Trajectories of Search - Mapping Scientific Frontiers: The Quest for Knowledge Visualization - page 157

Fig. 4.12 Overview first, zoom in, filtering, detail on demand. Accumulative trajectory maps of
user jbr in four consecutive sessions of tasks. Activated areas in each session reflect the changes of
the scope (clockwise: Task A to Task D)
Tabl e 4. 1 The state sequence generated by the HMM for user jbr. Relevant documents are
in bold type
67 57 120 199 65 61 61 61 73 73 73 87 170 134 105 170 142 172 156 112 192 77 47 138
128 114 186 30 13 13 18 114 135 50 161 50 43 50 66 50 50 66 161 66 66 169 66 66 169
169 123 123 83 149 169 169 123 123 149 149 83 11 138 159 121 123 149 149 100 100 91
91 83 83 119 83 83 119 119 83 41 162 162 82 50 82 82 82 82 161 122 31 43 135 81 161 43
43 135 81 81 135 14 135 135 14 14 20 20 80 80 189 189 152 56 189 189 64 64 158
In addition to the above approach, one can derive an HMM by using the Baum-
Welch algorithm based on a given sequence of observed actions. We use user jbr's
action sequence as the input and generate an HMM.
Using the Hidden Markov model derived from user jbr's and user nol's actual
sequences, we can verify the internal structure of the model using the well-known
algorithm - the Viterbi algorithm. Given a Hidden Markov model œ and a sequence
of observed symbols, the Viterbi algorithm can be used to generate a sequence
of states. One can examine this state sequence and compare it with the original
sequence of events log from the user.
Tab le 4.1 shows the state sequence generated by the Viterbi algorithm based on
the HMM œstate, which returns the sequence of states that is most likely to emit the
observed symbols, i.e. the information foraging sequence. Relevant documents in
the state sequence are highlighted in bold. This sequence is of course identical to
the original sequence recorded in the session.
Based on the HMM œstate, user jbr's observed information foraging action
sequence as the input and applied the Viterbi algorithm to generate the optimal
state transition path. Figure 4.13 shows the path of the sequence generated by the