External Knowledge Enhanced Contrastive Learning for Chinese Short Text Matching
Ruiqiang Liu, Qiqiang Zhong, Mengmeng Cui, Hanjie Mai, Qiang Zhang, Shaohua Xu, Xiangzheng Liu, Yanlong Du · 2024
In recent years, Chinese Short Text Matching task has been widely applied in the fields of advertising search and recommendation. However, the short length of texts often led to the lack of semantic information and word ambiguity, which brings the challenge for Text Matching. Previous works have introduced complement sentences or knowledge bases to provide additional feature information. Most existing approaches suffer from various limitations, such as insufficient interaction between the original sentence and the complement sentence, and the lack of handling of the noise issue that may arise from the introduction of external knowledge. In this paper, we propose a knowledge enhanced contrastive learning model that effectively combines contrastive learning and external knowledge, in which we employ contrastive learning based on the complementary sentences for the interaction between the short texts and construct a knowledge graph to enhance the representation learning of short texts. Furthermore, we extensively evaluate our approach on BQ and LCQMC datasets, and the results demonstrate the effectiveness of our method.