项目作者: BshoterJ

项目描述 :
This repo contains some experiments of text matching on Chinese dataset LCQMC
高级语言: Python
项目地址: git://github.com/BshoterJ/Text-Matching.git
创建时间: 2019-11-09T11:45:23Z
项目社区:https://github.com/BshoterJ/Text-Matching

开源协议:

下载


Text matching models on LCQMC datasets

Requrement

  • python 3.6
  • tensorflow-gpu 1.12
  • gensim 3.8.1
  • jieba 0.39
  • numpy 1.16
  • pandas 0.23

To Do List

Single Model

  • DSSM
  • ABCNN
  • ESIM
  • BIMPM
  • DIIN
  • DRCN
  • RE2

    Classic Algorithm

  • TFIDF
  • BM25
  • VSM

    LM Fintune

  • ELMo
  • BERT
  • ALBERT

Result

Model accuracy loss word/char
DSSM 63.336% 0.64119714 char
ABCNN 79.928% 0.6421789 char
ESIM 81.8% 0.48200694 char
BIMPM
DIIN 84.472% 0.34605518 char + dynamic word
DRCN
RE2