Predicting user’s action on emails: improvement with ham rules and real-world dataset
Ha Nguyen Thanh, Quan Dang Dinh, Quang Anh Tran · 2018
Email is an important tool for Internet communication. One of the many problems that email users are facing is email overload. There are approaches to solve this problem, namely spam filtering, email prioritization and user action prediction. Amongst these, action prediction is a smart one but studies have shown little results. In this paper, the authors propose a new method which extends the result of a previous user action prediction study [4]. In this method, ham rules are used to improve prediction accuracy and the dataset is labelled automatically by extracting user actions from email properties. With this approach, it is feasible to collect a large dataset for the study because users no longer have to manually label their emails. Experiment result shows significantly lower false positive rates compared to the previous study.