零样本学习测评基准,中文版

Overview

ZeroCLUE

零样本学习测评基准,中文版

零样本学习是AI识别方法之一。 简单来说就是识别从未见过的数据类别,即训练的分类器不仅仅能够识别出训练集中已有的数据类别,
还可以对于来自未见过的类别的数据进行区分。 这是一个很有用的功能,使得计算机能够具有知识迁移的能力,并无需任何训练数据,
很符合现实生活中海量类别的存在形式。
GPT系列模型,除了展示了模型只提供提示或示例,而不进行任何参数更新情况下的学习能力。

任务描述和原始数据量统计

Corpus Train Dev Test Public Test Private Num Labels Unlabeled Task Metric Source
Single Sentence Tasks
EPRSTMT 32 32 610 753 2 19565 SntmntAnalysis Acc E-CommrceReview
CSLDCP 536 536 1784 2999 67 18111 LongTextClassify Acc AcademicCNKI
TNEWS 240 240 2010 1500 15 20000 ShortTextClassify Acc NewsTitle
IFLYTEK 928 690 1749 2279 119 7558 LongTextClassify Acc AppDesc
Sentence Pair Tasks
OCNLI 32 32 2520 3000 3 20000 NLI Acc 5Genres
BUSTM 32 32 1772 2000 2 4251 SemanticSmlarty Acc AIVirtualAssistant
Reading Comprhnsn Tasks
CHID 42 42 2002 2000 7 7585 MultipleChoice,idiom Acc Novel,EssayNews
CSL 32 32 2828 3000 2 19841 KeywordRecogntn Acc AcademicCNKI
CLUEWSC 32 32 976 290 2 0 CorefResolution Acc ChineseFictionBooks
EPRSTMT:电商评论情感分析;CSLDCP:科学文献学科分类;TNEWS:新闻分类;IFLYTEK:APP应用描述主题分类;
OCNLI: 自然语言推理;BUSTM: 对话短文本匹配;CHID:成语阅读理解;CSL:摘要判断关键词判别;CLUEWSC: 代词消歧
EPRSTMT,CSLDCP,BUSTM 为新任务;其他任务(TNEWS,CHID,IFLYTEK,OCNLI,CSL,CLUEWSC)来自于CLUE benchmark,部分数据集做了新的标注。

实验结果

实验设置:训练集和验证集使用32个样本,或采样16个,测试集正常规模。基础模型使用RoBERT12层chinese_roberta_wwm_ext(GPT系列除外)。

模型 score eprstmt bustm ocnli csldcp tnews wsc ifytek csl chid
Human 82.49 90.0N 88.0N 90.3N 68.0N 71.0N 98.0N 66.0N 84.0N 87.1N
FineTuning.B 39.35 61.9N 54.1N 33.6N 25.6N 40.5N 50.3N 22.6N 50.5N 15.0N
Zero-shot.G 43.36N 57.54N 50N 34.4N 26.23N 36.96N 50.31N 19.04N 50.14N 65.63N
Zero-shot.R 44.61N 85.2N 50.6N 40.3N 12.6N 25.3N 50.0N 27.7N 52.2N 57.6N
PET 57.36 87.2N 64.0 43.9N 56.9N 53.7N 59.2N 35.1N 55.0N 61.3N
PtuningB 51.81 88.5N 65.4 35.0N 44.4N 48.2N 51.0N 32.0N 50.0N 57.6N
PtuningGPT 46.44 75.65N 54.9N 35.75N 33.69N 45.3N 49.0N 24.0N 53.5N 13.7N
EFL 53.4 85.6N 67.6N 67.5N 46.7N 53.5N 54.2N 44.0N 61.6N 28.2N
LM-BFF 55.79 84.59 54.06 43.10 53.64 56.27 51.84 46.14 51.16 61.3
Human: 人类测评成绩;FineTuning: 直接下游任务微调;PET:Pattern Exploiting Training(完形填空形式); 
Ptuning: 自动构建模板; Zero-shot: 零样本学习;EFL:自然语言推理形式; ADAPET:PET改进版,带正确标签条件优化
FineTuning.B:FineTuningBert; FineTuningR:FineTuningRoberta; PtuningB:Ptuning_RoBERTa; PtuningGPT:Ptuning_GPT; 
Zero-shot.R,采用chinese_roberta_wwm_ext为基础模型的零样本学习;Zero-shot.G,GPT系列的零样本学习;N”,代表已更新;
报告的数字是每一个任务的公开测试集(test_public.json)上的实验效果;CLUE榜单已经可以提交;
由于CHID还在继续实验中,暂时未将CHID的分数纳入到最终的分数(Score)中。

