融合图片信息的“标题党”新闻识别研究  

Research on News Recognition of “Clickbait” Fusing Image Information

作　　者：杨林[1] 丁继超 朱胜 王帅[1] 

机构地区：[1]中国矿业大学(北京)机电与信息工程学院,北京

出　　处：《图像与信号处理》2020年第3期137-145,共9页Journal of Image and Signal Processing

摘　　要：近年来,网络新闻逐渐取代传统的纸质新闻,成为人们日常获取新闻的主要方式。网络新闻因此而成为一个产业,产业的主体是由新闻制作者、用户、和点击率构成,新闻制作者通过用户的点击转换为点击率获取利益,由此导致了“标题党”的产生。“标题党”对用户、行业乃至社会都有巨大的危害。以往的“标题党”识别方法都忽略了网络新闻的主体是由图像和文本两个部分组成这一特点,只针对新闻的文本信息进行检测,而忽略了新闻的图片语义信息。目前网络上有大量利用引人眼球或者与文章毫不相关的图片吸引用户点击的新闻,本文针对新闻中的图像信息利用深度学习相关技术设计了图像语义描述与信息提取模型,使用这一模型,对新闻中的图片进行信息提取,对提取到的信息进行特征设计,将图片特征融合进“标题党”识别模型中,最后通过实验验证了使用图片信息识别“标题党”新闻的必要性和有效性。In recent years, online news has gradually replaced the traditional paper news, becoming the main way for people to get news daily. Internet news has thus become an industry. The main body of the industry is composed of news producers, users, and click-through rates. News producers convert users’ clicks into click-through rates to obtain benefits, which leads to the emergence of the “Clickbait”. The “Clickbait” is a huge hazard to users, industries and even society. In the past, the identification method of “Clickbait” ignored the feature that the main body of online news was composed of two parts: image and text. It only detected the text information of the news and ignored the image information of the news. At present, there is a lot of news on the Internet that uses eye-catching or irrelevant images to attract users to click. This article designs an image information extraction model using deep learning related technology for the image information in the news. Using this model, the news Extract the information of the image, design the features of the extracted information, integrate the image features into the “Clickbait” recognition model, and finally verify the necessity and effectiveness of using the image information to identify the “Clickbait” news.

关 键 词：图像描述 标题党 机器学习 文本相似度 

分 类 号：G21[文化科学—新闻学]