Text Classification Using Lifelong Machine Learning

This paper proposes a novel lifelong machine learning model for text classification. The proposed model tries to solve problems as humans do i.e. it learns small and simple problems, retains the knowledge learnt from those problems, mines the useful information from the stored knowledge and reuses the extracted knowledge to learn future problems. The proposed approach adopts rule based learning classifier systems and a new encoding scheme is proposed to identify building units of knowledge which can be reused for future learning. The fitter building units from the learning system trained against small problems of text classification domain are extracted and utilized in high dimensional social media text classification problems to achieve scalable learning. The experimental results show that proposed continuous learning approach successfully solves complex high dimensional problems by reusing the previously learned fitter building blocks of knowledge.