Traffic Event Identification And Information Reconstruction Based On Social Media

In recent years,as China’s social and economic development continues,people’s travel needs are increasing,causing further deterioration of urban road traffic problems.Social media has become an important source of data for traffic research,as social media content is semantically rich,so social media data can be used to identify not only when and where traffic anomalies occur,but also the reasons behind traffic events,i.e.the specific types of traffic events.Therefore,the traffic information extracted through social media provides traffic public opinion for traffic management departments,and provides cross-validation and post-evaluation functions for traffic governance.Based on this data,the impact time and scope of different traffic incident categories on the road can be analyzed,so as to take corresponding measures.This paper therefore investigates methods for traffic event identification and traffic event information reconstruction from social media data.This paper proposes a method for acquiring traffic event data from social media.Taking Sina Weibo as an example,firstly,we obtain microblog data from the Sina Weibo website through crawler technology and analyse the characteristics of microblog data,pre-process the data according to the data characteristics,including de-duplication,de-noise and word separation.In addition,this paper proposes a microblog classification method based on text classification,use the BERT model to encode the text and classify the microblogs related to traffic events,and the F1 value on the dataset reaches 0.9120,proving the effectiveness of the model.On the basis of obtaining traffic event data,this paper proposes a traffic event identification method based on multi-label classification,classifying seven topics: traffic accidents,traffic congestion,road construction,traffic control,bad weather,smooth traffic flow and other events.One microblog can belong to more than one topic.This paper combines the BERT model with the Bi LSTM model,and combines the generated feature vectors with the keyword vectors extracted by TF-IDF to propose a BERT-Bi LSTM-Attention-Keyword model-based traffic event identification.The hammingloss reaches 0.02703 on the dataset,proving the effectiveness of the model.On the basis of traffic event identification,this paper proposes a method for reconstructing traffic event information based on named entity recognition,combining the BERT model with the Bi GRU model and CRF model to identify the time and location of traffic events.The F1 values of location and time recognition in the dataset reach 0.8267 and0.8892,proving the effectiveness of the model,and the time and location identified are filtered and standardised.Social media data may contain traffic events falsely described by people or provide incorrect information about traffic events,such as wrong time,location and event type.In this paper,based on the identified traffic event categories and the reconstructed traffic event times and locations,the geo-coding function of the Gaode API is used to convert the normalised location entities into latitude and longitude coordinates to match the corresponding road condition data,and the road condition data is used for cross-validation to determine whether the average travel speed of the incident section is abnormal,taking the Beijing Olympic Sports Centre area as an example to prove the authenticity of the Weibo data.