The Identification Of Online Rumors Based On Deep Learning

With the development of mobile internet and the rise of social platforms,the speed and scope of information dissemination have undergone unprecedented development.Especially in the field of health information,a large amount of health related information is generated and disseminated on social platforms.At the same time,the health sector has also become a breeding ground for rumors.Compared to other themed rumors,health rumors are more sensitive and dangerous because they involve people’s lives and health.During the spread of rumors,the audience is extremely vulnerable,and even a slight carelessness may trigger a public safety crisis.At the same time,with the popularity of fragmentation reading,the spread form of health rumors also began to change,often using short,certain conclusions,so that people can quickly remember conclusions in fragmentation reading,making health rumors more hidden and influential.Therefore,early detection and identification of rumors is an important way to reduce the harm of rumors.Currently,major online platforms mainly rely on netizens to report and manually screen for judgment.However,this method not only increases the workload of the platform,but also fails to identify the source of rumors,thereby reducing the impact of rumor propagation.To address this issue,this article proposes a detection model for short text health rumors.By mining the content features of short text rumors,including word features,emotional orientation features,topic type features,emotional differences features,and object emotional features,we can break away from the dependence of traditional network rumor recognition on user and propagation features,and achieve early recognition of rumors.Experiments have shown that this model has high accuracy and recall in identifying health rumors,effectively alleviating the harm caused by the spread of health rumors on social platforms.It mainly includes the following two parts:(1)Analysis of the Content Characteristics of Short Text Health Rumors.Based on the summary of existing research on the content features of online rumors and health rumors,this paper combines natural language processing technologies such as part of speech tagging,emotional orientation analysis,etc.,and extracts five content features from the short text health rumors,namely,word features,emotional orientation features,topic type features,emotional difference features,and object emotional features,Obtained a large number of popular science texts and constructed an object emotion knowledge base.In order to verify the effectiveness of the content features proposed in this chapter in rumor recognition,crawler technology was used to crawl a large amount of short text health rumor data from We Chat mini programs,and support vector machine classification algorithms were used to set up different experimental templates for classification training of each template feature set.The experimental results show that the newly proposed topic type features,emotional differences features,and object emotional features have good effects on rumor recognition.(2)Constructing a rumor recognition model based on deep learning.This article improves the text classification models Text CNN,Text RNN,and Text RCNN based on the characteristics of health rumor short texts.The extracted content features are fused into a deep learning network model for rumor recognition,and different reference groups are set up for experiments by fusing different content features.The experimental results show that the improved model that integrates content features performs better than the original text classification model and feature based classification methods,with the best recognition model being the Text CNN model that integrates topic type features,emotional differences features,and object emotional features.This article also conducted comparative experiments on the size of the object knowledge base for object emotional features,and found that the size of the knowledge base has a significant impact on the model’s performance,providing a reference for later model optimization.The model proposed in this article focuses on the recognition of short text health rumors,which can achieve early detection of rumors and identify them at the source,avoiding the risks brought by rumor propagation and effectively helping network platforms automatically identify rumors.