通过深度相关性查询实现新闻事件挖掘  

News event mining through deep correlation query

作　　者：李辰 LI Chen(News and Publicity Center, Shenhua Railway Wagon Transportation Company Limited, Beijing 100012, China)

机构地区：[1]神华铁路货车运输有限责任公司新闻宣传中心,北京100012

出　　处：《信息技术》2019年第5期93-96,100,共5页Information Technology

摘　　要：文中提出了基于给定的查询词以新闻文档和具有额外情感极性评论信息为排名特征的新闻事件排名算法框架。首先,通过语义相似度交互模块将查询关键词、新闻文档和带有情感色彩的新闻评论转换为语义向量表示,并计算查询词和新闻文档相似度以及查询词和评论语句相似度。然后,基于特征提取查询关键词重要性特征、查询关键词频率特征和新闻事件相关性特征。最后,通过特征聚合模块将提取的特征与一些辅助相关特征合并,产生全局相关性分数,并基于所得出的全局相关性分数对新闻事件进行排名聚类。大型新闻数据集上的实验证明了该算法框架与常见排名算法相比具有明显的性能优势。This paper proposes a news event ranking algorithm framework based on a given query term,which is with a news document and additional emotional polarity comment information as the ranking feature. Firstly,the algorithm framework converts query keywords,news documents and news comments with emotional colors into semantic vector representations through semantic similarity interaction modules,then calculates the similarity between query words and news documents,also between query words and comment sentences. Secondly,the query word importance feature,the query word frequency feature and the news event correlation feature are extracted. Finally,the extracted features are merged with some auxiliary related features by the feature aggregation module,and global relevance scores are generated,and the news events are ranked and clustered based on the obtained global relevance scores. Experiments on large news data sets prove that the proposed algorithm framework has obvious performance advantages compared with common ranking algorithms.

关 键 词：新闻事件排名 用户相关查询 深度新闻事件排名 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]