Register Linear Based Model for Question Classification Using Costa Level Questions

摘要

Question classification module of a Question Answering System plays a very important role in identifying and providing results according to the user expectations. Different methods are involved in the classification that can be applied to all kinds of domains like machine learning and lexical database. Identifying the relevant approach for question classification for a specific domain is one of the foremost tasks. A study on different levels of questions including Blooms taxonomy and Costa taxonomy made the researchers to focus more on different categories of questions. To overcome these issues, we employ a question classifier using Register Linear (RL) models for a specific domain. The Register Linear (RL) Classification Model classifies the complex questions in a linear manner where each input is assigned to only one class. The RL classification model identifies the role of semantics provided in the input space which is divided into decision regions with the decision surfaces to be of linear functions of input x (sentence) for different set of classes. Initially, the Register Linear model identifies the role of semantics in a sentence, and with these roles being identified, statistical relations between the concepts in the sentence are derived that produce a probability distribution over different set of classes. With these classifications, the exact answer type is identified. The model proposed gives better results in terms of execution time (time taken to categorize the queries), classification accuracy and result analyzing efficiency.