基于眼动跟踪的个性化文本摘要生成模型  

A Model to Personalized Text Summarization Generation Based on Eye Tracking

作　　者：揭志强 叶阳[1] 程时伟[1] Jie Zhiqiang;Ye Yang;Cheng Shiwei(School of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310012)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310012

出　　处：《计算机辅助设计与图形学学报》2023年第10期1620-1628,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(62172368,61772468).

摘　　要：为了满足不同用户对生成摘要的个性化需求,提出一种基于眼动跟踪的个性化文本摘要生成网络模型.首先使用阅读时的注视和扫视等眼动跟踪数据提取眼动关键信息,并使用长短期记忆模型对眼动关键信息进行编码;然后将眼动关键信息隐藏层状态融合到注意力机制中,结合内部注意力模型和指针生成器网络生成个性化的文本摘要;最后设计开发了相关的文本摘要生成系统.使用包含用户眼动数据的中文新闻数据集ADEGBTS进行实验的结果表明,与现有的文本摘要模型相比,所提模型在ROUGE指标评估上的得分更高,达到48.68%;用户测试结果表明,所提系统可以高效地生成个性化摘要.In order to meet the personalized demands of various users for summarization generation,a gaze-based key information guide network(GKGN)is proposed.Firstly,it uses eye-movement data,such as fixation and saccade in the reading process,to extract key eye-movement information.The key eye-movement information is encoded by long-short term memory(LSTM)networks.Secondly,the hidden layer state of the key eye-movement information is fused with mechanisms of attention.Apart from that,intra-attention model and pointer generator network(PGN)are fused to generate the personalized text summarization.The ADEGBTS dataset,which contains users’eye-movement data on reading Chinese news,is used for evaluating the GKGN model.The results show that,compared with existing text summarization models,the GKGN model scored 48.68%higher on the ROUGE.In addition,a text summarization system is designed and developed,and the user test results show that the system can efficiently generate the personalized summarization.

关 键 词：眼动跟踪 文本摘要 人机交互 自然语言处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]