Jointly Predicting Future Content in Multiple Social Media Sites Based on Multi-task Learning

User-generated contents (UGC) in social media are the direct expression of users’ interests, preferences, and opinions. User behavior prediction based on UGC has increasingly been investigated in recent years. Compared to learning a person’s behavioral patterns in each social media site separately, jointly predicting user behavior in multiple social media sites and complementing each other (cross-site user behavior prediction) can be more accurate. However, cross-site user behavior prediction based on UGC is a challenging task due to the difficulty of cross-site data sampling, the complexity of UGC modeling, and uncertainty of knowledge sharing among different sites. For these problems, we propose a Cross-Site Multi-Task (CSMT) learning method to jointly predict user behavior in multiple social media sites. CSMT mainly derives from the hierarchical attention network and multi-task learning. Using this method, the UGC in each social media site can obtain fine-grained representations in terms of words, topics, posts, hashtags, and time slices as well as the relevances among them, and prediction tasks in different social media sites can be jointly implemented and complement each other. By utilizing two cross-site datasets sampled from Weibo, Douban, Facebook, and Twitter, we validate our method’s superiority on several classification metrics compared with existing related methods.

Coupled CFD-FEM Simulation of Steel Box Bridge Exposed to Fire

Increasing fire-induced bridge failures are demanding more precise behavior prediction for the bridges subjected to fires. However, current numerical methods are limited to temperature curves prescribed for building structures, which can misestimate the fire impact significantly. This paper developed a framework coupling the computational dynamics (CFD) method and finite element method (FEM) to predict the performance of fire-exposed bridges. The fire combustion was simulated in CFD software, Fire Dynamic Simulator, to calculate the thermal boundary required by the thermomechanical simulation. Then, the adiabatic surface temperatures and heat transfer coefficient were applied to the FEM model of the entire bridge girder. A sequential coupled thermomechanical FEM simulation was then carried out to evaluate the performance of the fire-exposed bridge, thermally and structurally. The methodology was then validated through a real fire experiment on a steel beam. The fire performance of a simply supported steel box bridge was simulated using the proposed coupled CFD-FEM methodology. Numerical results show that the presented method was able to replicate the inhomogeneous thermomechanical response of box bridges exposed to real fires. The girder failed due to the buckling of a central diaphragm after the ignition of the investigated tanker fire in no more than 10 min. The framework presented in this study is programmatic and friendly to researchers and can be applied for the estimation of bridges in different fire conditions.

Vehicle Interaction Behavior Prediction with Self-Attention

The structured road is a scene with high interaction between vehicles, but due to the high uncertainty of behavior, the prediction of vehicle interaction behavior is still a challenge. This prediction is significant for controlling the ego-vehicle. We propose an interaction behavior prediction model based on vehicle cluster (VC) by self-attention (VC-Attention) to improve the prediction performance. Firstly, a five-vehicle based cluster structure is designed to extract the interactive features between ego-vehicle and target vehicle, such as Deceleration Rate to Avoid a Crash (DRAC) and the lane gap. In addition, the proposed model utilizes the sliding window algorithm to extract VC behavior information. Then the temporal characteristics of the three interactive features mentioned above will be caught by two layers of self-attention encoder with six heads respectively. Finally, target vehicle’s future behavior will be predicted by a sub-network consists of a fully connected layer and SoftMax module. The experimental results show that this method has achieved accuracy, precision, recall, and F1 score of more than 92% and time to event of 2.9 s on a Next Generation Simulation (NGSIM) dataset. It accurately predicts the interactive behaviors in class-imbalance prediction and adapts to various driving scenarios.

Model of predicting customer behavior based on big data analysis technologies

The analysis of the available methods and models of formation of recommendations for the potential buyer in network information systems for the purpose of development of effective modules of selection of advertising is executed. The effectiveness of the use of machine learning technologies for the analysis of user preferences based on the processing of data on purchases made by users with a similar profile is substantiated. A model of recommendation formation based on machine learning technology is proposed, its work on test data sets is tested and the adequacy of the RMSE model is assessed. Keywords: behavior prediction; advertising based on similarity; collaborative filtering; matrix factorization; big data; machine learning

Mesoscale cortex-wide neural dynamics predict self-initiated actions in mice several seconds prior to movement

Volition - the sense of control or agency over one's voluntary actions - is widely recognized as the basis of both human subjective experience and natural behavior in non-human animals. To date, several human studies have found peaks in neural activity preceding voluntary actions, e.g. the readiness potential (RP), and some have shown upcoming actions could be decoded even before awareness. These findings remain controversial with some suggesting they pose a challenge to traditional accounts of human volition while others proposing that random processes underlie pre-movement neural activity. Here we seek to address part of this controversy by evaluating whether pre-movement neural activity in mice contains structure beyond that expected from random processes. Implementing a self-initiated water-rewarded lever pull paradigm in mice while recording widefield [Ca++] neural activity we find that cortical activity changes in variance seconds prior to movement and that upcoming lever pulls or spontaneous body movements could be predicted between 1 second to more than 10 seconds prior to movement, similar to but even earlier than in human studies. We show that mice, like humans, are biased towards initiation of voluntary actions during specific phases of neural activity oscillations but that the pre-movement neural code in mice changes over time and is widely distributed as behavior prediction improved when using all vs single cortical areas. These findings support the presence of structured multi-second neural dynamics preceding voluntary action beyond that expected from random processes. Our results also suggest that neural mechanisms underlying self-initiated voluntary action could be preserved between mice and humans.