Clickbait Analysis And Detection In News Media

In recent years,the way people read news has been constantly changing.However,with the emergence of electronic news media platforms,the influence of personalized and fragmented self-media is gradually expanding.Some self-media hopes to obtain high clickthrough rates through exaggerated headlines,to increase the number of advertisements read on the webpage,and improve their own revenue.It will cause great problems to readers,society and social media.Therefore,it is necessary to detect such clickbait news sooner and block its propagation.In this thesis,clickbait in news media is studied.Existed researches lacked deep mining of the summary relationship and dependence between the headline and body.A multi-strategy is proposed to solve these problems.Starting from the aspect of both headline and body,the model analyzes the abnormal features in headline characteristics and takes the relationship of matching degree of them.The main work includes:First,this article extracts and models the text features of the headline.Clickbait has specific language features.For example,clickbait title is more likely to have exaggeration,question,etc.tone,or have a specific sentence structure.Based on the idea of headline features,this paper proposes a full convolutional neural network feature extraction module,which uses convolution kernels to extract local headline features,and then have multi-layer convolution overlay to obtain multi-granular information.Convert the language features of the headline into a dense text feature vector.Secondly,this article also considers the summarization of the headline to the body,and models the association between them.In view of the existing research,the information size of the headline and body is unbalanced,and the dependency relationship between them is not considered.This paper proposes a new headline-body matching model to solve these problems.In addition,the self-attention mechanism is used to generate a semantic expression with internal information in the sentence.At the same time,the cross-attention mechanism is used to make the title and body pay attention to each other to capture the potential dependence of the two sentences to form a semantic expression with text-related supplementary information.The above two expressions are used to calculate the similarity of headline and body,and then the matching vector of the them is obtained.Finally,combining the results of the above two models,based on the latest news dataset,the effectiveness of the algorithm is experimentally verified compared with related algorithms.Through the analysis of the experimental results,it can be concluded that the clickbait detection model proposed in this paper can detect clickbait news more accurately than other algorithms.At the same time,the headline-body matching model proposed in this paper can effectively evaluate the generalization of the headline.In addition,the headline-body matching model can also be used in other tasks to evaluate the similarity between long and short texts.This article's research on clickbait detection can not only help readers stay away from clickbait news intrusion,but also create a healthy and harmonious social media environment.It can also help social media platforms themselves to improve content quality and user stickiness.Finally,it can also create a better news reading environment for users.