摘要
语义文本相似度计算在自然语言处理实际应用中有着重要的作用,但由于当前可用的中文语义文本相似度数据集缺乏,因此目前的中文语义文本相似度研究还存在很多局限性。针对这一问题,本文引入EDA和反向翻译方法共计五种数据增强策略对中文语义相似度数据集进行数据增强,并基于ABCNN和ESIM模型进行实验,实验数据显示:五种数据增强策略均有助于模型性能的提升且在数据集数量越少时效果越明显,其中在最小比例的训练集上使用同义词替换数据增强策略达到了3.6%的准确率提升。
Semantic text similarity calculation plays an important role in the practical application of natural language processing.However,due to the lack of available Chinese semantic text similarity data sets,the current research on Chinese semantic text similarity still has many limitations.To solve this problem,this paper introduces five kinds of data enhancement strategies,EDA and reverse translation,to enhance the Chinese semantic similarity data set,and carries out experiments based on ABCNN and ESIM model.The experimental data show that the five kinds of data enhancement strategies are helpful to improve the performance of the model,and the less the number of data sets,the more obvious the effect is,in the minimum proportion of the training set,synonyms are used to replace the data,and the accuracy of the enhancement strategy is improved by 3.6%.
作者
张豪
张华熊
ZHANG Hao;ZHANG Huaxiong(Zhejiang Sci-Tech University,Hangzhou Zhejiang 310018)
出处
《软件》
2021年第5期125-127,共3页
Software
关键词
中文语义相似度
数据增强
EDA
反向翻译
chinese semantic similarity
data enhancement
EDA
back translation
