Using sentiment analysis on Twitter data to identify place experience in New York City, USA

Organised by Laboratory of Geo-information Science and Remote Sensing

Tue 6 June 2017 09:00 to 09:30

Venue Gaia, gebouwnummer 101
Room 1

By Rogier Eldering (The Netherlands)

Public space in cities are the glue that holds a society together. Researchers, policy makers and designers stand to gain from an improved insight in how people experience a public space to improve existing spaces and create new ones at a higher quality. Additionally, extra insight in the spatial distribution of the space experiences is useful, as it could serve as a tool for targeted research. Many methods that are developed to determine how people experience preselected space, have limitations in one way or another. The classic methods (like Mindmaps or interviews) are either not scalable to study groups of people or the results cannot be aggregated due to their qualitative nature. With other methods, like Sentiment Analysis, it is difficult to target a specific space. The purpose of this thesis is to develop a model for adapting Sentiment Analysis to identify the experience of a space in Twitter data, for large groups of people who share a location. This research consists of a literature study to identify how people experience their environment and how this can be captured. The concept ‘place identity’, as part of space-place of Relph (1976), is used as foundation for the model. The model consists of harvesting tweets and classifying them using an adapted Sentiment Analysis. The classification is an approximation to the place identity. The spatial application potential of the model is explored with a series of projects located in New York City, which focus on the themes: spatial aggregation, spatial scales, rhythms, time and personal information. The results of the model are encouraging for theoretical application, with 72% of the tweets accurately classified. The results for the spatial application of the model leaves to be desired, they are not sufficient for a practical application. Areas of improvements are identified for the model, which should be studied before practical applications of the model.

Keywords: space, place, place identity, space experience, Natural Language Processing, Sentiment Analysis, Twitter, Tweet