Personalization and Adaptation
Central to the proposed architecture is the VR-mall personalization engine, comprising the user profiles database and the user-modeling engine. The user profiles database stores information regarding the profiles of individual users, and associations between user profiles and specific items or item properties. This information is utilized in the process of VR-mall creation to include in the mall the merchandise that most closely matches the interests of the current e-shopper. For each e-shopper, the user profiles database hosts both static and dynamic information.
Static information reflects characteristics that remain constant, at least in the context of the current visit (e.g. preferred language, connection speed, age, etc.). This information is either entered by the user (e.g., a response to a “Language selection” prompt) or deduced by the system (e.g., connection speed is estimated by measuring the download time for an image of known size). Dynamic information pertains to the interaction of the user with the virtual environment and is collected by the user activity recorder. This information describes certain actions that the e-shopper has performed in the VR-mall, including moving close to an item and moving away from it, start and end of item manipulation, resetting activities (probably due to disorientation problems), acquiring and losing visibility for items, requesting specific resources or resource types, and so forth. This information is collected within the user browser, and communicated to the user activity recorder periodically. When the user activity recorder receives a group of event information, it first arranges to combine “activity beginning—activity end” records, to compute the duration of each activity. For events that are instantaneous by nature (e.g., request for an image of a product), only the count of these events is computed. The combined information is inserted into the user profiles database, and the user-modeling engine is invoked to update the profile of the user.
The user-modeling engine is a separate architectural module that examines the user activities observed within the virtual environment, and deduces the preferences of the user towards certain items or item categories. Upon invocation, the user modeling engine extracts from the user profiles database the records that describe activities of the current user and processes them as follows:
1. Items that have come into visibility are assigned a grade in the range of (-5) to (10), depending on the time that they have attracted the visitor’s attention (-5 = not at all, 10 = very long).
2. If for some item some resources have been explicitly requested (e.g., 3D models, detailed text, images), an extra amount is added to the item’s interest grade (1 to 3, depending on the time the extra resource was viewed).
3. Items that have not come into visibility are not assigned any grade, as the user may ignore altogether that the items were present in the VR-world.
The final grade for each item is computed by multiplying the above computed grade with an aging factor, which ranges from 1.0 (for recent activities) and 0.1 (for activities that occurred a long time ago). This step effectively assigns a greater importance to recent activities, allowing for modeling of changing user interests (Kilfoil, Ghorbani, Xing, Lei, Lu, Zhang, et al., 2003).
The last phase of the user profile update procedure is the mapping of the grades computed in the previous step to preferences towards item categories, or—more generally— item properties. To this end, the semantic information associated with the items (in the form of propertyvalue pairs) is extracted from the digital items representation repository. For each property-value pair retrieved, the grades of all items associated with it are summed up to form the score of the specific property-value pair. This information is finally inserted into the user profile database. Similar algorithms are used to deduce user preferences towards specific media types and interaction methods.
Static information reflects characteristics that remain constant, at least in the context of the current visit (e.g. preferred language, connection speed, age, etc.). This information is either entered by the user (e.g., a response to a “Language selection” prompt) or deduced by the system (e.g., connection speed is estimated by measuring the download time for an image of known size). Dynamic information pertains to the interaction of the user with the virtual environment and is collected by the user activity recorder. This information describes certain actions that the e-shopper has performed in the VR-mall, including moving close to an item and moving away from it, start and end of item manipulation, resetting activities (probably due to disorientation problems), acquiring and losing visibility for items, requesting specific resources or resource types, and so forth. This information is collected within the user browser, and communicated to the user activity recorder periodically. When the user activity recorder receives a group of event information, it first arranges to combine “activity beginning—activity end” records, to compute the duration of each activity. For events that are instantaneous by nature (e.g., request for an image of a product), only the count of these events is computed. The combined information is inserted into the user profiles database, and the user-modeling engine is invoked to update the profile of the user.
The user-modeling engine is a separate architectural module that examines the user activities observed within the virtual environment, and deduces the preferences of the user towards certain items or item categories. Upon invocation, the user modeling engine extracts from the user profiles database the records that describe activities of the current user and processes them as follows:
1. Items that have come into visibility are assigned a grade in the range of (-5) to (10), depending on the time that they have attracted the visitor’s attention (-5 = not at all, 10 = very long).
2. If for some item some resources have been explicitly requested (e.g., 3D models, detailed text, images), an extra amount is added to the item’s interest grade (1 to 3, depending on the time the extra resource was viewed).
3. Items that have not come into visibility are not assigned any grade, as the user may ignore altogether that the items were present in the VR-world.
The final grade for each item is computed by multiplying the above computed grade with an aging factor, which ranges from 1.0 (for recent activities) and 0.1 (for activities that occurred a long time ago). This step effectively assigns a greater importance to recent activities, allowing for modeling of changing user interests (Kilfoil, Ghorbani, Xing, Lei, Lu, Zhang, et al., 2003).
The last phase of the user profile update procedure is the mapping of the grades computed in the previous step to preferences towards item categories, or—more generally— item properties. To this end, the semantic information associated with the items (in the form of propertyvalue pairs) is extracted from the digital items representation repository. For each property-value pair retrieved, the grades of all items associated with it are summed up to form the score of the specific property-value pair. This information is finally inserted into the user profile database. Similar algorithms are used to deduce user preferences towards specific media types and interaction methods.
posted by Jane at 1:14 PM
0 Comments:
Post a Comment
<< Home