• 文本摘要的一些研究概念


    文本摘要的一些研究概念

    主要翻译自github,特别适合新手查看。我也是新手,看下面的翻译就知道了。

    生成方式(Generation Way)

    • gen-ext:提取式摘要?第一个就遇到了困难。
    • gen-abs:抽象式摘要?
    • gen-2stage:两个混合,压缩和混合

    回归方式(Regressive Way)

    • regr-auto: Autoregressive Decoder (Pointer network) 自回归解码器,指针网络
    • regr-nonauto: Non-autoregressive Decoder (Sequence labeling) 非自回归解码器,序列标签

    任务设定(Task Settings)

    • task-singleDoc: Single-document Summarization 单文本摘要
    • task-multiDoc: Multi-document Summarization 多文本摘要
    • task-senCompre: Sentence Compression 句子压缩
    • task-sci: Scientific Paper 科技论文
    • task-radiologyReport: Radiology Reports 放射科报告??这玩意怎么跑到这的?
    • task-multimodal: Multi-modal Summarization 多模型摘要/多模型汇总
    • task-aspect: Aspect-based Summarization 基于方面的摘要???
    • task-opinion: Opinion Summarization 可选择摘要
    • task-review: Review Summarization 摘要综述???
    • task-meeting: Meeting-based Summarization 基于会议的摘要
    • task-conversation: Consersation-based Summarization 基于会话的摘要
    • task-medical: Medical text-related Summarization 关于医学文本摘要
    • task-covid: COVID-19 related Summarization 关于新冠病毒的摘要
    • task-query: query-based Summarization 基于查询的摘要
    • task-question: question-based Summarization 基于问答的摘要
    • task-video: Video-based Summarization 基于视频的摘要
    • task-code: Source Code Summarization 源码摘要
    • task-control: Controllable Summarization 可控制的摘要
    • task-event: Event-based Summarization 基于事件的摘要
    • task-longtext: Summarization for Long Text 长文本摘要
    • task-knowledge: Text Summarization with External Knowledge 可提取知识文本摘要
    • task-highlight: Pick out important content and emphasize 选出重要内容并强调
    • task-analysis: Model Understanding or Interpretability 模型的可理解性或可解释性
    • task-novel: Novel Chapter Generation 新章节的产生,novel在这里做形容词吧
    • task-argument: Automatic Argument Summarization 自动参数摘要

    架构-Architecture (Mechanism)

    • arch-rnn: Recurrent Neural Networks (LSTM, GRU) 递归神经网络
    • arch-cnn: Convolutional Neural Networks (CNN) 循环神经网络
    • arch-transformer: Transformer 翻译器
    • arch-graph: Graph Neural Networks or Statistic Graph Models 图神经网络或者统计图模型
    • arch-gnn: Graph Neural Networks 图神经网络
    • arch-textrank: TextRank 不翻译
    • arch-att: Attention Mechanism 注意力机制
    • arch-pointer: Pointer Layer 在这里应该不是指针层,不是输入层,不是输出层,肯定就是隐藏层了。
    • arch-coverage: Coverage Mechanism 覆盖机制???

    训练(Training)

    • train-sup: Supervised Learning 监督学习
    • train-unsup: Unsupervised Learning 非监督学习
    • train-weak: (implies train-sup): Weakly Supervised Learning 弱监督学习
    • train-multitask: Multi-task Learning 多任务学习
    • train-multilingual: Multi-lingual Learning 多语言学习
    • train-multimodal: Multi-modal Learning 多模型学习
    • train-auxiliary: Joint Training 连接学习
    • train-transfer: Cross-domain Learning, Transfer Learning, Domain Adaptation 跨领域学习,转移学习,领域适应
    • train-active: Active Learning, Boostrapping 主动学习,助人为乐?什么翻译
    • train-adver: Adversarial Learning 对抗学习
    • train-template: Template-based Summarization 基于模板的摘要
    • train-augment: Data Augmentation 数据参数
    • train-curriculum: Curriculum Learning 课程学习?
    • train-lowresource: Low-resource Summarization 低资源摘要
    • train-retrieval: Retrieval-based Summarization 基于检索的摘要
    • train-meta: Meta-learning 元学习

    预训练模型(Pre-trained Models)

    • pre-word2vec: word2vec
    • pre-glove: GLoVe
    • pre-bert: BERT
    • pre-elmo: ELMo
    • pre-hibert: HiBERT
    • pre-bart: BART
    • pre-pegasus: PEGASUS
    • pre-unilm: UNILM
    • pre-mass: MASS
    • pre-T5: Text-to-Text Transfer Transformer
    • pre-S2ORC: Pretrained model on semantic scholar open research corpus
    • pre-sciBERT: Scientific paper based pre-trained model
    • pre-SPECTER: Scientific Paper Embeddings using Citationinformed TransformERs

    不可微函数的松弛/训练方法(Relaxation/Training Methods for Non-differentiable Functions)

    这里应该是针对不可导不可微的一些处理方法。softmax曾经看到过。

    • nondif-straightthrough: Straight-through Estimator
    • nondif-gumbelsoftmax: Gumbel Softmax
    • nondif-minrisk: Minimum Risk Training
    • nondif-reinforce: REINFORCE

    对抗方法(Adversarial Methods)

    • adv-gan: Generative Adversarial Networks 生成对抗网络
    • adv-feat: Adversarial Feature Learning 对抗特征学习
    • adv-examp: Adversarial Examples 对抗样例
    • adv-train: Adversarial Training 对抗训练

    潜在变量模型(Latent Variable Models)

    • latent-vae: Variational Auto-encoder 可变自动编码器
    • latent-topic: Topic Model

    数据集(Dataset)

    • data-new: Constructing a new dataset 组件新的数据集
    • data-annotation: Annotation Methodology 注释方法

    评价(Evaluation)

    这里应该就是说你的实验出来结果,怎么评价你的文本摘要出来是符合标注还是不符合标注的,有机构有人去专门评价你的工作。

    • eval-human: Human Evaluation 人类评价
    • eval-metric-rouge: ROUGE 一个机构
    • eval-metric-bertscore: BERTScore
    • eval-aspect-coherence: Coherence
    • eval-aspect-redundancy: Redundancy of Summary
    • eval-aspect-factuality: Factuality
    • eval-aspect-abstractness: Abstractness
    • eval-referenceQuality: Reference Quality
    • eval-metric-learnable: Metrics are Learnable
    • eval-optimize-humanJudgement: Optimization towards human judgement
    • eval-reference-less: Reference-less Approach to Automatic Evaluation
    • eval-metric-unsupervised: Unsupervised Automatic Evaluation

    Survey

    • survey-2020: A survey paper in 2020
  • 相关阅读:
    面试题目以及注意事项
    jQuery Ajax 实例 ($.ajax、$.post、$.get)
    前端知识大全
    jquery实现2级联动
    [转]那些年我们一起清除过的浮动
    使用kubeadm在CentOS上搭建Kubernetes1.14.3集群
    企业优秀运维人员20道必会iptables面试题
    通过nginx日志利用shell统计日pv和uv
    php访问mysql接口pdo-mysql安装
    何查看已经安装的nginx、apache、mysql和php的编译参数
  • 原文地址:https://www.cnblogs.com/chenyameng/p/13280768.html
Copyright © 2020-2023  润新知