The thesis contains all four chapters of my Ph.D. research on deep learning and text mining. The first chapter, "Temporal Topic Analysis with Endogenous and Exogenous Processes'', proposes a topic model which mines temporal economy-related documents with an exogenous economic indicator, and finds the relationship between document topics and the economic background. The second chapter, "Regularization for Unsupervised Deep Neural Nets'', discusses and compares different regularization methods for unsupervised deep neural nets, such as deep belief networks, and proposes a new approach to refine Dropout. The third chapter, "An Attention-Based Deep Net for Learning to Rank'', proposes a list-wise attention-based learning-to-rank mechanism for image and document retrieval. The fourth chapter, "Generative Adversarial Nets for Multiple Text Corpora'', proposes two applications of the generative adversarial net on text data with multiple corpora, i.e. refining word embeddings and generating multi-corpora document embeddings.

Last modified

• 02/15/2019

Creator

• Baiyang Wang

DOI

• https://doi.org/10.21985/N2KB4R

Subject

• Industrial Engineering and Management Sciences

Keyword

• deep learning
• text mining

Date created

• 2017-01-01

Resource type

• Dissertation

Rights statement

• In Copyright