Developing a Location Distortion Model to Improve Reverse Geocoding with Weather Data

Reverse geocoding is the process of assigning a readable place name or address to a point location. Common reverse geocoding methods assign the shared location to the closest venue based on Euclidean distance. In recent years, due to the advancement in positioning technology, a huge amount of spatial data has been generated by location-based social networks such as Yelp and Swarm. Additionally, various services offer the ability to provide online weather data in any coordinate and time. These data can be a valuable source of behavior patterns of different people in different weather conditions. Our study efforts to enhance the reverse geocoding based on spatial distance with the help of these data. In this way, weather condition data were used to make behavior patterns of people and check-in data were collected with the help of Swarm service. Swarm service which was used in our study is a new service from Foursquare that enables gathering check-in through the Twitter Streaming API. After gathering each check-in with Twitter streaming API, Weather data were provided instantly using the OpenWeatherMap API. Weather data were included various attributes that four of them were used in our study. Weather data were used in four categories, including; air humidity, air temperature, wind speed, and cloudiness to produce weather semantic signatures. In our study, linear, rational and sinusoidal functions were used for distorting the spatial distance with weather check-in probability in the process of reverse geocoding. In addition, two training and test data sets have been used in our case study (New York State) to specify the values of the model parameters and to evaluate the result. For the training process of location distortion functions, the check-in data were collected for New York State for one year from 01/03/2017 to 01/03/2018. The results showed that with the linear model and weather semantic signatures, the reverse geocoding results (based on spatial distance) of MRR and First Position indices (New York State) could be improved by 18.64% and 111.49%, respectively. For the process of evaluating linear location distortion function, the check-in data were collected for New York State for seven days from 01/03/2018 to 07/03/2018. The results showed that the reverse geocoding results (based on spatial distance) of MRR and First Position indices (New York State) could be improved by 13.40% and 66.96%, respectively. These results indicated the high capability of the presented model to be used outside of the timeframe of training data. In our study, one of the important challenges in the geolocation services, named the reverse geocoding process, was investigated. The model presented in this study was able to modify the distance between individuals and venues by linear location distortion function. Given that, this model has demonstrated its ability to be used with weather (and temporal) semantic signatures. It can be expected that future studies use other contextual data by location distortion functions.