Facilitating knowledge interaction based on topic schema modeling

Abstract

The text is an important medium to acquire knowledge. The knowledge representation in a text is verbal and sequential, while the knowledge in long-term memory is the network of mental representations. When searching, comprehending, learning, and recalling the information in a text, a human utilizes prior knowledge to manipulate the information. So, the difference between two types of knowledge representation induces high cognitive load that contributes to mental exhaustion. Thus, it is essential to present the text in a similar form to the human knowledge structure. To this end, I propose a topic schema network to represent the information in a text, and I proposed the method to model the topic schema network as a computable form from a document. To model the topic schema network, I proposed to use burst durations of a concept as a source instead of the individual occurrence of a word, and elaborated the method for extracting topic schema networks based on the co-burst durations. When evaluating against the human-created concept map, the method based on co-burst durations effectively extracted association relationships. The results show that the burst durations of a word and co-burst durations can be a useful source for detecting topic schema model. Furthermore, I developed the system which visualizes the topic schema, and the experiment result shows that the topic schema helps users to seek information in the particular part of a long document by providing the efficient way to find the location of information and guess the gist of a specific part.