A novel framework for multiclass supervised classification of location-sensitive events

Abstract In the past couple of years, location-sensitive information retrieval has gained significant attention in terms of extracting and utilizing location information present in the unstructured text. It requires analysis of documents both geographically and thematically that makes it a challengi...
Ausführliche Beschreibung

Abstract In the past couple of years, location-sensitive information retrieval has gained significant attention in terms of extracting and utilizing location information present in the unstructured text. It requires analysis of documents both geographically and thematically that makes it a challenging task. The semantics of text needs to be associated with location features present in the text. Such information association is beneficial in conducting fine-grained analysis of events reported in the text, e.g., Tourist location recommendation, Disaster surveillance, Political activeness and Happiness index, etc. Recently, context-based vector space models have attained much importance in text mining as they intelligently preserve semantics of the text while representing text in vector space of desired dimension. In this paper, a framework for multiclass supervised classification of location-sensitive events, namely, LDoc2Vec is proposed that integrates context-based vector space models with geographic scope resolution of events reported in the text documents. Variants of the Doc2Vec model have been integrated with location features and their performance for multiclass supervised event classification is analysed. Experimental results with various machine learning classifiers indicate that the proposed framework outperforms baseline Doc2Vec models for multiclass classification of location-sensitive events as expressed by renowned performance measurement metrics viz. precision, recall and F1-score. Ausführliche Beschreibung