We propose an intelligent document title classification agent based on a theory of information inference. The information is represented as vectorial spaces computed by a cognitively motivated model, namely Hyperspace Analogue to Language (HAL). A combination heuristic is used to combine a group of concepts into one single combination vector. Information inference can be performed on the HAL spaces via computing information flow between vectors or combination vectors. Based on this theory, a document title is treated as a combination vector by applying the combination heuristic to all the nonstop terms in the title. Two methodologies for learning and assigning categories to document titles are addressed. Experimental results on Reuters-21578 corpus show that our framework is promising and its performance achieves 71% of the upper bound (which is approximated by using whole documents).
展开▼
