Skip to content

Latest commit

 

History

History
126 lines (106 loc) · 4.3 KB

summ_concept.md

File metadata and controls

126 lines (106 loc) · 4.3 KB

Research Concepts for Text Summarization

Generation Way

  • gen-ext: Extractive Summarization
  • gen-abs: Abstractive Summarization
  • gen-2stage Two-stage Summarization (compressive, hybrid)

Regressive Way

  • regr-auto: Autoregressive Decoder (Pointer network)
  • regr-nonauto: Non-autoregressive Decoder (Sequence labeling)

Task Settings

  • task-singleDoc: Single-document Summarization
  • task-multiDoc: Multi-document Summarization
  • task-senCompre: Sentence Compression
  • task-sci: Scientific Paper
  • task-radiologyReport: Radiology Reports
  • task-multimodal: Multi-modal Summarization
  • task-aspect: Aspect-based Summarization
  • task-opinion: Opinion Summarization
  • task-review: Review Summarization
  • task-meeting: Meeting-based Summarization
  • task-conversation: Consersation-based Summarization
  • task-medical: Medical text-related Summarization
  • task-covid: COVID-19 related Summarization
  • task-query: query-based Summarization
  • task-question: question-based Summarization
  • task-video: Video-based Summarization
  • task-code: Source Code Summarization
  • task-control: Controllable Summarization
  • task-event: Event-based Summarization
  • task-longtext: Summarization for Long Text
  • task-knowledge: Text Summarization with External Knowledge
  • task-highlight: Pick out important content and emphasize
  • task-analysis: Model Understanding or Interpretability
  • task-novel: Novel Chapter Generation
  • task-argument: Automatic Argument Summarization

Architecture (Mechanism)

  • arch-rnn: Recurrent Neural Networks (LSTM, GRU)
  • arch-cnn: Convolutional Neural Networks (CNN)
  • arch-transformer: Transformer
  • arch-graph: Graph Neural Networks or Statistic Graph Models
  • arch-gnn: Graph Neural Networks
  • arch-textrank: TextRank
  • arch-att: Attention Mechanism
  • arch-pointer: Pointer Layer
  • arch-coverage: Coverage Mechanism

Training

  • train-sup: Supervised Learning
  • train-unsup: Unsupervised Learning
  • train-weak: (implies train-sup): Weakly Supervised Learning
  • train-multitask: Multi-task Learning
  • train-multilingual: Multi-lingual Learning
  • train-multimodal: Multi-modal Learning
  • train-auxiliary: Joint Training
  • train-transfer: Cross-domain Learning, Transfer Learning, Domain Adaptation
  • train-active: Active Learning, Boostrapping
  • train-adver: Adversarial Learning
  • train-template: Template-based Summarization
  • train-augment: Data Augmentation
  • train-curriculum: Curriculum Learning
  • train-lowresource: Low-resource Summarization
  • train-retrieval: Retrieval-based Summarization
  • train-meta: Meta-learning

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

Relaxation/Training Methods for Non-differentiable Functions

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

Adversarial Methods

  • adv-gan: Generative Adversarial Networks
  • adv-feat: Adversarial Feature Learning
  • adv-examp: Adversarial Examples
  • adv-train: Adversarial Training

Latent Variable Models

  • latent-vae: Variational Auto-encoder
  • latent-topic: Topic Model

Dataset

  • data-new: Constructing a new dataset
  • data-annotation: Annotation Methodology

Evaluation

  • eval-human: Human Evaluation
  • eval-metric-rouge: ROUGE
  • 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