Climate Event Detection Algorithm Based On Climate Category Word Embedding

By extracting information from text,text representing and cluster,event detection techniques can summarize events effectively from abudant and complex text information,and demonstrate corresponding details to people efficiently and accurately and provide effective information of decision support to goverments and enterprises.By studying the characteristics of climate text,climate event detection can detect climate events timely and accurately,and have practical application value in fields such as agricultural production and activity organization.Given that ordinary event detection algorithms is inadequate mining of climate semantics and domain knowledge,the results of climate event detection are not satisfactory.This paper proposes a climate event detection algorithm based on climate category word embedding(CEDCWE),which use climate knowledge to mining semantic information and construct climate text representing model.The algorithm contains four mainly parts,namely,climate category word embedding model(Cc WE),words topic cluster model,climate document reprensentation model and climate event detection model.The main work of each part is as follows:1.In the Cc WE model,the paper proposes a word embeddiing model using climate catrgories as the objective,and enhances semantic information of word vector.2.In the words topic cluster model,by cluster word vectors from the Cc WE,the model acquire the result of words' topic with better semantic similarity.The results of this model can be used for climate text representing,effectively avoiding the problem of high dimension and sparse feature of traditional methods.3.In the climate document reprensentation model,the model combines the result of words' topic and three important characteristics of climate events,namely,climate category,occurrence time,and occurrence position as the climate document reprensentation.4.In the climate event detection model,the model utilizes Single-Pass algorithm which is a sort of incremental clusters to detect climate events.Experimental results show that the CEDCWE is effective in climate document representation and outperforms typical methods.In additional,the idea of CEDCWE can implement in other domains' event detection,the CEDCWE has expandability.