EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2927. Using integer programming, Markov logic networks and inference to help find missing shareholders in social networks for a New Zealand Māori indigenous incorporation

Invited abstract in session TC-31: Analytics and the link with stochastic dynamics III, stream Analytics.

Tuesday, 12:30-14:00
Room: 046 (building: 208)

Authors (first author is the speaker)

1. Andrew J Mason
Dept Engineering Science, University of Auckland
2. Karl Zhu
University of Auckland
3. Anthony Downward
Engineering Science, University of Auckland

Abstract

Parininihi ki Waitotara (PkW) is a New Zealand indigenous organisation lead by Māori that manages land and other assets on behalf of thousands of shareholders. Many of these shareholders are “lost” in the sense that PkW has no contact details for them. As a consequence, PkW has millions of dollars of dividends waiting to be paid if these shareholders can be found. We have developed a system to collect public social network data and search this for the missing PkW shareholders. This involves solving the entity resolution problem to determine whether multiple document records refer to the same person or people. Traditional entity resolution approaches treat all matching decisions independently and only compare attribute similarities between reference pairs. However, this independence assumption omits valuable relational information in situations where the references describe a rich network of relationships between people. Collective entity resolution - where entities are resolved jointly - incorporates this previously ignored information and can predict matches with greater accuracy. We use Markov logic networks to convert our domain knowledge and evidence data into a Markov network. Prediction is made by performing most probable explanation (MPE) inference on the network using integer programming formulations and a mixed IP solver. We report experiments that demonstrate how our collective entity resolution approach is able to resolve complex relational information.

Keywords

Status: accepted

Back to the list of papers