GLUE list Address: https://gluebenchmark.com/leaderboard/
If you use a word to describe the wide application of text classification task in NLP is, to some extent, probably best suited to this sentence:
NLP are all classified.
Generally speaking, NLP can be divided into natural language understanding (the NLU) and natural language generation (NLG). At NLU, we take the most popular GLUE (General Language Understanding Evaluation) chart, for example, on a collection of nine NLU task, namely,
- CoLA (at The Corpus of Linguistic Acceptability): New York University published data sets related to syntax, the main task for a given sentence, whether or not its syntax is correct, so CoLA belong to a single sentence of two text classification tasks;
- SST (at The Stanford Sentiment Treebank), Stanford University released a sentiment analysis data sets, mainly for movie reviews do sentiment classification, SST therefore belong to a single sentence of text classification task (which is a binary SST-2, SST-5 is five-emotional polarity distinguished more detailed SST-5);
- MRPC (in the Microsoft Research, of Paraphrase Corpus), published by Microsoft, judgment given two sentences have the same semantics, belong to the sentence of the text dichotomous task;
- B-the STS (the Semantic Similarity Textual Benchmark), mainly from the calendar SemEval a task (and the data set is also included in the SentEval ), particularly with a score of 1 to 5, characterized semantic two similar sentences sex is a regression in nature, but can still use methods of classification do, it can be classified as a sentence of five categories of text task;
- QQP (Quora Question Pairs), whether issued by Quora two sentences semantically consistent set of data, text belongs to a binary classification task sentence;
- MNLI (Multi-Genre Natural Language Inference), also released by the New York University, is a text contains a task at a given premise (Premise), need to determine hypothesis (Hypothesis) is established, which since MNLI main selling point is a collection of many different styles of text fields, it is also divided into two matched and mismatched versions of MNLI dataset, the former refers to the same training and test sets of data sources, while the latter refers to the inconsistent sources. This task belongs to a sentence of three text classification problem.
- QNLI (Question Inference Natural Language), which is predecessor SQuAD 1.0 dataset, given a question needs to judge whether a given text contains the answer to the question is correct. It belongs to a text sentence of two classification tasks;
- RTE (Recognizing Textual entailments), and the like MNLI, text also contains a task, except that the three classification MNLI, RTE determines whether only two sentences to infer or aligned, belonging to two sentences of text classification task;
- WNLI (Winograd Natural Language Inference), also contains a text task, but it seems GLUE on this data set still have some questions;