations from incomplete data.

Authors 1

  1. Leiden University · University of Pisa · IMT School for Advanced Studies Lucca

    Affiliation as printed

    Computer Science Department, University of Pisa, Pisa, Tuscany, Italy, Faculty of Science, Leiden University, Leiden, Netherlands, and IMT School for Advanced Studies Lucca, Lucca, Italy

Cited by 0 stored of 0

References 14

14 results