摘要
个性化评论摘要旨在针对一篇评论文本,面向不同用户产生反映他们不同偏好的摘要,具有较高的应用价值.现有工作存在新用户偏好无法增量学习、忽略用户属性关联对偏好的影响等不足.为了解决上述问题,本文提出了融合用户属性交互的个性化评论摘要生成算法.该算法自动学习用户各方面属性的嵌入表达,在此基础上利用自注意力机制捕捉所有属性之间的交互关联,从而使获得的用户偏好表达更加准确.得到的偏好表达被用于捕捉评论中用户可能感兴趣的信息,进而指导模型生成符合用户个性化偏好的摘要文本.实验结果表明,本文提出的算法在评价指标ROUGE上明显高于已有的先进算法.
Personalized review summarization generates summaries reflecting users'preferences with the same review content for dif-ferent users and is of high application value.Existing works suffer from several serious deficiencies,including incapability in incremen-tally learning preference for new users and overlooking interaction among attributes of a user for capturing his preference.To overcome these deficiencies,we propose an algorithm for generating personalized review summarization,taking interaction among user attributes into consideration for user modeling.The algorithm learns single atribute embedding automatically,and then employs a self-attention mechanism to capture interaction among all attributes of a user,thus resulting into his preference representation more accurately.Such preference representation is used to capture the information interested in by a user in a review,and accordingly guiding decoder of the algorithm to generate summarization satisfying the preference representation.The experimental results show that our proposed algo-rithm outperforms several state-of-the-art baselines in the ROUGE evaluation.
作者
陶嘉鸿
卢永美
何东
卜令梅
陈黎
于中华
TAO Jia-hong;LU Yong-mei;HE Dong;BU Ling-mei;CHEN Li;YU Zhong-hua(College of Computer Science,Sichuan University,Chengdu 610065,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2023年第8期1649-1655,共7页
Journal of Chinese Computer Systems
基金
国家重点研究项目(2020YFB0704502)资助。
关键词
评论摘要
用户属性交互
个性化摘要生成
自注意力机制
review summarization
user attribute interaction
personalized summarization
self-attention mechanism
