Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Ontology-based semantic reminiscence support system

Shi, Lei 2012. Ontology-based semantic reminiscence support system. PhD Thesis, Cardiff University.
Item availability restricted.

PDF - Accepted Post-Print Version
Download (7MB) | Preview
[img] PDF - Supplemental Material
Restricted to Repository staff only

Download (2MB)


This thesis addresses the needs of people who find reminiscence helpful in focusing on the development of a computerised reminiscence support system, which facilitates the access to and retrieval of stored memories used as the basis for positive interactions between elderly and young, and also between people with cognitive impairment and members of their family or caregivers. To model users’ background knowledge, this research defines a light weight useroriented ontology and its building principles. The ontology is flexible, and has simplified knowledge structure populated with semantically homogeneous ontology concepts. The user-oriented ontology is different from generic ontology models, as it does not rely on knowledge experts. Its structure enables users to browse, edit and create new entries on their own. To solve the semantic gap problem in personal information retrieval, this thesis proposes a semantic ontology-based feature matching method. It involves natural language processing and semantic feature extraction/selection using the user-oriented ontology. It comprises four stages: (i) user-oriented ontology building, (ii) semantic feature extraction for building vectors representing information objects, (iii) semantic feature selection using the user-oriented ontology, and (iv) measuring the similarity between the information objects. To facilitate personal information management and dynamic generation of content, the system uses ontologies and advanced algorithms for semantic feature matching. An algorithm named Onto-SVD is also proposed, which uses the user-oriented ontology to automatically detect the semantic relations within the stored memories. It combines semantic feature selection with matrix factorisation and k-means clustering to achieve topic identification based on semantic relations. The thesis further proposes an ontology-based personalised retrieval mechanism for the system. It aims to assist people to recall, browse and re-discover events from their lives by considering their profiles and background knowledge, and providing them v with customised retrieval results. Furthermore, a user profile space model is defined, and its construction method is also described. The model combines multiple useroriented ontologies and has a self-organised structure based on relevance feedback. The identification of person’s search intentions in this mechanism is on the conceptual level and involves the person’s background knowledge. Based on the identified search intentions, knowledge spanning trees are automatically generated from the ontologies or user profile spaces. The knowledge spanning trees are used to expand and reform queries, which enhance the queries’ semantic representations by applying domain knowledge. The crowdsourcing-based system evaluation measures users’ satisfaction on the generated content of Sem-LSB. It compares the advantage and disadvantage of three types of content presentations (i.e. unstructured, LSB-based and semantic/knowledgebased). Based on users’ feedback, the semantic/knowledge-based presentation is considered to have higher overall satisfaction and stronger reminiscing support effects than the others.

Item Type: Thesis (PhD)
Status: Unpublished
Schools: Engineering
Subjects: Q Science > QA Mathematics > QA76 Computer software
T Technology > T Technology (General)
Uncontrolled Keywords: user-oriented ontolog; knowledge modelling; semantic system; reminiscence support; personalised information retrieval.
Date of First Compliant Deposit: 30 March 2016
Last Modified: 19 Mar 2016 23:19

Actions (repository staff only)

Edit Item Edit Item


Downloads per month over past year

View more statistics