Topic Based Machine Learning Summarizer

摘要

The study develops a summarizer relying on domain-based corpus to enhance classification. Three contributions are brought forth: (1) summarization is extended to a context of spotting a text on food insecurity; (2) use of suitable corpus to promote topical summarization; and (3) shows that extractive summarization considered less effective can be dependable in some situations. These study aspirations were experimented on a task of a human identifying whether an article is on food insecurity or not. Findings showed that a human could still label an article correctly at 98% of Area Under the Curve (AUC) after applying the summarization compared to random classifier performance of 74%. By summarizing the articles, it becomes easy to identify and track conversations on food insecurity in the articles. Trends of these conversations can inform food relief agencies on appropriate actions as dictated by prevailing circumstances. Future work considers blending this summarization with Machine Learning Classification to enhance performance.