Personalization is managed by the humanIT Dynamic Personalization Server (DPS) [4], an open, standards-based, and platform-independent tool that provides essential user-modeling services to user-adaptive applications (see Figure 2). To produce the personalized list of top news, the publishing system queries the DPS Directory Component for a customer’s (presumed) interests. It uses the DPS response to in turn query the content management system and news database for relevant news items that match the interest profile returned by the DPS.
Whereas previous user-modeling systems stored data about users in database and knowledge representation systems [1, 2], DPS employs a Lightweight Directory Access Protocol (LDAP)-based directory management system [3]. This offers significant advantages with respect to the:
- Management and retrieval of (user-related) information, which is compliant with widely established standards allowing for easy integration.
- Addition of new information types to the set of predefined types, which allows for flexibility as sites and personalization goals evolve.
- Distribution of information across a network leading to better performance, scalability, availability, and reliability.
- Provision of facilities for authentication, signing, encryption, access control, auditing, and resource control for ensuring information security and users’ privacy.
The user profile acquisition process is implicit except for the initial registration. The system reports selected interaction events, enriched by content information, to the DPS. Different interaction types (for example, viewing the headline versus requesting the full text of a news item) carry different weights.
The DPS Directory Component stores user profiles as nodes in an LDAP directory, making the representation explicit and human-readable. User profiles can be inspected and/or modified via a browser interface and exported for analysis purposes or used as input to operative CRM applications.
The DPS component integrates multiple learning components that continuously process incoming event information in the background. The user-learning component determines and tracks particular interests of individual users over the time span of multiple sessions. The collaborative filtering component determines a “peer group” for each user based on user-interest profiles that are most similar to the profile of a particular user. This data can be used efficiently to predict the likelihood that a user is interested in a certain topic in cases where no prediction can be made based on the user’s individual interaction record. By integrating these techniques into a single server, we can leverage several synergistic effects between these techniques and compensate for their well-known deficits with regard to performance, scalability, integration of domain knowledge, and sparsity of data.
The humanIT personalization solution for N24 can be deployed in a wide range of e-business scenarios. The only requirements are a content management system offering classified content and a site’s ability to track individual customers.
Join the Discussion (0)
Become a Member or Sign In to Post a Comment