Introduction:

Tourism is an important industry to national economy and world economy, and its
comprehensive contribution accounts for 11% of GDP in China. However, through the analysis
of data from several travel websites, we notice that the distribution of travelers is uneven among
the scenic spots. This characteristic of scenic spots’ popularity exhibits the long tail effect. In
other words, nearly 90% scenic spots are untraversed, while the rest attract almost all of the
tourists. As a matter of fact, most of these unpopular spots actually receive high evaluation from
the tourists, which implies the precious values of these unpopular spots may be substantially
underestimated. POI popularity prediction aims at analyzing the popularity condition in the next
period of time based on the features of scene spots. This research is valuable and important, for it
not only improves the accuracy of attractions recommendation and route planning for visitors,
but also provides reference information to mining commercial value hidden in the unpopular
spots. In addition, there exist many problems in traditional popular tourist attractions, such as
congestion, disappearance of characteristics, rising of entrance fee, excessive commercialization
and so on. Therefore, more and more tourists prefer to visit some unpopular scenic spots with
exceptional features. In recent years, several research endeavors have been dedicated to
popularity prediction and its related applications. Considerable studies principally focus on
content like tweets, images and videos. Most researches pay little attention to the field of
popularity prediction of tourist sites. Related work involving popularity of scenic spots mainly
concentrates on popularity prediction of attractions with sufficient information. In application
fields involving forecasting popularity of scenic spots, most approaches mainly focus on
establishing data-based prediction model by mining popular travel POIs through social media,
descriptions from a large number of travelers and highly ranked results from search engines.
Although these studies have raised some innovative approaches of applying popularity prediction
to attractions, they ignore the significant influence of emerging scenic spots and undiscovered
attractions on tourism tendency, i.e., attractions with freshness and uniqueness are likely to be
discovered by travelers and become popular in the near future. However, there are three main
challenges for POI popularity prediction working on real-world data. 1) the description
information of POI in social network is very sparse. Even in famous and widely-used websites, a
large proportion of POIs only have few photos and/or associated text. 2) Visual ambiguity is
common in different types of scene spots. For instance, it is difficult to distinguish picking
garden from natural scenery only through their images. Therefore, it is partial to predict
popularity by utilizing only low-level features. 3)Seldom researches have been made to
effectively fuse multi-modal features from multiple sources to model POI data. It is non-trivial to
effectively integrate multiple social clues for modelling POIs.