Browsing by Subject "sports"

Being physically active is one of the key aspects of health. Thus, equal opportunities for exercising in different places is one important factor of environmental justice and segregation prevention. Currently, there are no openly available scientific studies about actual physical activities in different parts of the Helsinki Metropolitan Area other than sports barometers. In the lack of comprehensive official data sources, user-generated data, like social media, may be used as a proxy for measuring the levels and geographical distribution of sports activities. In this thesis, I aim to assess 1) how Twitter tweets could be used as an indicator of sports activities, 2) how the sports tweets are distributed spatially and 3) which socio-economic factors can predict the number of sports tweets. For recognizing the tweets related to sports, out of 38.5 million tweets, I used Named Entity Matching with a list of sports-related keywords in Finnish, English and Estonian. Due to the spatial nature of my study, I needed tweets that contain a geotag, meaning that the tweet is attached to coordinates that indicate a location. However, only about 1% of tweets contain a geotag, and since 2019 Twitter doesn’t support precise geotagging anymore with some exceptions. Therefore, I implemented geoparsing methods to search for location names in the text and transform them to coordinates if the mentioned place was within the study area. After that, I aggregated the posts to postal code areas and used statistical and spatial methods to measure spatial autocorrelation and correlation with different socio-economic variables to examine the spatial patterns and socio-economic factors that affect the tweeting about sports. My results show that the sports tweets are concentrated mainly in the center of Helsinki, where the population is also concentrated. The distribution of the sports tweets exhibits local clusters like Tapiola, Leppävaara, Tikkurila and Pasila besides the largest cluster in the center of Helsinki. Sports-wise mapping of the tweets reveals that for example racket sport and skiing tweets are heavily concentrated around the corresponding facilities. Statistical analyses indicate that the number of tweets per inhabitant does not correlate with the education level or the amount of average income in the postal code area. The factors that predict the number of tweets per inhabitant are number of sports facilities per inhabitant, employment, and percentage of children (0-14 years old) in the postal code area. Keys to a successful study when analyzing Twitter data are geoparsing, having enough data, and a good language model to process it. Despite the promising results of this study, Twitter as indicator of physical activity should be studied more to better understand the kind of bias it inherently has before basing real-life decisions on Twitter research.