Generating User-Engaging News Headlines

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 3265–3280 July 9-14, 2023 ©2023 Association for Computational Linguistics Generating User-Engaging News Headlines Pengshan Cai,1∗ Kaiqiang Song,2 Sangwoo Cho,2 Hongwei Wang,2 Xiaoyang Wang,2 Hong Yu,1,3 Fei Liu,4 Dong Yu2 1University of Massachusetts, Amherst 2Tencent AI Lab, Bellevue, WA 3University of Massachusetts, Lowell {pengshancai,hongyu}@cs.umass.edu4Emory University fei.liu@emory.edu {riversong,swcho,hongweiw,shawnxywang,dyu}@global.tencent.com Abstract The potential choices for news article headlines are enormous, and finding the right balance be- tween conveying the essential message and cap- turing the reader’s attention is key to effective headlining. However, presenting the same news headline to all readers is a suboptimal strategy, because it does not take into account the differ- ent preferences and interests of diverse readers, who may be confused about why a particular article has been recommended to them and do not see a clear connection between their inter- ests and the recommended article. In this paper, we present a novel framework that addresses these challenges by incorporating user profil- ing to generate personalized headlines, and a combination of automated and human evalua- tion methods to determine user preference for personalized headlines. Our framework utilizes a learnable relevance function to assign person- alized signature phrases to users based on their reading histories, which are then used to person- alize headline generation. Through extensive evaluation, we demonstrate the effectiveness of our proposed framework in generating person- alized headlines that meet the needs of a diverse audience. Our framework has the potential to improve the efficacy of news recommendations and facilitate creation of personalized content.1 1 Introduction Personalized news recommendation systems, such as Google News and Yahoo News, help users dis- cover articles that align with their interests (Karimi et al., 2018). However, these systems often present the same article headline to all users, making it diffi- cult for them to understand the connection between their interests and the recommended article, poten- tially reducing the effectiveness of the recommen- dation system. To address this, we propose a new framework for generating personalized, engaging ∗*Work completed during an internship at Tencent AI Lab 1Our code can be accessed publicly at: https://github. com/pengshancai/user-engaging-headlines. headlines that clearly show the connection between a user’s reading history and a recommended article. Our framework has the potential to improve the ef- ficacy of personalized news recommendations, and recommendations for short videos, articles, recipes, etc. (Majumder et al., 2019; Kanouchi et al., 2020; Gosangi et al., 2021) Generating personalized headlines is a challeng- ing task due to the constraints of conciseness and the need to capture the reader’s attention. A person- alized headline should (a) effectively convey the main message of the article and (b) provide a clear link to the user’s reading history, using only about 10 words on average (Bernstein et al., 2020). There are two main challenges in this task. First, a head- line that entices users to click, but only presents limited information and fails to convey the essential story, becomes clickbait rather than a useful …

Content truncated to 500 words