![]() | Only 14 pages are availabe for public view |
Abstract Classifying crisis events like hurricanes, earthquakes, football matches, etc. is a challenging issue in the field of Natural Language Processing (NLP) as it doesn’t have a rule. For example, if you search for “hurricane Michael”, the results may contain information about what is a hurricane or about a person called Michael. This illustrates the importance of the problem in Search Engine Optimization (SEO) and in Information Retrieval (IR). So, this research dives deep into this problem as a one-class classification problem, discusses the previous research in the topic with a comparison in accuracy illustrates the best scoring measure, and introduces a technique of modeling a deep autoencoder as a one-class classifier by adjusting its hyper parameters to detect the relevance of news to a certain rare event by thresholding the reconstruction error of the news document word-embedding after being fed to the model. |