Predicting text features of social temporal point process
Date
2021
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
This thesis studies the problem of predicting the text feature of temporal point processes. Temporal point process is common in our daily life - for example, the time sequence of multivariate discussions in online forums with features like text and user profile, and the coordinate sequence of human joint position in movement with features like relationship with other joints. Events in a temporal point process are usually marked by text features, thus predicting text feature is a required step for modeling temporal point process. Predicting text in such a process is challenging due to the long discourse structure of the text feature in temporal point process in social life. This research explores techniques combining grid discretization and temporal convolutional network, together with a new definition of Social Temporal Point Process to predict the text features. Specifically, these techniques are applied to a case study on Reddit online discussion forum.
Description
Keywords
Deep learning, Temporal point process, Text prediction