Generating And Measuring Textual Context Based On Cognition

Cognitive linguistics indicates that context is just co-text, environment, profile, and mutual relations between speakers. Simply, textual context is just co-text, and the comprehension of sentences must depend on two aspects: local-context and whole-context. At present, many evidences show that human can master the means of a text with helps of semantic relations and context. However, available machine-based models of text analysis often neglect the contextual effect, which is caused by reasons that the text-contextual generation is related with a complicated psychological process and the text-contextual effect is subject to human's subjective sense. Available methods of textual context either depend on human's manual works, i.e. OWL, or generate unsatisfied textual context, i.e. VSM, and all methods lack the measure of the subjective cognitive sense to textual context in order to machine-based generate textual context and measure text-contextual effect on human's subjective cognitive sense, this thesis focuses on the generating and measuring textual context from cognitive science perspective.1. We discuss with the generations of local-text-context, whole-text-context, and domain-text-context in three levels respectively. 1) Local-text-context is generated based on Fuzzy Cognitive Map (FCM); 2) whole-text-context is based on the addition of FCMs; 3) domain-text-context is based on the reduction of plenty of FCMs. So the characteristics of text-contextual multi-levels and multi-granularities are represented completely.2. In the process of the generations of multi-level textual contexts, 1) our visualized methods based on the addition and the reduction of FCMs have the characteristics of composition and decomposition, which represents knowledge's evolved characteristics; 2) the local-text-contextual effect on whole-text-context is computed in order to acquire high-level semantic relations; 3) the validity of domain textual context is evaluated based on statistics in order to acquire high-level knowledge.3. The complexity and information of textual context are proposed based on the principle of cognitive economy. 1) According to the complexity of human's conceptual learning, the complexity of textual context is proposed in order to measure the text-contextual effect on human's subjective cognitive sense; 2) according to relevance theory, the information of textual context is proposed in order to provide machine-based personified text analysis from cognitive science perspective.4. For verifying the measure of textual context, we adopt linguistic and cognitive science viewpoints, 1) verifying experiments transform quality analysis into quantity one in order to provide the effective approach for machine-based text understanding; 2) based on the measure of textual context machine could mimic the human's reading-text experience and measure the capacity of text understanding so that our measuring method could provide a personified approach for machine-based text analysis.5. In order to optimize and reduce the computational complexities of text-contextual measure, our research is developed in the following three aspects: 1) optimizing sentence combination of maximal text-contextual information is solved by Genetic Algorithm; 2) an approximate computational approach is proposed to measure text-contextual information in order to reduce the computational complexities of text-contextual information from ( )O n 2 to O ( n ); 3) the relation between the complexity and the information is studied in order to select the optimum measuring method. By above studies the computational complexities of text-contextual complexity and information are decreased mostly, so these methods could be adopted in large-scale network environment.Textual knowledge can be acquired in multi-levels with the generation of textual context, and the effect of textual context can be measured from cognitive science perspective, so we provide a new approach for textual knowledge acquisition and the measure of the text-contextual effect; with the combination between traditional statistics-based semantic analysis and cognitive science overviews, we provide a theoretical foundation for machine-based text understanding in order to provide an effective approach for Internet advertisement delivering, question-answer system, merchandise recommendation in e-Business, knowledge service in e-Science, and automatic organization of textual segments, Web intelligent browsing.