使用的GPT模型: NEZHA-Gen

Datasets 数据集

需要预测的测试文件地址: ./datasets/

Baseline 基线模型

说明:./baselines/readme.md

测试流程

使用验证集(dev.json)做验证,在测试集(test.json)上做预测,并生成的预测文件。格式为:<task>_predict.json,如:bustm_predict.json

Submit Examples 提交样例

./resources/zeroclue_submit_examples

排行榜

排行榜,及开启测评

数据集介绍

分类任务 Single Sentence Tasks

1. EPRSTMT(EPR-sentiment) 电商产品评论情感分析数据集 E-commerce Product Review Dataset for Sentiment Analysis
     数据量:训练集(32),验证集(32),公开测试集(610),测试集(753),无标签语料(19565)
     
     例子:
     {"id": 23, "sentence": "外包装上有点磨损,试听后感觉不错", "label": "Positive"}
     每一条数据有三个属性,从前往后分别是 id,sentence,label。其中label标签,Positive 表示正向,Negative 表示负向。
2. CSLDCP 中文科学文献学科分类数据集
     中文科学文献学科分类数据集,包括67个类别的文献类别,这些类别来自于分别归属于13个大类,范围从社会科学到自然科学,文本为文献的中文摘要。
     数据量:训练集(536),验证集(536),公开测试集(1784),测试集(2999),无标签语料(67)
     
     例子:
     {"content": "通过几年的观察和实践,初步掌握了盆栽菊花的栽培技术及方法,并进行了总结,以满足人们对花卉消费的需求,提高观赏植物的商品价值,为企业化生产的盆菊提供技术指导。",
     "label": "园艺学", "id": 1770}
     {"content": "GPS卫星导航定位精度的高低很大程度上取决于站星距离(即伪距)的测量误差.载波相位平滑伪距在保证环路参数满足动态应力误差要求的基础上。。。本文详细论述了载波相位平滑伪距的原理和工程实现方法,并进行了仿真验证.", 
     "label": "航空宇航科学与技术", "id": 979}

     每一条数据有三个属性,从前往后分别是 id,sentence,label。其中label标签,Positive 表示正向,Negative 表示负向。
3.TNEWS 今日头条中文新闻(短文本)分类数据集 Toutiao Short Text Classificaiton for News
 该数据集来自今日头条的新闻版块,共提取了15个类别的新闻,包括旅游、教育、金融、军事等。
     
     例子:
     {"label": "102", "label_des": "news_entertainment", "sentence": "江疏影甜甜圈自拍,迷之角度竟这么好看,美吸引一切事物"}
     每一条数据有三个属性,从前往后分别是 分类ID,分类名称,新闻字符串(仅含标题)。
4.IFLYTEK 长文本分类数据集 Long Text classification
该数据集关于app应用描述的长文本标注数据,包含和日常生活相关的各类应用主题,共119个类别:"打车":0,"地图导航":1,"免费WIFI":2,"租车":3,….,"女性":115,"经营":116,"收款":117,"其他":118(分别用0-118表示)。
    
    例子:
    {"label": "110", "label_des": "社区超市", "sentence": "朴朴快送超市创立于2016年,专注于打造移动端30分钟即时配送一站式购物平台,商品品类包含水果、蔬菜、肉禽蛋奶、海鲜水产、粮油调味、酒水饮料、休闲食品、日用品、外卖等。朴朴公司希望能以全新的商业模式,更高效快捷的仓储配送模式,致力于成为更快、更好、更多、更省的在线零售平台,带给消费者更好的消费体验,同时推动中国食品安全进程,成为一家让社会尊敬的互联网公司。,朴朴一下,又好又快,1.配送时间提示更加清晰友好2.保障用户隐私的一些优化3.其他提高使用体验的调整4.修复了一些已知bug"}
    每一条数据有三个属性,从前往后分别是 类别ID,类别名称,文本内容。

Sentence Pair Tasks

5.OCNLI 中文原版自然语言推理数据集 Original Chinese Natural Language Inference
OCNLI,即原生中文自然语言推理数据集,是第一个非翻译的、使用原生汉语的大型中文自然语言推理数据集。
数据量:训练集(32),验证集(32),公开测试集(2520),测试集(3000),无标签语料(20000)

例子:
{
"level":"medium",
"sentence1":"身上裹一件工厂发的棉大衣,手插在袖筒里",
"sentence2":"身上至少一件衣服",
"label":"entailment","label0":"entailment","label1":"entailment","label2":"entailment","label3":"entailment","label4":"entailment",
"genre":"lit","prem_id":"lit_635","id":0
}
6.BUSTM 小布助手对话短文本匹配数据集 XiaoBu Dialogue Short Text Matching
对话短文本语义匹配数据集,源于小布助手。它是OPPO为品牌手机和IoT设备自研的语音助手,为用户提供便捷对话式服务。
意图识别是对话系统中的一个核心任务,而对话短文本语义匹配是意图识别的主流算法方案之一。要求根据短文本query-pair,预测它们是否属于同一语义。

