Project 2

This project aims to investigate the correlation between social media sentiment and crime rates within local government areas (LGAs) in Victoria, Australia. The underlying hypothesis is that negative sentiments reflected in social media posts may indicate a hostile environment, potentially correlating with elevated crime rates. To explore this hypothesis, we make use of a variety of data sources including a large Twitter corpus, crime statistics from the Spatial Urban Data Observatory (SUDO), geospatial data from an open dataset, and harvested data from Mastodon?s general channel. The project involves developing a cloud-based solution on the University of Melbourne?s Research Cloud. Our setup employs four instances for data storage, processing, and visualization of the results. The architecture leverages Docker containers for deploying CouchDB databases, analytics scripts, and an NGINX web application. By utilizing Ansible, we can ensure scalability and ease of deployment across our infrastructure.

Heat Map of Tweets & Total Sentiment Produced

The Heat Map visualization provides a striking graphical representation of the distribution of Tweets and their associated sentiment across Victoria, Melbourne. Areas of higher intensity or 'heat' are concentrated mainly in the city center and a few areas in the north east, indicating a higher volume of Twitter activity and sentiment emanating from these locales. Interestingly, the pattern seems to align somewhat with population density, suggesting a potential relationship between the number of inhabitants in a given area and the intensity of Twitter activity and sentiment. This juxtaposition of digital sentiment with physical geography underscores the pervasive impact of social media on our daily lives, and offers intriguing possibilities for further study into the interplay between location, population density, and digital sentiment.

Joint Geographical Visualisation representing Sentiment & Subjectivity of Tweets and their Average Word count

The Joint Geographical Visualization offers an intriguing perspective on the relationship between the sentiment, subjectivity, and average word count of Tweets across Victoria, Melbourne. This visualisation reveals that the majority of Tweets fall within an average length of 14-19 words, regardless of the sentiment and subjectivity expressed within them. Moreover, the sentiment analysis clearly demonstrates that some Local Government Areas (LGAs) emit more negative tweets than others, suggesting a potential correlation between the mood of the tweets and the geographical region from which they originate. This geographical disparity in sentiment and subjectivity, when cross-referenced with tweet length, provides a unique lens through which to understand the dynamics of social media communication across different regions.

Heat map of the kind of Crimes Division Statistics recorded across different suburbs

The Heat Map of Crime Division Statistics provides a comprehensive overview of the types and distribution of crimes committed across different suburbs in Victoria, Australia. The visualization reveals that the most prevalent crimes come from Division B, which includes property and deception offences such as arson, property damage, burglary/break and enter, theft, deception, and bribery. Following this, Division E crimes, centered around justice procedures offences and breaches of orders, are the second most common. Interestingly, drug offences (Division C) and public order and security offences (Division D), although significant, are markedly less common. The map also uncovers that certain suburb clusters have a higher crime rate than others, suggesting a potential geographical concentration of certain crime types. This finding might overlap with the presence of more hostile sentiments in tweets originating from these areas, indicating a possible correlation between social media sentiment and crime rates.

Spatial Viz of Crime occurrence

toot Distributions

Sentiment and subjectivity Distributions

Team