{"id":500,"date":"2021-06-15T13:55:25","date_gmt":"2021-06-15T13:55:25","guid":{"rendered":"http:\/\/hnr2021.historicalnetworkresearch.org\/?page_id=500"},"modified":"2021-06-18T09:29:28","modified_gmt":"2021-06-18T09:29:28","slug":"modeling-social-context-improves-role-detection-in-communication-network-of-16th-century-european-reformers","status":"publish","type":"page","link":"http:\/\/hnr2021.historicalnetworkresearch.org\/?page_id=500","title":{"rendered":"Modeling social context improves role detection in communication network of 16th century European reformers"},"content":{"rendered":"\n

Ramona Roller <\/em>and Frank Schweitzer <\/em><\/h2>\n\n\n\n

Time and Place:<\/strong> Thursday, 01.07., 11:55\u201312:15, Room 2
Session:<\/strong> Networking Correspondences<\/p>\n\n\n\n

The Reformation was a major transformative movement in early modern Europe which overthrew the established clerical order.1<\/sup> It also promoted social dynamics such as confessionalisation<\/em>, where religious issues became institutionalised and state-controlled and professionalisation<\/em>, where people adopted unique professions.2<\/sup> These social dynamics unfold their effect on the scale of society but they originate from the behavioural patterns of individual people who lived in the 16th century. What do these behavioural patterns look like and how can we capture them are the questions addressed in this work.<\/p>\n\n\n\n

We try to capture behavioural patterns by extracting communication patterns of people from their letter correspondences. Let us illustrate some communication patterns taking the example of the Strasbourg re former Martin Bucer. Bucer educated many students from Switzerland in Strasbourg. One of these students, Simon Sulzer, had accused Bucer of withholding an important letter from him.3, 4<\/sup> This letter was sent from Heinrich Bullinger, the head of the Zurich church, to Bucer. According to Sulzer\u2019s allegations, Bullinger \u00a8 had asked Bucer in that letter to also forward the letter to Bullinger\u2019s compatriot Sulzer which Bucer had not done. Assuming Sulzer is right, Bucer\u2019s communication behaviour resembles that of a censor<\/em>, a person who deliberately withholds information from others and ensures that she controls the information flow. Did other reformers also communicate as censors<\/em>? The first purpose of our study is to identify communication patterns by detecting people with similar communication behaviour. We are interested in whether and why two reformers with similar communication patterns ended up pursuing different or similar lives.<\/p>\n\n\n\n

In the case of Bucer, we know from historical records that he did not remain a censor <\/em>in all of his communication but also adopted other patterns. For example, he acted as a mediator <\/em>in the Eucharistic Controversy <\/em>in Bern in 1537.5 <\/sup>This controversy, as many of its sort, dealt with the question of whether during the Lord\u2019s Supper Christ\u2019s presence is real or symbolic. One faction supported the Lutheran view (symbolic) whereas the other faction supported the Zwinglian view (real). Bucer who supported a middle-way tried to reconcile the two factions. The second purpose of our study is to identify changes in the communication behaviour of people over time. We can capture fast changes because letters were exchanged within days whereas other forms of interactions such as contracts or certificates were issued over months or years. We are interested in the extent to which dynamics in communication patterns can explain social dynamics during the Reformation such as confessionalisation or professionalisation.\u00a0<\/p>\n\n\n\n

In this analysis, we call those communication patterns communication roles<\/em>. Role detection is the process of (1) identifying communication patterns by grouping similar communication behaviours together and (2) labeling these patterns. Our role examples of censor <\/em>and mediator <\/em>illustrate that our extracted roles go beyond the binary division of \u2018senders\u2019 and \u2018recipients\u2019 by capturing more entangled communication pat terns. It is important to distinguish our roles from \u2018formal roles\u2019 a concept from sociology6, 7 <\/sup>and also often used in colloquial language. Formal roles <\/em>are behavioural patterns which are predefined by some authority (e.g. society) and often relate to professions, such as ambassador, merchant, pastor. In contrast, our commu nication roles are a form of informal roles<\/em>, behavioural patterns which emerge in a specific context through the interplay of people. Although the role concept is a term from modern sociology, we do not try to map specific 21st century roles to a 16th century setting. In contrast, our aim is to extract roles for a specific data set. As a consequence, our extracted roles are not universal and cannot be generalised to other data sets, even if the two data sets are from the same historical period.\u00a0<\/p>\n\n\n\n

