The Integration of Self-extensible Rule-based Reasoning and Case-Based Reasoning and

Its Applications to Natural Language Processing and Computer-Assisted Instruction

@

Rey-Long Liu

@

ABSTRACT

Case-Based Reasoning (CBR) and Rule-Based Reasoning (RBR) are two major reasoning techniques in intelligent systems. They have shown many promising results in various applications. Recently many researchers have also noticed that the integration of the them may improve efficiency and accuracy in many domains. In this project, we will study the integration problem from the viewpoint of machine learning (ML), natural language processing (NLP), and computer-assisted instruction (CAI). We attempt to propose an integration model (integrating CBR and RBR) that may be applied to both NLP and CAI. The RBR component acts as the front-end processor for the CBR component. It is invoked first to retrieve suitable cases, and then the CBR component is triggered to derive the solution to the input problem. A machine learning mechanism is attached to the RBR component so that the RBR component may become self-extensible in order to improve the performance of case retrieval. In the application to NLP, we plan to construct a front-end language processor to understand natural language texts from different domains. In the application to CAI, we attempt to develop a test item manager to give suitable test items to each individual student (i.e. adaptive testing). In the two applications, the texts (students) of different domains (background) will need different kinds of processing (test items). Here domain and background correspond to the cases in CBR. As a text (student) is determined (by the RBR component) to be of a particular domain (background), CBR is invoked to process the text (give suitable test items to the student). As to the learning mechanism for building a self-extensible RBR component, we plan to use a learning technique called SEP that we had developed for building a self-extensible parser. SEP allows the computer to automatically derive missing rules based on top-down expectation and bottom-up partial result collection. Experiment will be done to explore the performance of the model.

Keywords: Case-Based Reasoning, Rule-Based Reasoning, Machine Learning, Natural Language Processing, Computer-Assisted Instruction

     Back to Research Project