Navigation mit Access Keys

Contenuto principale

 

Using a model trained on text and image data from social media content to predict and map leisure activities

Autori
Hartmann, M.
Anno di pubblicazione
2019
Volume
153 pagine
Citazione:

Hartmann, M., 2019: Using a model trained on text and image data from social media content to predict and map leisure activities. Master thesis. 153 p.

 

Hartmann, M., 2019: Using a model trained on text andimage data from social mediacontent to predict and map leisureactivities. Master Thesis ETHZ, D-USYS. Faculty Representative: Dr. Marcel Hunziker (ETH and WSL). Supervision: Dr. Flurina Wartmann (WSL), Prof. Dr. Ross Purves (UZH), Dr. Rahul Deb Das (UZH)

 

What can we learn from novel sources of data such as social media were people perform leisure activities? Typically, surveys or observations are used to answer these questions, but such methods are not feasible to implement across large areas. This master thesis therefore investigates the potential use of social media data in the form of text and georeferenced images from the photosharing platforms "Flickr" and "Instagram" for identifying areas where people perform different recreation activities (e.g. jogging or mountain biking). A machine learning model was implemented that predicts the recreational activities depicted in the images using a combination of automated image classification and analysis of the user-generated tags. Ground truth data in the form of interviews and passive observations at several locations in the Canton Zug was acquired to validate the machine-learning classification models and the mapping of different recreation activities . The results show that both methods identify recreational activities such as walking, jogging, biking and dog walking as prevalent. Social media data covered longer time periods than the observation and surveys, but there were differences between the Flickr and Instagram platforms in terms of data distribution. Although the analysis of social media data has potential applications for recreation research, data availability often limits more fine-grained spatial and temporal analysis that would be needed to inform recreational planning.