TY  - GEN
ID  - heidok24112
TI  - Social Network Extraction and Exploration of Historic Correspondences
Y1  - 2018///
A1  - Li, Hui
UR  - https://archiv.ub.uni-heidelberg.de/volltextserver/24112/
AV  - public
N2  - Historic correspondences, in the form of letters, provide a scenario in which historic figures and events are reflected and thus play a ubiquitous role in the study of history. Confronted with the digitization of thousands of historic letters and motivated by the potentially valuable insights into history and intuitive quantitative relations between historic persons, researchers have recently focused on
the network analysis of historic correspondences. However, most related research constructs the correspondence networks only based on the sender-recipient relation with the objective of visualization. Very few of them have proceeded beyond the above stage to exploit the detailed modeling of correspondence networks, let alone to develop novel concepts and algorithms derived from network analysis or formal approaches to the data uncertainty issue in historic correspondence.

In the context of this dissertation, we develop a comprehensive correspondence network model, which integrates the personal, temporal, geographical, and topic information extracted from letter metadata and letter content into a hypergraph structure. Based on our correspondence network model, we analyze three types of person-person relations (sender-recipient, co-sender, and co-recipient) and two types of person-topic relations (author-topic and sender-recipient-topic) statically and dynamically. We develop multiple measurements, such as local and global reciprocity for quantifying reciprocal behavior in weighted networks, and the topic participation score for quantifying interests or the focus of individuals or real-life communities. We investigate the rising and the fading trends of topics in order to find correlations among persons, topics, and historic events. Furthermore, we develop a novel probabilistic framework for refinement of uncertain person names, geographical location names, and temporal expressions in the metadata of historic letters.

We conduct extensive experiments using letter collections to validate and evaluate the proposed models and measurements in this dissertation. A thorough discussion of experimental results shows the effectiveness, applicability and advantages of our developed models and approaches.
ER  -