Because communication roles are context-specific role detection in networks cannot rely on topological measures alone but also has to pay attention to the multi-dimensional embedding of nodes in the social system.8 <\/sup>Previous research on role detection in networks has used dimensionality-reduction techniques on a pool of node-specific network measures. In order to account for the social context of roles, authors have developed node-specific measures from the content of messages in communication networks,9, 10 <\/sup>and used them together with pure topological measures.11, 12 <\/sup>However, this approach is problematic since message content is not always available in the data. Modeling the social context of roles in different ways may or may not affect role detection. Of specific interest is how network properties on the meso-scale, such as communities, and their relation with individual nodes, e.g. embeddedness of a node within a community, can be use to approximate the social context of roles.\u00a0<\/p>\n\n\n\n

We pose the following research question:\u00a0<\/p>\n\n\n\n

How can we extract context-specific roles of European reformers from their letter correspondence network?<\/p>\n\n\n\n

We extract topological and context-specific measures from a directed multi-edge letter correspondence network which is based on 17,000 letters, exchanged between 2,500 people between 1510 and 1575. These data are based on the letter correspondences of seven reformers which were crawled from open source repositories.13, 14 15, 16 17, 18, 19 <\/sup>We use the umbrella term \u2018reformers\u2019 to refer to all people in the data independent of whether they supported Protestantism and of the estate (clergy, nobles, bourgeoisie) they belonged to. <\/p>\n\n\n\n

As part of our data-preprocessing, we disambiguated the names of all senders and recipients for all seven letter collections together. This step was accomplished by a Historian who used a semi-automated computer programme which we wrote specifically for this purpose. The programme displayed two different names from our data set (e.g. \u2018Martin Luther\u2019 and \u2018Luther\u2019) and the Historian checked in the respective letters whether these names belonged to the same person or not. For example, we managed to disambiguate \u2018Justus Jonas\u2019 who may refer to \u2018the Elder\u2019 or \u2018the Younger\u2019 who are father and son, respectively. We also identified \u2018John Frederick I, Elector of Saxony\u2019 and \u2018Duke John Frederick\u2019 to be the same person at two different points in time: before and after the Schmalkaldic War. John Frederick had lost that war and, as a consequence, had to cede his electoral dignity leaving him with the less powerful title of a duke. In addition, we construct a spatial data set of the territories of the Holy Roman Empire, the main domain of power in the 16th century. The attributes of and relations between these territories provide a detailed view on the geo-political situation at that time.\u00a0<\/p>\n\n\n\n

We present a methodology to extract communication patterns from network measures and to provide names to these patterns, our communication roles. We use topological measures to account for the position dependent aspect of a role and time as well as homophily measures to represent the social context. Figure 1 shows how we represent social roles as a multi-layer network. (1) Topological measures<\/strong> correspond to centralities and the size of the local clustering coefficient. (2) Time-dependent measures<\/strong> account for the temporal ordering of links which can be derived by the sending date of the letters. These temporal measures can be used to quantify possible social influence and possible information spread by label propagation and the Infomap algorithm, respectively. (3) Homophily-based measures<\/strong> refer to different kinds of node similarity: structural, spatial and social similarity. Structural similarity considers the formation of communities in the letter correspondence network as well as the structure within and between these communities. Spatial similarity is based on the geographic distance between senders and recipients using the sending and receiving locations of letters in the data. Social similarity infers social attributes of the reformers from the territories they were located in.\u00a0<\/p>\n\n\n\n

This elaborate choice of network measures is necessary for three reasons. First, we require as many topological network measures as possible in our analysis to increase the coverage of communication patterns which we can possibly extract. A large coverage is necessary because we do not know in advance which roles can be extracted from the data set at hand. If we decreased the coverage of network measures we may miss important roles.\u00a0<\/p>\n\n\n\n

Second, we require temporal network measures to capture chronological information flow between senders and recipients via letters. If information flow was only captured by topological measures the chronology of letter correspondences would be distorted and our interpretation of communication patterns may be incorrect.\u00a0<\/p>\n\n\n\n

Third, we require homophily-based measures to account for sociological ingroup-outgroup principles in human interactions. These principles influence human interactions across cultures and historical periods.20<\/sup> If we did not include homophily-based measures we would falsely ascribe communication patterns to pure dyadic interactions without group effects. <\/p>\n\n\n\n

Social Role<\/p>\n\n\n\n

\"\"<\/a>
Figure 1: Exemplary layers of multi-dimensional embedding of social roles 
<\/figcaption><\/figure><\/div>\n\n\n\n

Using the example of the censor <\/em>we can characterise and model her by the following attributes and network measures, respectively. <\/p>\n\n\n\n

The Censor <\/em><\/p>\n\n\n\n