Assignment:
You will go through the process of creating decision networks from meeting transcripts. This will involve an introduction to decision network theory, to content analysis, and the construction and analysis of decision networks using ORA.
The deliverables are the following:
1. Coded excerpt in spreadsheet form.
You will segment the excerpt into decision discussions according to the coding guidelines, then assign codes to each segment.
2. Inter-rater reliability analysis.
You will compare your segmentation and coding with the work of one other student, and discuss the differences. You will agree on a segmentation, and independently code the excerpt, and compute the Cohen's kappa score as a measure of inter-rater agreement.
3. Network constructed from excerpt.
Show a visualization of the network you construct from the data you just coded. As briefly discussed in class (see "relevant slides" below), the idea is to create networks of components* from the "logical" dependencies, where, in this case, we infer a logical dependency when two components are mentioned in the discussion of the same decision. So, for example, if "power" and "mobility" are mentioned in conjunction with the same decision, then we infer a link between them. These networks can be binary or weighted -- use whichever you feel best achieves the purpose of the assignment, which is to understand how the components are technically related to one another. For this part of the assignment, you may create the network however you like, by hand or with your favorite tool.
If you are not finding sufficient links to make an interesting network, then your "decision discussions" may be too small. You may need to create larger chunks so that there are co-occurrences of components within a discussion.
4. Analysis of provided network.
Explore network data that is provided (the "network data" link, below), and describe insights into the relationships observed. The provided data is from one full meeting, coded in much the same manner that you coded the brief excerpt for tasks 1-3. To what extent do components form independent clusters? What do the patterns you observe suggest about coordination among the teams working on the various components? The data can be downloaded and opened in ORA. The network has three types of nodes, segments, decision discussions, and components (see slides, link below). Try various visualizations to understand how the components are related. The simplest possible network is the components by themselves, so you can see how they are linked with minimal clutter. Including the decisions allows you to see more detail about the basis of the links in the particular decision discussions.
* Unfortunately, the word "components" is ambiguous in this context. It could mean parts of the robotic system being designed, or it could mean independent parts of the network structure. In this assignment, I use the term "components" only in the first sense, as parts of the robotic system being designed.
Resources:
Brief introduction to content analysisCoding guidelinesExcerpt to codeNetwork datarelevant slides