4.3 Adaptivity of the user model In CRUMPET an adaptive user model learns user interests from the user's interaction with the system. When a user asks for more information about an object, this adds a small amount to the evidence that a user is interested in objects with these features, more than in others. If a user asks for more and more details about the same object, or even asks for directions to a site/restaurant, this adds a greater amount to the evidence that she or he is interested in such services. The user model, in order to learn user interests, needs reliable and intelligent feedback from the user interface. Another idea would be to use a user‘s movements to infer her or his interests. So, for example, it seems a good idea to infer from a user’s visit to a church that this user is interested in churches. (Again, this only adds a little to the statistical evidence of this user’s interest.) While this approach may be useful in a small and closed world, such as a museum, it is risky in the open world, such as a city. The localisation is not precise enough to determine the user’s topological position without doubt. There may also be several reasons why a user stays for a while at a certain position, and these reasons may not be obvious to the system. So, for instance, the visit in or near a church may be because of a concert in the church, or an exceptional view from its tower, or even the little café in the aisle may attract tourists, who are not generally interested in churches; and none of these reasons can be modelled and determined by the system. So in CRUMPET, which is a guide in the open world, we decided to rely on the user’s interaction with the system to learn her or his interests over time. Learning user interests unfortunately requires a number of events until statistically significant tendencies can be postulated. Several mechanisms to accelerate this are possible. One "shortcut" would be asking the user to confirm early guesses explicitly. Another, complementary, solution will be to initialise the user model using stereotypes. A stereotype is a (small) set of demographic data correlating to a set of
typical interests of such tourists. The most appropriate stereotype to start with can then be identified by a few demographic attributes that the user states when registering to the system. In case a user’s profile has been initialised by an inappropriate stereotype, this would be adjusted over time (implicitly) by learning, or can also be corrected by the user explicitly. The validity of stereotypes, i.e. the correlation of certain demographic data with “typical” user interests, needs to be established by a separate empirical study.
4.3 Adaptivity of the user model In CRUMPET an adaptive user model learns user interests from the user's interaction with the system. When a user asks for more information about an object, this adds a small amount to the evidence that a user is interested in objects with these features, more than in others. If a user asks for more and more details about the same object, or even asks for directions to a site/restaurant, this adds a greater amount to the evidence that she or he is interested in such services. The user model, in order to learn user interests, needs reliable and intelligent feedback from the user interface. Another idea would be to use a user‘s movements to infer her or his interests. So, for example, it seems a good idea to infer from a user’s visit to a church that this user is interested in churches. (Again, this only adds a little to the statistical evidence of this user’s interest.) While this approach may be useful in a small and closed world, such as a museum, it is risky in the open world, such as a city. The localisation is not precise enough to determine the user’s topological position without doubt. There may also be several reasons why a user stays for a while at a certain position, and these reasons may not be obvious to the system. So, for instance, the visit in or near a church may be because of a concert in the church, or an exceptional view from its tower, or even the little café in the aisle may attract tourists, who are not generally interested in churches; and none of these reasons can be modelled and determined by the system. So in CRUMPET, which is a guide in the open world, we decided to rely on the user’s interaction with the system to learn her or his interests over time. Learning user interests unfortunately requires a number of events until statistically significant tendencies can be postulated. Several mechanisms to accelerate this are possible. One "shortcut" would be asking the user to confirm early guesses explicitly. Another, complementary, solution will be to initialise the user model using stereotypes. A stereotype is a (small) set of demographic data correlating to a set oftypical interests of such tourists. The most appropriate stereotype to start with can then be identified by a few demographic attributes that the user states when registering to the system. In case a user’s profile has been initialised by an inappropriate stereotype, this would be adjusted over time (implicitly) by learning, or can also be corrected by the user explicitly. The validity of stereotypes, i.e. the correlation of certain demographic data with “typical” user interests, needs to be established by a separate empirical study.
การแปล กรุณารอสักครู่..
