Adversarial Learning for Distant Supervised Relation Extraction
Recently, many researchers have concentrated on using neural networks to learn features for Distant Supervised Relation Extraction (DSRE). These approaches generally
use a softmax classifier with cross-entropy loss, which inevitably brings the noise of artificial class NA into classification process. To address the shortcoming, the classifier with
ranking loss is employed to DSRE. Uniformly randomly selecting a relation or heuristically selecting the highest score among all incorrect relations are two common methods
for generating a negative class in the ranking loss function. However, the majority of the generated negative class can be easily discriminated from positive class and will contribute
little towards the training. Inspired by Generative Adversarial Networks (GANs), we use a neural network as the negative class generator to assist the training of our desired model,
which acts as the discriminator in GANs. Through the alternating optimization of generator and discriminator, the generator is learning to produce more and more indistinguishable
negative classes and the discriminator has to become better as well. This framework is independent of the concrete form of generator and discriminator. In this paper, we use a two
layers fully-connected neural network as the generator and the Piecewise Convolutional Neural Networks (PCNNs) as the discriminator. Experimental results show that our proposed GAN-based method is effective and performs better than state-of-the-art methods.
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
All authors should sign the copyright agreement by downloading the file at
Manuscript reference number:
Mailing address and Telephone/Fax numbers:
Each author must read and sign the following statements.
RETAINED RIGHTS: Except for copyright, other proprietary rights related to the work shall be retained by the authors. To reproduce any text, figures, tables, or illustrations from this work in future works of their own, the authors must obtain written permission from Tech Science Press. ORIGINALITY: Each author warrants that the submission to the work is original. Neither this work has been already published nor shall be submitted for publication elsewhere while under consideration for the publication by this journal.
AUTHORSHIP RESPONSIBILITY: Each author certifies that he or she has participated sufficiently in the preparation of the present work to take public responsibility for it. Each has reviewed the final version of the work, believes it is a valid work and approves it for publication. Moreover, they should produce data upon which the work is based if requested by the Editors of the journal.
DISCLAIMER: Each author warrants that this work contains no libelous or unlawful statements and does not infringe on the rights of others. If excerpts (text, figures, tables, or illustrations) from copyrighted works are included, a written release will be secured by the authors prior to submission, and credit to the original publication will be properly acknowledged. Each author warrants that he or she has obtained, prior to submission, written releases from patients whose names or photographs are submitted as part of the work.
TRANSFER OF COPYRIGHT
AUTHORS’ OWN WORK: In consideration of Tech Science Press’s publication of the work, the authors hereby transfer, assign, and otherwise convey all copyright ownership worldwide, in all languages, and in all forms of media now or hereafter known, including electronic media such as CD-ROM, Internet, and Intranet, to Tech Science Press. If Tech Science Press should decide for any reason not to publish an author’s submission to the work, Tech Science Press shall give prompt notice of its decision to the corresponding author, this agreement shall terminate, and neither the author nor Tech Science Press shall be under any further liability or obligation. The authors grant Tech Science Press the rights to use their names and biographical data (including professional affiliation) in the work and in its or the publication’s promotion.
FINANCIAL DISCLOSURE: Each author certifies that he or she has no commercial associations that might pose a conflict of interest in connection with the submitted article, except as disclosed in a separate attachment. All funding sources supporting the work and all institutional or corporate affiliations of the authors are acknowledged in a footnote in the work.
INSTITUTIONAL REVIEW BOARD/ANIMAL CARE COMMITTEE APPROVAL: Each author certifies that his or her institution has approved the protocol for any investigation involving humans or animals and that all experimentation was conducted in conformity with ethical and humane principles of research.
I certify that the above article has been written in the course of the author’s employment by the United States Government, so that it is not subject to U.S. copyright laws, or that it has been written in the course of the author’s employment by the United Kingdom Government (Crown Copyright).
Note to U.S. Government Employees:
If the above article was not prepared as part of the employee’s duties, it is not a U.S. Government work. If the above article was prepared jointly, and any co-author is not a U.S. Government employee, it is not a U.S. Government work.