Database Reference
In-Depth Information
Different users tend to select different sequences for accomplishing the
same goal. Specifically, the user's attributes and context (such as age or the
operating system) indicate which sequence the user will eventually perform.
For example, if a young male uses a system, it is possible to use the model
to predict the goal that he intends to accomplish and provide him with a
user interface (UI) more appropriate to his usage and intentions.
The reason for using hidden Markov models is that the sequence of
actions that users perform is not always observable but can only be observed
through another set of stochastic processes that produce the sequence of
observations. For example, in a mobile device the sequence of buttons that
the user pressed is unknown (hidden) but the sequence of screens that was
created from using these buttons is known (observable). Thus, we can learn
the sequence of buttons from the sequence of screens using HMM.
Figure 8.1 illustrates a HMM tree. In this case, the tree is branched
according to the user's attributes and each node in the tree consists of
Hidden
state 1
Hidden
state 2
Hidden
state3
0.4
0.7
0.3
0.5
0.3
1.0
0.5
0.4
Observed
state a
Observed
state c
Observed
state b
Age group
Young
Old
Hidden
state 1
Hidden
state 1
Hidden
state 2
Hidden
state 2
Hidden
state 3
Hidden
state 3
0.3
0.1
0.85
0.5
0.4
0.7
1.0
0.4
0.3
0.2
0.8
1.0
0.2
0.6
0.4
Observed
state a
Observed
state c
Observed
state a
Observed
state c
Observed
state b
Observed
state b
Device
Gender
Fig. 8.1 Structure of attribute-driven HMM tree. Each node in the hierarchical structure
contains a model and an attribute. All nodes contain the same model but with different
transition probabilities.
Search WWH ::
Custom Search