FEW-SHOT TEXT CLASSIFICATION WITH DISTRIBUTIONAL SIGNATURES

任务

Given an N-way K-shot classification task

论文设计图像

数据集

• 20 Newsgroups is comprised of informal discourse from news discussion forums (Lang, 1995).Documents are organized under 20 topics.
• RCV1 is a collection of Reuters newswire articles from 1996 to 1997 (Lewis et al., 2004). These articles are written in formal speech and labeled with a set of topic codes. We consider 71 secondlevel topics as our total class set and discard articles that belong to more than one class.
• Reuters-21578 is a collection of shorter Reuters articles from 1987 (Lewis, 1997). We use the
  standard ApteMod version of the dataset. We discard articles with more than one label and consider
  31 classes that have at least 20 articles.
  *** Amazon product data** contains customer reviews from 24 product categories (He & McAuley,
  2016). Our goal is to classify reviews into their respective product categories. Since the original
  dataset is notoriously large (142.8 million reviews), we select a more tractable subset by sampling
  1000 reviews from each category.
• HuffPost headlines consists of news headlines published on HuffPost between 2012 and
  2018 (Misra, 2018). These headlines split among 41 classes. They are shorter and less grammatical
  than formal sentences.
• FewRel is a relation classification dataset developed for few-shot learning (Han et al., 2018). Each
  example is a single sentence, annotated with a head entity, a tail entity, and their relation. The goal is to predict the correct relation between the head and tail. The public dataset contains 80 relation
  types.

Baselines

Representations

• AVG
• IDF
• CNN

Learning algorithms

• the ridge regressor (RR)
• Prototypical network (PROTO)
• FT pre-trains a classifier
• MAML
• NN
• biLSTM

执行细节

• pre-trained fastText embeddings

• sentence-level datasets (HuffPost, FewRel)

• pre-trained BERT embeddings using HuggingFaces codebase

• a biLSTM with 50 hidden units and apply dropout of 0.1

• In the ridge regressor

• Adam

• early stopping

• 1000

• testing episodes

• the average accuracy over 5 different random seeds.

• 5-way 1-shot and 5-way 5-shot classification

操作

• Ablation study
• Contextualized representations
• PCA visualization of the input representation for the query set of a testing episode
• Attention generator
• Ridge regressor
• Visualizing the embedding space
• Cosine similarity to oracle word importance
• Visualizing the generated attention

创新

• the source pool

涉及领域

• 元学习
• 迁移学习