A continuous learning method for recognizing named entities by integrating domain contextual relevance measurement and Web farming mode of Web intelligence

Web farming can advance computational social science into a never-end learning process, in which social phenomena are dynamically and scientifically understood based on continuously produced, updated and expired data in the connected hyper world. Named entity recognition is a basic and core task of Web farming. However, the existing named entity recognition methods mainly depend on the complete, high-quality and well-labelled data sets and cannot meet the requirements of real-world applications. This paper proposes a continuous learning method for recognizing named entity by introducing the Web farming mode of Web Intelligence into the recognizing process. During the on-line stage, the domain contextual relevance of candidate entities is calculated by using the domain discrimination degree and the domain dependence function for recognizing the target entities. During the off-line stage, an active learning approach is designed to continuously improve the target corpus set by binding density-based clustering with semantic distance measurement. Experimental results show that the proposed method can effectively improve the accuracy of entity recognition and is more suitable for real-world applications.