Time-aware Graph Neural Network for Entity Alignment between Temporal Knowledge Graphs

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 913–919 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 913 Learning Universal Authorship Representations Rafael A. Rivera Soto Olivia Miano Juanita Ordonez Barry Chen Lawrence Livermore National Laboratory {riverasoto1,miano3,ordonez2,chen52}@llnl.gov Aleem Khan Marcus Bishop Nicholas Andrews Johns Hopkins University {akhan141,marcus.bishop,noa}@jhu.edu Abstract Determining whether two documents were composed by the same author, also known as authorship verification, has traditionally been tackled using statistical methods. Recently, au- thorship representations learned using neural networks have been found to outperform al- ternatives, particularly in large-scale settings involving hundreds of thousands of authors. But do such representations learned in a par- ticular domain transfer to other domains? Or are these representations inherently entangled with domain-specific features? To study these questions, we conduct the first large-scale study of cross-domain transfer for authorship verification considering zero-shot transfers in- volving three disparate domains: Amazon re- views, fanfiction short stories, and Reddit com- ments. We find that although a surprising de- gree of transfer is possible between certain do- mains, it is not so successful between others. We examine properties of these domains that influence generalization and propose simple but effective methods to improve transfer. 1 Introduction Authorship verification offers a useful capability for a number of problems, such as plagiarism de- tection, moderation of user-generated content, his- torical authorship attribution, and forensic analysis. Authorship verification techniques have tradition- ally relied on modeling stylometric linguistic prop- erties, such as punctuation and whitespace usage and the frequencies of function words, parts-of- speech, and character n-grams (Stamatatos, 2009; Stolerman et al., 2014). More recently, end-to- end models built with convolutional neural net- works (Shrestha et al., 2017; Andrews and Bishop, 2019) and recurrent neural networks (Boenninghoff et al., 2019) have been proposed. However, neural methods introduce a tradeoff: although they obvi- ate the need of manual feature design by learning relevant features, these features are not explicitly identified. Not knowing how a system arrived at its authorship conclusions makes it difficult to assess how closely tied these features are to the domain from which the training data was drawn. In addition, neural methods are less applicable in low-resource domains due to their requirement of large training datasets. On the other hand, if authorship features could be learned in a domain- independent fashion, it would reduce the need for in-domain training sets by exploiting transfer be- tween domains: authorship representations could be learned from a large but out-of-domain corpus and subsequently deployed in a target domain. In prior work, Barlas and Stamatatos (2020) perform a study on cross-domain author verification in a closed world of 21 authors. In contrast, we consider an open-world setting with several orders of magni- tude more authors. With a view towards addressing the deeper question of whether neural approaches are capable of learning universal authorship rep- resentations that are effective in unseen domains, we conduct a systematic study involving three dis- parate domains: Amazon reviews, fanfiction, and Reddit comments. We propose the combination of an …

Content truncated to 500 words