27.06.2013 Views

Volume Two - Academic Conferences

Volume Two - Academic Conferences

Volume Two - Academic Conferences

SHOW MORE
SHOW LESS

Create successful ePaper yourself

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

Cristina Wanzeller and Orlando Belo<br />

sufficient processes’ descriptions, properly structured and applicable, according to the organization’s<br />

own needs. If we have such description, we will ask for an effective and flexible mechanism to help<br />

the matching of the current problem with the most promising mining processes. Knowledge<br />

management systems focus on knowledge and experience sharing. They enhance organization<br />

capabilities by externalizing knowledge of its employees and combine them in novel forms. It breaks<br />

up rigid processes and enables much richer interaction possibilities and creativity.<br />

3. Representing usage analysis experience<br />

Case representation is a key issue in CBR. The case based representation model must support a<br />

comprehensive description of WUM experiences, regarding the nature and requirements of such<br />

process development. A case is defined as a WUM process, corresponding to a successful usage<br />

analysis example. The case description is structured according to the domain problem and the<br />

applied solution (Table 1). We also defined a set of fundamental dimensions (D, T, P, A and K) to<br />

organize items of case’s description. The P dimension contains general attributes of WUM process<br />

description, involving problem (Pp) and solution aspects (Ps).<br />

Table 1: WUM process main item descriptors<br />

Case representation<br />

Problem Dataset and variables metadata (D)<br />

Problem type (T): activity, analysis goals and application areas<br />

Process date and requisites (Pp)<br />

Solution Mining Activities (A): DM tool and modelling and transformation stages<br />

Knowledge (K): Discoveries and general and specific facts<br />

Process general information (Ps)<br />

The case problem description comprises: metadata characterizing the data (D), at dataset and<br />

individual variables level; categorizations of the WUM problem type (T), mainly in terms of<br />

abstractions such as Web site main activity, analysis goals and application areas; and process data<br />

and evaluation criteria (Pp). The site activity may be “higher education” or something more detailed.<br />

An analysis goal reflects a kind of WUM problem, for example some general like “determine access<br />

order of pages and items”. The application areas are meant as the possible uses of the discoveries<br />

(e.g. “impact analysis”). To allow several levels of subdivision in more sub-areas, the application<br />

areas are organized in a dynamic hierarchy. This hierarchy provides a flexible way to define more<br />

subtle and refined application areas concerning the eLearning domain. The process requisites<br />

respects to criteria desired for the process results (e.g. level of precision).<br />

The applied solution is essentially defined by: a sequence of activities (A), including transformation<br />

and modelling stages, the involved data items and the model parameter settings; prior and derived<br />

knowledge (K), concerning to facts that affected the analysis, the extracted knowledge and the<br />

relations to such facts; and general information about the WUM process (Ps). The most important<br />

aspects of the solution description are the details which may be used to reproduce with fidelity the<br />

WUM process. Such details are usually the main ingredients of the DM activity success.<br />

We also have a context description item to organize cases in terms of a site or particular sections<br />

(e.g. curse). This item is a logic container for cases description features. The context may be<br />

associated with some aspects, namely dataset, activity, specific fact and general fact. Besides CBR<br />

principles (Kolodner 1993; Aamodt and Plaza 1994; Richter 1995; Mantaras et al. 2005), the<br />

Predictive Model Markup Language (PMML) standard was an important orientation. PMML is a XML<br />

based standard to define and share statistical and DM models across compliant applications The<br />

DM/KD/WUM domains are complex, the respective vocabulary is diverse, even at basic aspects,<br />

hindering the knowledge modelling. So, PMML provided accepted vocabulary to adopt and insights on<br />

how to represent DM processes.<br />

4. Acquiring and reusing experience<br />

A straight reuse of one solution is quite possible in the WUM domain, since recurrent problems are<br />

common. Still, becomes necessary to enable to help users on identifying the most plausible strategies<br />

to address the problem at hands. We also need simple ways for acquiring items of problem and<br />

solution description, in order to submit easily new problems and gather efficiently the relevant know<br />

how of successful processes. The CBR paradigm brings a key opportunity to our knowledge base,<br />

providing inherently a proper way for attending these demands. Coupling CBR intrinsic functionalities<br />

with components to attend additional needs, we devised an integrated system to fulfil our intents.<br />

856

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

Saved successfully!

Ooh no, something went wrong!