Research On Highway Traffic Incidents Based On Microblog Data

With the rise of online social media,social media data of major platforms at home and abroad have become an important source of research data.How to effectively classify traffic events and how to extract traffic related information from them has become an urgent problem to be solved.This paper takes microblog data of traffic events as the research object,studies the characteristics of its webpage,uses crawler technology to obtain microblog traffic text;according to the text characteristics and research fields,creates an exclusive corpus,and carries out data preprocessing such as text denoising,Chinese word segmentation,part of speech tagging,etc.The topic of microblog traffic text is mined by LDA topic modeling algorithm.Based on the confusion degree,the text is divided into topics,and the topic vocabulary matrix is obtained.The content of the divided topics is analyzed and summarized into reasonable classification categories as the basis of classification.By evaluating the classification results of machine learning classification algorithm and deep learning classification algorithm,the convolution neural network(CNN)is selected to build an automatic text classifier,which realizes the automatic classification of microblog traffic text.On this basis,a named entity recognition expression framework is constructed to recognize the time entity and place entity in the text based on conditional random field;the time dimension of traffic events is integrated into the traditional spatial kernel density,and the spatiotemporal kernel density model is constructed to identify the event prone points.At the same time,Arc Scene is used for three-dimensional visualization display to more intuitively analyze the characteristics of highway traffic events Spatial and temporal agglomeration characteristics.Based on Arc GIS platform,it realizes the construction of spatiotemporal cube model of expressway traffic events and the storage of spatiotemporal data;based on Getis Ord Gi * hot spot model is used to detect and identify expressway traffic incident points with statistical significance in time and space;based on Mann Kendall time trend test model,the long-term measurement of temporal and spatial change trend is carried out.For further analysis,the change patterns of spatiotemporal cold and hot spots are divided into 17 types to extract the spatiotemporal law of traffic events.Finally,taking the microblog related to Chongqing Expressway as the research data source,this paper analyzes and studies the geological traffic events,construction traffic events and accident traffic events of Chongqing Expressway from 2015 to 2019.The results show that: in terms of the total amount of traffic events,the greater the density of road network,the higher the total amount of traffic events;The total amount of traffic events is higher at the intersection of high-speed and high-speed,high-speed and ordinary highway;Different types of traffic events have different characteristics;In addition,the spatiotemporal kernel density is easier to detect the incident prone sections than the traditional hot spot map.The spatiotemporal cube model can model and analyze traffic events from the micro and medium perspectives.The spatiotemporal cold hot spot method is also effective in evaluating the different evolution rules of traffic incident points.