News Headlines Categorization Scheme for Unlabelled Data

Text categorization without training data is a difficult task and requires enough amount of hand labelled data to apply supervised methods, while manual labelling is a tedious job. In this paper a news categorization scheme is proposed to filter out and categorize news headlines related to Pakistan Stock Exchange (PSX) using negligible manual effort. By using domain knowledge, category names are selected manually then these category names are used as seed keyword to filter out news headlines. Natural Language Processing (NLP) based technique is used to extract context of seed keyword from initially filtered news headlines. These context terms are added in keyword list for string matching that further refines news filtration. Each news headline in a filtered news group is labelled and assigned a seed keyword term as a category label. Finally, a supervised classification technique is used to ensure the segregation of news categories as well as validates the performance of multiclass classification. Prepared dataset will be published in near future for potential uses explored by research community.