Knowledge management is a decision-making approach for facilitating the development and application of a variety types of knowledge assets. There are a number of key questions in the field, including “how can we gather knowledge assets?” and “How can we evaluate knowledge management initiatives planned for improving user experiences?”. The identification of the key knowledge asset value drivers and their relations allows stakeholders to define priorities.
It is also important to utilize existing knowledge effectively in the proper knowledge management of knowledge-based assets. Accordingly, building a knowledge-based system to solve new and similar problems is a research challenge that this thesis aims to address.
Although search engines and question-answering systems already serve as crucial tools for knowledge workers, understanding texts and using knowledge obtained from the texts for problem-solving is far from routine. Thus, this work addresses the problem of developing a collaborative knowledge-based system that can learn from user experience and knowledge assets.
The research described in this dissertation involved an investigation of the use of word association strength based on the statistical cohesions between words to build a semantic profile of a text. This approach in the retrieval of relevant information can provide reasoning information from a text in a manner that has traditionally required the use of human experts; this information then be reused in the analysis of new problems. In developing an artificial intelligence
(AI)-based problem-solving technique, this study investigated the use of case-based reasoning (CBR), a methodology in which data representing information on solved problems is stored for reuse in new problem-solving processes. The choice of past cases to be reused is based on similarity measures in the retrieval process as extracted from all stored cases in the case base. Each similarity measure characterizes a set of heuristics for approximating
the unidentified utility of a case, and the quality of similarity measures can be improved by integrating as much knowledge regarding the specific application domain as possible into them. Features relations from ontology and fuzzy logic can also be integrated into CBR similarity measures to handle the ambiguities and uncertainties that are characteristically present in knowledge-intensive processes.
The system developed in this research – DePicT CLASS – is based on the DePicT
concept, in which diseases are detected and predicted using image classification and text information from personal health records. DePicT CLASS was developed to serve as a collaborative case-based system to support caregivers and patients’ relatives by preparing relevant references and learning material to help them understand the patients’ medical issues. The main characteristics of DePicT and DePicT CLASS are demonstrated in this work using instances from two disease domains: dementia and melanoma.