The Research Of Learning Algorithm In Reltion Extraction

As Information Era is coming, varied information emerges in a big amount. People hardly can find information they need rapidly. In order to save people's precious time, we urgently need an automated facility to find the needed information rapidly and accurately. Research on information extraction emerges under this background. Relation extraction is on task of information extraction.The task of relation extraction is judging that whether an entity pair in a sentence can form a predefined relation. Usually, the relation extraction problem was converted to a classification problem. As solving other classification problem, we firstly research on the feature extraction of relation extraction problem, viz. how to express the relation in the natural text and how to compute the similarity between relation instances. The feature extraction task was accomplished by analyzing the data. The result of feature extraction is estimated with two typical learning algorithms: SVM and NN.In the classifying process, features have different relevance. Through assigning every feature an appropriate relevant weight, a better classification performance can be reached. Feature weighting methods are researched in this paper. An algorithm called PSO-NN that based on NN algorithm is designed to learn every feature's relevant weight. Using the learned feature weights in the similarity function, the performance of relation extraction is improved with NN algorithm.Lastly, a research platform of relation extraction is designed and implemented. Researcher can focus on feature extraction of relation extraction problem and algorithms designing under this platform. The use of this platform saves the researcher's time and accelerates the research progress.