Maximum Entropy-Based Named Entity Recognition Method for Multiple Social Networking Services

Abstract

Given a certain question, named entity recognition (NER) methods are regarded as an efficient strategy to extract correct answers. The goal of this work is to extend such conventional NER methods for analyzing a set of microtexts of which lengths are relatively short. These microtexts are streaming through several different social networking services, e.g., Twitter and FaceBook. To do so, we propose three heuristics for determining contextual associations between the microtexts, and discovering contextual clusters of microtexts, which can be expected to improve the performance of conventional NER tasks. Experimental results show the feasibility of the proposed mechanisms which extend the maximum entropy-based NER tasks for extracting relevant information in online social network applications.