Collaborative annotation, search and categorisation
thesisposted on 09.09.2016, 10:59 by Yi Hong
The purpose of this research is to develop a collaborative framework for annotation, search and categorisation. The basis of this research is to define an ontology-based data model that allows users to create semantic tags, establish relationships among tags and provide contextual information by hierarchical concepts and properties structure derived from a lexical knowledge base. A computational model is introduced to record uncertainty, establish user credibility and compute the truthfulness or reliability of the statements, which can then be used for ranking search results. A method to transform a relational database to the ontology-based repository is developed to populate the proposed data model. The second stage of the research is to develop an expressive yet intuitive querying technique for searching semantically annotated data. There are many questions that arise when querying complex datasets. For example, how to help average non-tech users to write queries without excessive reliance on external technical support? How to utilise a rich knowledge base and how to enable members of a collaborative team to construct queries collectively, considering their opinions on the importance of searching criteria? Traditional keyword-based or form-based approaches fail to address these issues due to lack of expressive power or flexibility. A visual querying technique is presented for the collaborative team, based on graph pattern matching. This method allows members of a collaborative team to collectively construct complex queries in a more convenient manner. Then the possibility of applying various categorisation techniques to help sort annotated objects is investigated. A new workflow model is proposed that help a collaborative team build a universally-accepted categorisation system and develop a systematic way for team members to create a training data set, taking into account various criteria and degrees of uncertainty in human decisionmaking. Eventually, a modified Naive Bayes classifier was built for storing a large number of objects. In the end, in collaboration with members of an archaeological research team, a series of experiments was conducted to evaluate our methodologies.