Search

Abstract With the rapid development of economy, the sharp increase in the number of urban cars and the backwardness of urban road construction lead to serious traffic congestion of urban roads. Many scholars have tried their best to solve this problem by predicting traffic congestion. Some traditional models such as linear models and nonlinear models have been proved to have a good prediction effect. However, with the increasing complexity of urban traffic network, these models can no longer meet the higher demand of congestion prediction without considering more complex comprehensive factors, such as the spatio-temporal correlation information between roads. In this paper, we propose a traffic congestion index and devise a new traffic congestion prediction model spatio-temporal transformer (STTF) based on transformer, a deep learning model. The model comprehensively considers the traffic speed of road segments, road network structure, the spatio-temporal correlation between road sections and so on. We embed temporal and spatial information into the model through the embedding layer for learning, and use the spatio-temporal attention module to mine the hidden spatio-temporal information within the data to improve the accuracy of traffic congestion prediction. Experimental results based on real-world datasets demonstrate that the proposed model significantly outperforms state-of-the-art approaches.

Abstract The outliers in traffic flow represent the anomalies or emergencies in the road. The detection and research of outliers will help to reveal the mechanism of such events. Aiming at the problem of outlier detection in urban traffic flow, this paper innovatively proposes a highly efficient traffic outlier detection framework based on the study of road traffic flow patterns. The main research works are as follows: (1) data pre‐processing, the road traffic flow matrix of the roads is calculated based on the collected GPS data, the non‐negative matrix factorisation algorithm is chosen to reduce the dimension of the matrix. (2) Road traffic flow pattern extraction, the fuzzy C‐means clustering algorithm with the Optimal k‐cluster centre (K‐FCM) is adopted to cluster the roads with the same road traffic flow pattern. (3) Outlier detection model training and evaluation, kernel density estimation is introduced to fit the probability density of roads traffic flow matrices which are used to train the back propagation neural network based on particle swarm optimisation to obtain the outlier detection and evaluation model, and a threshold is introduced to optimise the precision and recall of the model. The experimental results show that: the average precision and recall of the proposed method in this paper are 95.38% and 96.23%, respectively, and the average detection time is 28.4 seconds. The method has high accuracy, high efficiency and good practical significance.

Most existing unsupervised approaches to detect topic sentiment in social texts consider only the text sequences in corpus and put aside social dynamics, as leads to algorithm’s disability to discover true sentiment of social users. To address the issue, a probabilistic graphical model LDTSM (Long-term Dependence Topic-Sentiment Mixture) is proposed, which introduces dependency distance and uses the dynamics of social media to achieve the perfect combination of inheriting historical topic sentiment and fitting topic sentiment distribution underlying in current social texts. Extensive experiments on real-world SinaWeibo datasets show that LDTSM significantly outperforms JST, TUS-LDA and dNJST in terms of sentiment classification accuracy, with better inference convergence, and topic and sentiment evolution analysis results demonstrate that our approach is promising.

Script identification in natural scene images is a key pre-step for text recognition and is also an indispensable condition for automatic text understanding systems that are designed for multi-language environments. In this paper, we present a novel framework integrating Local CNN and Global CNN both of which are based on ResNet-20 for script identification. We first obtain a lot of patches and segmented images based on the aspect ratios of the images. Subsequently, these patches and segmented images are used as inputs to Local CNN and Global CNN for training, respectively. Finally, to get the final results, the Adaboost algorithm is used to combine the results of Local CNN and Global CNN for decision-level fusion. Benefiting from such a strategy, Local CNN fully exploits the local features of the image, effectively revealing subtle differences among the scripts that are difficult to distinguish such as English, Greek, and Russian. Moreover, Global CNN mines the global features of the image to improve the accuracy of script identification. The experimental results demonstrate that our approach has a good performance on four public datasets.