روش پیش بینی فعالیت فضایی و زمانی
ABSTRACT
I. INTRODUCTION
II. RELATED WORK
III. PROBLEM DEFINITION
IV. THE FRAMEWORK
V. FAST LIGHTWEIGHT SPATIOTEMPORAL ACTIVITY PREDICTION
VI. EXPERIMENT AND ANALYSIS
VII. CONCLUSION AND FUTURE WORK
REFERENCES
ABSTRACT
How to predict spatiotemporal activity from geo-tagged social media is an urgent problem. Existing methods don’t make full use of spatiotemporal information and text sequence features. In view of above problem, we design a Fast Lightweight Spatiotemporal Activity Prediction method(FLSAP) based on Gated Recurrent Unit(GRU) neural network. While GRU structure can extract text sequence features, the model takes up a lot of space due to the numerous parameters. At the same time, due to the long sequence in the text, the convergence speed of GRU is slow. So, we design a novel GRU neuron, GRU with Tiny and Skip(GTS), which can quickly generate a lightweight model with higher accuracy. In GTS, we add a scalar weighted residual connection to stabilize the training. Furthermore, we extend the residual connection to a gate by reusing the parameter matrices to compress the model size. At last, in order to make the model converge faster, we add a binary gate, which determine whether to skip the current state update. According to the experimental results, compared with ReAct [1] in the spatiotemporal activity prediction task, FLSAP improves the accuracy by 3.3%, reduces the model space by 98.79% and accelerates 74.4% of convergence speed.
INTRODUCTION
Recently, big cities face a big challenge when people try to find their desired activities. Imagine if a tourist is in a strange city, how can he/she get information about the popular activity in his/her neighborhood at the time being quickly and accurately? Especially in the age of increasing information, even a local person can hardly answer this question accurately in a short time. However, geo-tagged social media(GTSM) has made it possible to solve this problem. Some studies [2]–[۸] have demonstrated that GTSM has great potential in predicting spatiotemporal activity. GTSM includes not only timestamp and geographic coordinates, but also text generated by users using social media. Twitter is a geo-tagged social media, a large number of users use Twitter to generate a large number of messages with time and location tags every day. And these messages are adopted by studies [9]–[15] as data source. These messages contain information about main local activity. For instance, if there are many restaurants in a region, the chances of tweets related to food in this area will be much greater than areas with fewer restaurants. In addition to time and place, text plays a crucial role in the activity prediction process. So, capturing more information from the text will provide more help for activity prediction. In addition, GTSM typically relies on mobile smart terminals. Although the computing power of mobile intelligent terminals is gradually improving, how to quickly get a model that can accurately predict activity while occupying as little space as possible is still an urgent problem to be solved.