To help individuals express themselves better, quotation recommendation is receiving growing attention. Nevertheless, most prior efforts focus on modeling quotations and queries separately and ignore the relationship between the quotations and the queries. In this work, we introduce a transformation matrix that directly maps the query representations to quotation representations. To better learn the mapping relationship, we employ a mapping loss that minimizes the distance of two semantic spaces (one for quotation and another for mapped-query). Furthermore, we explore using the words in history queries to interpret the figurative language of quotations, where quotation-aware attention is applied on top of history queries to highlight the indicator words. Experiments on two datasets in English and Chinese show that our model outperforms previous state-of-the-art models.
View on arXiv