当前位置: X-MOL 学术Future Gener. Comput. Syst. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Entity-aware capsule network for multi-class classification of big data: A deep learning approach
Future Generation Computer Systems ( IF 6.2 ) Pub Date : 2020-11-20 , DOI: 10.1016/j.future.2020.11.012
Amit Kumar Jaiswal , Prayag Tiwari , Sahil Garg , M. Shamim Hossain

Named entity recognition (NER) is one of the most challenging natural language processing (NLP) tasks, as its performance is related to constantly evolving languages and dependency on expert (human) annotation. The diverse and dynamic content on the web significantly raises the need for a more generalized approach—one that is capable of correctly classifying terms in a corpus and feeding subsequent NLP tasks, such as machine translation, query expansion, and many other applications. Although extensively researched in recent times, the variety of public corpora available nowadays provides room for new and more accurate methods to tackle the NER problem. This paper presents a novel method that uses deep learning techniques based on the capsule network architecture for predicting entities in a corpus. This type of network groups neurons into so-called capsules to detect specific features of an object without reducing the original input unlike convolutional neural networks and their ‘max-pooling’ strategy. Our extensive evaluation on several benchmarked datasets demonstrates how competitive our method is in comparison with state-of-the-art techniques and how the usage of the proposed architecture may represent a significant benefit to further NLP tasks, especially in cases where experts are needed. Also, we explore NER using a theoretical framework that leverages big data for security. For the sake of reproducibility, we make the codebase open-source2 .



中文翻译:

实体感知胶囊网络用于大数据的多类分类:一种深度学习方法

命名实体识别(NER)是最具挑战性的自然语言处理(NLP)任务之一,因为其性能与语言的不断发展以及对专家(人类)注释的依赖有关。Web上多样化和动态的内容极大地提高了对一种更通用方法的需求,该方法能够正确地对语料库中的术语进行分类并提供后续的NLP任务,例如机器翻译,查询扩展和许多其他应用程序。尽管最近进行了广泛的研究,但如今可用的各种公共语料库为解决NER问题提供了新的更准确方法的空间。本文提出了一种新颖的方法,该方法使用基于胶囊网络架构的深度学习技术来预测语料库中的实体。与卷积神经网络及其“最大池”策略不同,这种类型的网络将神经元分组为所谓的胶囊,以检测物体的特定特征而不会减少原始输入。我们对多个基准数据集的广泛评估表明,与最先进的技术相比,我们的方法具有竞争优势,并且所提出的体系结构的使用可能对进一步的NLP任务产生重大好处,尤其是在需要专家的情况下。此外,我们使用理论框架探索NER,该理论框架利用大数据来保证安全性。为了重现性,我们将代码库设为开源 我们对多个基准数据集的广泛评估表明,与最先进的技术相比,我们的方法具有竞争优势,并且所提出的体系结构的使用可能对进一步的NLP任务产生重大好处,尤其是在需要专家的情况下。此外,我们使用理论框架探索NER,该理论框架利用大数据来保证安全性。为了重现性,我们将代码库设为开源 我们对多个基准数据集的广泛评估表明,与最先进的技术相比,我们的方法具有竞争优势,并且所提出的体系结构的使用可能对进一步的NLP任务产生重大好处,尤其是在需要专家的情况下。此外,我们使用理论框架探索NER,该框架利用大数据来保证安全性。为了重现性,我们将代码库设为开源2

更新日期:2020-11-26
down
wechat
bug