Spectral Clustering of Multiple Choice Questions

Process

Input

Answers (a-e) on 10 questions from the midterm in Computer Networks (UCR CS164) for 53 students.

Processing

Processing was identical to the processing done in my first multiple choice clustering experiement, with the additional step of performing spectral clustering as a pre-processing step before running the hierarchical clustering algorithm. (NOTE also that the spectral clustering outputs points located in Rk, rather than the distances between points in a (possibly non-Euclidean) space. So a distance-finding step akin to the initial M.I. step was also performed to allow maximal code re-use.)

About the dataset

The first 5 questions are T/F, the remaining questions are multiple choice (5 responses).
10-node dendrogram

Independent Groupings

I asked for perceived groupings of the questions from people I know that have a strong networks background. These people were given only the question text, as listed below, they did NOT have access to the score information. (Conversely, this clustering is done with only score information, not question text.) I got 5 responses, listed here: In this groupings, a number of pairings recur: Only one of these identified pairings is held up in the clustering. Several of the more spurious clusterings are present though, such as 3, 4 @ 80% (1 level off), but many are not.
  1. Throughput and Latency mean basically the same thing
  2. In the Python socket API, the listen method of Socket objects blocks until a new connection has been established and returns the new connection along with the address making the connection
  3. The error-detection algorithm used in Ethernet depends on the physical medium that it is used on.
  4. An Ethernet adaptor will only send out Ethernet frames that have the correct Ethernet source address
  5. The number of TCP ports available on a host is equal to the number of wires in the physical cable connecting the host to the network.
  6. Which of the following is more of a concern in ATM than Ethernet?
  7. Which of the following errors can NOT be introduced by an unreliable delivery mechanism?
  8. For a Stop-and-Wait protocol to transmit n frames, it requires a sequence number of how many bit(s)?
  9. In a Sliding-Window protocol, the RWS governs
  10. Hosts on an Ethernet have to process the following fields in an Ethernet frame header