Put Your Voice on Stage: Personalized Headline Generation for News Articles

Seldom prevailing news headline generation approaches take the interests of readers into account due to the elusive personalized style-related patterns and the insufficient hand-written headlines by individual readers. In this paper, we study the problem of personalized news headline generation from the point of view of the readers by utilizing distant supervision in their past click behaviors. We propose the first personalized headline generation approach for news articles named PNG (Personalized News headline Generator). First, user preference representations are learned through a knowledgeaware user encoder to comprehensively capture the genuine, sequential and flash interests of users, which are reflected in their historical clicked news. Then, a user-perturbed pointer-generator network is devised to accomplish the headline generation in which the learnt user representations implicitly affect the word generation. The proposed model is optimized by reinforcement learning solvers where indicators on linguistics and personalization aspects of the generated headline are regarded as rewards. Extensive experiments are conducted on a real-world dataset collected from [COMPANY NAME] 1 news. Both quantitative and qualitative results on our dataset validate the effectiveness of our approach.