Urban big data fusion creates huge values for urban computing in solving urban problems. In recent years, various models and algorithms based on deep learning have been proposed to unlock the power of knowledge from urban big data. To clarify the methodologies of urban big data fusion based on deep learning (DL), this paper classifies them into three categories: DL-output-based fusion, DL-input-based fusion and DL-double-stage-based fusion. These methods use deep learning to learn feature representation from multi-source big data. Then each category of fusion methods is introduced and some examples are shown. The difficulties and ideas of dealing with urban big data will also be discussed.
Our life and the city we live in affect each other. In the era of big data, it is urgent to effectively use urban big data to solve problems in the city, such as traffic congestion [1,2], noise pollution [3,4], air pollution [5,6], etc., to improve our life experience. Nowadays, many urban computing methods based on deep learning have been put forward to solve urban problems, such as urban traffic flow prediction [7,8], urban crowd flows prediction [9,10], urban air prediction [11,12], urban water quality prediction [13,14], etc. In these urban computing methods, the big data used by the researchers are all from different sources, such as meteorological stations, taxi detectors, online weather web sites, etc. Moreover, urban big data shows different representations, such as text, numbers and symbols. Bello et al.  and Zhang et al. summarized five characteristics of big data, that is, large volume, large velocity, large variety, veracity and value, which are called 5V’s features. The 5V’s features of the data indirectly indicate a big explosion in data amount. On the one hand, how to sense, obtain and manage these big data is a challenge; On the other hand, how to analyze and excavate the value of these big data is another significant challenge. Apparently, the urban big data with 5V’s characteristics brings great challenges to urban computing. Fig. 1 depicts the urban big data. Firstly, urban big data comes from many sources. When studying the real-time city-wide traffic volume, the data usually come from taxi sensor, exploratory data, monitoring data and Internet web data. For example, Meng et al.