arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2002.00745cs.CLcs.LGstat.ML
收藏

基于递归最优传输的结构感知句子相似度

Structural-Aware Sentence Similarity with Recursive Optimal Transport

Research Institute of Information Technology(信息技术研究所);Beijing National Research Center for Information Science and Technology (BNRist)(北京信息科学与技术国家研究中心);Tsinghua University(清华大学)

Zihao Wang, Yong Zhang, Hao Wu

摘要:

句子相似度度量是自然语言处理中的一个经典课题。即使深度学习模型在许多其他任务中取得了成功,轻量级相似性仍然具有特别的实际意义。一些具有更多理论见解的轻量级相似性已被证明比监督式深度学习方法更强大。然而,成功的轻量级模型,如单词移动器的距离[Kusner等人,2015]或平滑逆频率[Arora等人,2017]未能检测到句子结构的差异,即单词的顺序。为了解决这个问题,我们提出了递归最优运输(ROT)框架,将结构信息与经典的OT。此外,我们进一步开发递归最佳相似度(ROTS)的句子与有价值的语义见解之间的连接词向量的加权平均值和最佳运输的余弦相似度。ROTS是结构感知的,并且与最优传输相比具有低时间复杂度。我们在20个句子纹理相似度(STS)数据集上的实验表明,ROTS比所有弱监督方法都具有明显的优势。详细的消融研究证明了ROT的有效性和语义见解。

英文摘要:

Measuring sentence similarity is a classic topic in natural language processing. Light-weighted similarities are still of particular practical significance even when deep learning models have succeeded in many other tasks. Some light-weighted similarities with more theoretical insights have been demonstrated to be even stronger than supervised deep learning approaches. However, the successful light-weighted models such as Word Mover's Distance [Kusner et al., 2015] or Smooth Inverse Frequency [Arora et al., 2017] failed to detect the difference from the structure of sentences, i.e. order of words. To address this issue, we present Recursive Optimal Transport (ROT) framework to incorporate the structural information with the classic OT. Moreover, we further develop Recursive Optimal Similarity (ROTS) for sentences with the valuable semantic insights from the connections between cosine similarity of weighted average of word vectors and optimal transport. ROTS is structural-aware and with low time complexity compared to optimal transport. Our experiments over 20 sentence textural similarity (STS) datasets show the clear advantage of ROTS over all weakly supervised approaches. Detailed ablation study demonstrate the effectiveness of ROT and the semantic insights.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