Embedding Geographic Locations for Modelling the Natural Environment Using Flickr Tags and Structured Data

Abstract

Meta-data from photo-sharing websites such as Flickr can be used to obtain rich bag-of-words descriptions of geographic locations, which have proven valuable, among others, for modelling and predicting ecological features. One important insight from previous work is that the descriptions obtained from Flickr tend to be complementary to the structured information that is available from traditional scientific resources. To better integrate these two diverse sources of information, in this paper we consider a method for learning vector space embeddings of geographic locations. We show experimentally that this method improves on existing approaches, especially in cases where structured information is available.

A GloVe-Based POI Type Embedding Model for Extracting and Identifying Urban Functional Regions

“…To solve the this drawback of word2vec, Pennington et al [16] proposed a word embedding model named Global Vectors (GloVe) that integrates the global statistical information on the basis of the local context window-based method, thus combining the advantages of both topic models and neural network-based models. Jeawak et al [34] combined Flickr (a photo sharing website) tag data and traditional structured data (e.g., temperature, land use) to embed geographical locations into vectors, verifying that GloVe is more effective than are traditional bag-of-words models in ecology-related tasks (e.g., predicting species distribution, soil types, and land cover). To the best of our knowledge, the feasibility of applying the GloVe model to urban computing remains to be proven.…”

Predicting environmental features by learning spatiotemporal embeddings from social media

“…We found that the predictive value of Flickr tags is roughly on a par with that of standard commonly available environmental datasets, and that combining both types of information leads to significantly better results than using either of them alone. In (Jeawak et al, 2019), we proposed the EGEL (Embedding GEographic Locations) model that integrates both Flickr and environmental data into low-dimensional vector space embeddings. This model was found to outperform the bag-of-words model for all the evaluation experiments.…”

“…Intuitively, to determine whether a given tag is time and/or location specific, we assess to what extent the distribution of its occurrences across all spatiotemporal cells diverges from the overall distribution of all tag occurrences. To this end, we use a method based on Kullback-Leibler (KL) divergence, which was previously found to be effective in Van Laere et al 2014and Jeawak et al (2019). In particular, we select those tags ⊆ which maximize the following score:…”

“…This method represents each location as a concatenation of two feature vectors: a bag-of-words representation derived from Flickr and a feature vector encoding the numerical and categorical features obtained from the structured dataset. Following on from that approach we experimented in (Jeawak et al, 2019) with the EGEL (Embedding Geographic Locations) model, which learns vector space embeddings of geographic locations by integrating the textual information derived from Flickr with the numerical and categorical information derived from environmental datasets. We found that this approach led to more accurate predictions than the previous approach from (Jeawak et al, 2017) that concatenated the bag of words data with the structured data.…”

“…In this paper, we extend our approach from (Jeawak et al, 2019) by considering a spatiotemporal representation of regions. In particular, we learn a vector space embedding for each geographic region and each month of the year, which allows us to capture environmental phenomena that may depend on monthly or seasonal variation.…”