04.11.2014 Views

elektronická verzia publikácie - FIIT STU - Slovenská technická ...

elektronická verzia publikácie - FIIT STU - Slovenská technická ...

elektronická verzia publikácie - FIIT STU - Slovenská technická ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

236 Selected Studies on Software and Information Systems<br />

and provides a ranked list of similar users from the very beginning of user’s work with the<br />

system (so the system is able to provide also collaborative filtering based on users similarity).<br />

During the usage of the system, user browsing behavior is monitored via a proxy server,<br />

where each URL browsed during normal work activity is logged. A classifying algorithm is<br />

used to classify browsed URL’s based on a training set of labeled example papers, storing<br />

each new paper in a central database. The profiling algorithm performs correlation between<br />

paper topic classifications and user browsing logs. Whenever a research paper is browsed<br />

that has been classified as belonging to a topic or an explicit feedback on recommendation is<br />

provided, it accumulates an interest score for that topic. This is the way Quickstep contributes<br />

back to the bootstrap ontology.<br />

The weak point of the approach is the assumption that ontology used for bootstrap contains<br />

knowledge about the new user and that this user has already written some publications<br />

which could be compared with publications in the information space.<br />

User Model Mediation<br />

The concept of mediation of partial user models between several adaptive systems was proposed<br />

in [8]. The basic idea is that different applications would benefit from enriching their<br />

UMs through importing and aggregating user models which were built by other, possibly<br />

related applications. It is different from the centralized user model approach using user<br />

modeling server, where each application extracts the required data from the central UM<br />

and updates it later. The described approach is designed to be decentralized and providing<br />

ad-hoc (for a specific purpose) generation of user model for the target application through<br />

translation and aggregation of partial UMs built by other applications. It is therefore useful<br />

for solving cold-start problem.<br />

The mediation to be successful needs to solve several issues among which is user’s<br />

privacy, the structural heterogeneity and incompleteness of the user models content, since<br />

every application refers to a specific application domain only.<br />

The mediation process is partitioned to the following stages (see Figure 8-8):<br />

1. A target application, required to provide personalization to a user, queries the mediator<br />

for the user model related to the application domain.<br />

2. The mediator identifies the required personalization domain and the user model<br />

representation in the target application.<br />

3. The mediator determines a set of other applications that can potentially provide partial<br />

domain-related user models of the given user and queries them.<br />

4. Applications, actually storing the needed data, answer the query, and send to the<br />

mediator their partial user models of the given user.<br />

5. The mediator translates and aggregates the acquired partial user models (using the<br />

knowledge base (KB)) into a single domain-related user model, represented according<br />

to the target application.<br />

6. The generated domain-related user model is sent to the target application, which is<br />

capable of providing more accurate personalized service.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!