数据量:训练集(32),验证集(32),公开测试集(1772),测试集(2000),无标签语料(4251)
例子:
{"id": 5, "sentence1": "女孩子到底是不是你", "sentence2": "你不是女孩子吗", "label": "1"}
{"id": 18, "sentence1": "小影,你说话慢了", "sentence2": "那你说慢一点", "label": "0"}

Reading Comprehension Tasks

7.ChID 成语阅读理解填空 Chinese IDiom Dataset for Cloze Test
以成语完形填空形式实现,文中多处成语被mask,候选项中包含了近义的成语。 https://arxiv.org/abs/1906.01265
数据量:训练集(42),验证集(42),公开测试集(2002),测试集(2000),无标签语料(7585)

例子:
{"id": 1421, "candidates": ["巧言令色", "措手不及", "风流人物", "八仙过海", "平铺直叙", "草木皆兵", "言行一致"],
"content": "当广州憾负北控,郭士强黯然退场那一刻,CBA季后赛悬念仿佛一下就消失了,可万万没想到,就在时隔1天后,北控外援约瑟夫-杨因个人裁决案(拖欠上一家经纪公司的费用),
导致被禁赛,打了马布里一个#idiom#,加上郭士强带领广州神奇逆转天津,让...", "answer": 1}
8.CSL 论文关键词识别 Keyword Recognition
中文科技文献数据集(CSL)取自中文论文摘要及其关键词,论文选自部分中文社会科学和自然科学核心期刊,任务目标是根据摘要判断关键词是否全部为真实关键词(真实为1,伪造为0)。
数据量:训练集(32),验证集(32),公开测试集(2828),测试集(3000),无标签语料(19841)

例子: 
{"id": 1, "abst": "为解决传统均匀FFT波束形成算法引起的3维声呐成像分辨率降低的问题,该文提出分区域FFT波束形成算法.远场条件下,
以保证成像分辨率为约束条件,以划分数量最少为目标,采用遗传算法作为优化手段将成像区域划分为多个区域.在每个区域内选取一个波束方向,
获得每一个接收阵元收到该方向回波时的解调输出,以此为原始数据在该区域内进行传统均匀FFT波束形成.对FFT计算过程进行优化,降低新算法的计算量,
使其满足3维成像声呐实时性的要求.仿真与实验结果表明,采用分区域FFT波束形成算法的成像分辨率较传统均匀FFT波束形成算法有显著提高,且满足实时性要求.",
 "keyword": ["水声学", "FFT", "波束形成", "3维成像声呐"], "label": "1"}
 
每一条数据有四个属性,从前往后分别是 数据ID,论文摘要,关键词,真假标签。
9.CLUEWSC WSC Winograd模式挑战中文版
Winograd Scheme Challenge(WSC)是一类代词消歧的任务,即判断句子中的代词指代的是哪个名词。题目以真假判别的方式出现,如:  
句子:这时候放在[床]上[枕头]旁边的[手机]响了,我感到奇怪,因为欠费已被停机两个月,现在[它]突然响了。需要判断“它”指代的是“床”、“枕头”,还是“手机”?
从中国现当代作家文学作品中抽取,再经语言专家人工挑选、标注。  

数据量:训练集(32),验证集(32),公开测试集(976),测试集(290),无标签语料(0)
例子:  
 {"target": 
     {"span2_index": 37, 
     "span1_index": 5, 
     "span1_text": "床", 
     "span2_text": "它"}, 
 "idx": 261, 
 "label": "false", 
 "text": "这时候放在床上枕头旁边的手机响了,我感到奇怪,因为欠费已被停机两个月,现在它突然响了。"}
 "true"表示代词确实是指代span1_text中的名词的,"false"代表不是。 

问题反馈

发送邮件:[email protected],或提交issue

引用

@article{CLUEbenchmark,
  title={CLUE: A Chinese Language Understanding Evaluation Benchmark},
  author={Liang Xu, Hai Hu, Xuanwei Zhang, Lu Li, Chenjie Cao, Yudong Li, Yechen Xu, Kai Sun, Dian Yu,Cong Yu,et al.},
  journal={In Proceedings of the 28th International Conference on Computational Linguistics, pages 4762–4772.},
  year={2020}
 }
Owner
CLUE benchmark
Organization of Language Understanding Evaluation benchmark for Chinese: tasks & datasets, baselines, pre-trained Chinese models, corpus and leaderboard
CLUE benchmark
Discord QR Scam Code Generator + Token grab mobile device.

