论文地址:Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
数据集:LCQMC
句子对的形式 text_a text_b label 喜欢打篮球的男生喜欢什么样的女生 爱打篮球的男生喜欢什么样的女生 1 我手机丢了,我想换个手机 我想买个新手机,求推荐 1 大家觉得她好看吗 大家觉得跑男好看吗? 0
模型: bert+2层隐藏层+全连接层
做句子的语义分类
模型model/sentence_bert.py——更新了模型输出向量的均值方法pooling
similarity_valdation.py——相似度验证
query_topn_search.py——相似度搜索
python train_sentence_bert.py
或者
bash run_train_sentence_bert.sh
最新数据采用业务数据,效果记录在
train_model_record.txt中