Detecting interaction links in a collaborating group using manually annotated data

Identification of network linkages through direct observation of human interaction has long been a staple of network analysis. It is, however, time consuming and labor intensive when undertaken by human observers. This paper describes the development and validation of a two-stage methodology for automating the identification of network links from direct observation of groups in which members are free to move around a space. The initial manual annotation stage utilizes a web-based interface to support manual coding of physical location, posture, and gaze direction of group members from snapshots taken from video recordings of groups. The second stage uses the manually annotated data as input for machine learning to automate the inference of links among group members. The manual codings were treated as observed variables and the theory of turn taking in conversation was used to model temporal dependencies among interaction links, forming a Dynamic Bayesian Network (DBN). The DBN was modeled using the Bayes Net Toolkit and parameters were learned using Expectation Maximization (EM) algorithm. The Viterbi algorithm was adapted to perform the inference in DBN. The result is a time series of linkages for arbitrarily long segments that utilizes statistical distributions to estimate linkages. The validity of the method was assessed through comparing the accuracy of automatically detected links to manually identified links. Results show adequate validity and suggest routes for improvement of the method. Network data come from a variety of sources, including sureys, email depositories, analysis of documents and archives, and irect observation. Direct observation of networks has a long hisory, stretching back to the original sociometric studies (Moreno, 951), the Bank Wiring Room studies conducted by Hawthorne esearchers and analyzed by Homans (1951), and early anthropoogical work (e.g., Kapferer, 1969). In recent years, gathering of network data through direct bservation is less common than collecting data via surveys, recontructing links from archival or media documents, and analysis f digital data on connections. Direct observation is extremely Please cite this article in press as: Mathur, S., et al., Detecting interaction lin (2012), http://dx.doi.org/10.1016/j.socnet.2012.04.002 ime and resource-intensive, and the expense becomes almost proibitive if we want to study network dynamics over time. ∗ Corresponding author. E-mail addresses: [email protected] (S. Mathur), [email protected] M.S. Poole), [email protected] (F. Peña-Mora), [email protected] M. Hasegawa-Johnson), [email protected] (N. Contractor). 378-8733/$ – see front matter © 2012 Elsevier B.V. All rights reserved. ttp://dx.doi.org/10.1016/j.socnet.2012.04.002 © 2012 Elsevier B.V. All rights reserved. However, there are compelling reasons for using direct observation to study networks. It offers a useful complement to self-reports of ties, especially if investigators can collect both types of data. If a permanent record of the observation can be made, for example by video or audio recording the observations, then additional facets of meaning can be adduced and used to supplement the network data. For example, semantic networks derived from transcriptions of interaction can be related to social or communication networks. One important barrier to direct observation of networks is the time and effort it takes. For even small networks, coding links observationally requires multiple coders to watch different subsets of subjects in real time. For larger networks, such as networks of emergency responders who may number in the hundreds, the task becomes truly formidable and forbidding. One way to reduce the time and effort required for direct observation of networks is to automate the process of link detection ks in a collaborating group using manually annotated data. Soc. Netw. as much as possible. This manuscript reports on the development of an algorithm for detection of network links from video data that is amenable to automation. The algorithm utilizes visual cues which current automated computing systems can detect. It