A Python script that automatically generates a Nitro scam QR code and grabs the Discord token when scanned.

Visual 9 Nov 22, 2022
Detect textlines in document images

Textline Detection Detect textlines in document images Introduction This tool performs border, region and textline detection from document image data

QURATOR-SPK 70 Jun 30, 2022
Convert PDF/Image to TXT using EasyOcr - the best OCR engine available!

PDFImage2TXT - DOWNLOAD INSTALLER HERE What can you do with it? Convert scanned PDFs to TXT. Convert scanned Documents to TXT. No coding required!! In

Hans Alemão 2 Feb 22, 2022
This is an API written in python that uses FastAPI. It is a simple API that can detect discord tokens in Images.

Welcome This is an API written in python that uses FastAPI. It is a simple API that can detect discord tokens in Images. Installation There are curren

8 Jul 29, 2022
This is used to convert a string to an Image with Handwritten Characters.

Text-to-Handwriting-using-python This is used to convert a string to an Image with Handwritten Characters. text_to_handwriting(string: str, save_to: s

Akashdeep Mahata 3 Aug 15, 2022
SRA's seminar on Introduction to Computer Vision Fundamentals

Introduction to Computer Vision This repository includes basics to : Python Numpy: A python library Git Computer Vision. The aim of this repository is

Society of Robotics and Automation 147 Dec 04, 2022
Code release for our paper, "SimNet: Enabling Robust Unknown Object Manipulation from Pure Synthetic Data via Stereo"

SimNet: Enabling Robust Unknown Object Manipulation from Pure Synthetic Data via Stereo Thomas Kollar, Michael Laskey, Kevin Stone, Brijen Thananjeyan

68 Dec 14, 2022
A tool for extracting text from scanned documents (via OCR), with user-defined post-processing.

The project is based on older versions of tesseract and other tools, and is now superseded by another project which allows for more granular control o

Maxim 32 Jul 24, 2022
Basic functions manipulating images using the OpenCV library

OpenCV Basic functions manipulating images using the OpenCV library. Reading Ima

Shatha Siala 3 Feb 17, 2022
huoyijie 1.2k Dec 29, 2022
Code related to "Have Your Text and Use It Too! End-to-End Neural Data-to-Text Generation with Semantic Fidelity" paper

DataTuner You have just found the DataTuner. This repository provides tools for fine-tuning language models for a task. See LICENSE.txt for license de

81 Jan 01, 2023
Tesseract Open Source OCR Engine (main repository)

Tesseract OCR About This package contains an OCR engine - libtesseract and a command line program - tesseract. Tesseract 4 adds a new neural net (LSTM

48.4k Jan 09, 2023
Ackermann Line Follower Robot Simulation.

Ackermann Line Follower Robot This is a simulation of a line follower robot that works with steering control based on Stanley: The Robot That Won the

Lucas Mazzetto 2 Apr 16, 2022
A machine learning software for extracting information from scholarly documents

GROBID GROBID documentation Visit the GROBID documentation for more detailed information. Summary GROBID (or Grobid, but not GroBid nor GroBiD) means

Patrice Lopez 1.9k Jan 08, 2023
Responsive Doc. scanner using U^2-Net, Textcleaner and Tesseract

Responsive Doc. scanner using U^2-Net, Textcleaner and Tesseract Toolset U^2-Net is used for background removal Textcleaner is used for image cleaning

3 Jul 13, 2022
aardio的opencv库

opencv_aardio dll库下载地址:https://github.com/xuncv/opencv-plugin/releases import cv2 img = cv2.imread("./images/Lena.jpg",1) img = cv2.medianBlur(img,5)

71 Dec 31, 2022
make a better chinese character recognition OCR than tesseract

deep ocr See README_en.md for English installation documentation. 只在ubuntu下面测试通过,需要virtualenv安装,安装路径可自行调整: git clone https://github.com/JinpengLI/deep

Jinpeng 1.5k Dec 28, 2022
[EMNLP 2021] Improving and Simplifying Pattern Exploiting Training

ADAPET This repository contains the official code for the paper: "Improving and Simplifying Pattern Exploiting Training". The model improves and simpl

Rakesh R Menon 138 Dec 26, 2022
Program created with opencv that allows you to automatically count your repetitions on several fitness exercises.

Virtual partner of gym Description Program created with opencv that allows you to automatically count your repetitions on several fitness exercises li

1 Jan 04, 2022
Course material for the Multi-agents and computer graphics course

TC2008B Course material for the Multi-agents and computer graphics course. Setup instructions Strongly recommend using a custom conda environment. Ins

16 Dec 13, 2022