1.实现方法
本次的实现方法是按照论文中所提出的 Enhanced Sequential Inference Model (ESIM)增强型顺序推理模型进行。(下图左半部分)
2.数据集
使用的数据集是题目要求的数据集,匹配关系共有四种:矛盾(contradiction),中立(neutral),蕴含(entailment),未知(-)。其中未知的关系在数据集中出现较少。
输入文本:A few people in a restaurant setting, one of them is drinking orange juice.
输入假设:The people are sitting at desks in school.
匹配关系:contradiction(矛盾)
输入文本:The school is having a special event in order to show the american culture on how other cultures are dealt with in parties.
输入假设:A high school is hosting an event.
匹配关系: neutral
输入文本:A man, woman, and child enjoying themselves on a beach.
输入假设:A family of three is at the beach.
匹配关系:entailment
3.神经网络
4.运行与结果
初始化:随机初始化 learning_rate = 0.001 len_feature = 50 len_hidden = 50 iter_times = 50 batch_size = 1000