融合语境分析的时序推特摘要方法

发布时间：2018-05-03 11:54

  本文选题：时序推特摘要 + 时序特性　； 参考：《软件学报》2017年10期

【摘要】：任务中的一个重要分支,旨在从热点事件相关的海量推特流中总结出随时间演化的简要推特集,以帮助用户快速获取信息.推特作为当今最流行的社交媒体平台,其信息量爆发式的增长以及文本碎片的非结构性,使得单纯依赖文本内容的传统摘要方法不再适用.与此同时,社交媒体的新特性也为推特摘要带来了新的机遇.将推特流视作信号,剖析了其中的复杂噪声,提出融合推特流随时序变化的宏微观信号以及用户社交上下文语境信息的时序推特摘要新方法.首先,通过小波分析对推特流全局时序信息建模,实现某一关键词相关的热点子事件时间点检测;接着,融入推特流局部时序信息和用户社交信息建立推特的随机步图模型摘要框架,为每个热点子事件生成推特摘要.在算法评估过程中,对真实推特数据集进行了专家时间点和专家摘要的人工标注,实验结果表明了小波分析和融合了时序-社交上下文语境的图模型在时序推特摘要中的有效性.
[Abstract]:An important branch of the task is to sum up a brief set of tweets that evolve over time from a massive stream of tweets related to hot events to help users quickly obtain information. Twitter is the most popular social media platform nowadays. With the explosive growth of information and the non-structural structure of text fragments, the traditional summary method which relies solely on text content is no longer applicable. At the same time, the new features of social media have opened up new opportunities for Twitter feeds. The Twitter stream is regarded as a signal, the complex noise is analyzed, and a new method of temporal Twitter summary is proposed, which combines the macro and micro signals of the Twitter stream and the context information of the user's social context. Firstly, the global temporal information of Twitter stream is modeled by wavelet analysis to detect the time point of a key word related to a hot sub-event. Based on the local temporal information of Twitter stream and the social information of users, a summary framework of random step graph model of Twitter is established to generate a Twitter summary for each hot sub-event. In the process of algorithm evaluation, the real Twitter data sets are annotated manually with expert time points and expert abstracts. The experimental results show the validity of wavelet analysis and graph model which combines temporal and social context in temporal Twitter summary.
【作者单位】： 天津大学计算机科学与技术学院;天津市认知计算与应用重点实验室;北京大学信息科学技术学院;
【基金】：国家重点基础研究发展计划(973)(2013CB329301) 国家自然科学基金(61472277)~~
【分类号】：TP391.1
，

本文编号：1838383