arXiv subjects

Arman Sarjou

Publications and source records attributed to Arman Sarjou.

2 recordsLinked to original sources

arXiv2021

The Power of Language: Understanding Sentiment Towards the Climate Emergency using Twitter Data

Understanding how attitudes towards the Climate Emergency vary can hold the key to driving policy changes for effective action to mitigate climate related risk. The Oil and Gas industry account for a significant proportion of global emissions and so it could be speculated that there is a relationship between Crude Oil Futures and sentiment towards the Climate Emergency. Using Latent Dirichlet Allocation for Topic Modelling on a bespoke Twitter dataset, this study shows that it is possible to split the conversation surrounding the Climate Emergency into 3 distinct topics. Forecasting Crude Oil Futures using Seasonal AutoRegressive Integrated Moving Average Modelling gives promising results with a root mean squared error of 0.196 and 0.209 on the training and testing data respectively. Understanding variation in attitudes towards climate emergency provides inconclusive results which could be improved using spatial-temporal analysis methods such as Density Based Clustering (DBSCAN).

cs.CL↗

arXiv2021

Violent Crime in London: An Investigation using Geographically Weighted Regression

Violent crime in London is an area of increasing interest following policing and community budget cuts in recent years. Understanding the locally-varying demographic factors that drive distribution of violent crime rate in London could be a means to more effective policy making for effective action. Using a visual analytics approach combined with Statsitical Methods, demographic features which are traditionally related to Violent Crime Rate (VCR) are identified and OLS Univariate and Multivariate Regression are used as a precursor to GWR. VIF and pearson correlation statistics show strong colinearity in many of the traditionally used features and so human reasoning is used to rectify this. Bandwidth kernel smoothing size of 67 with a Bi-Square type is best for GWR. GWR and OLS regression shows that there is local variation in VCR and K-Means clustering using 5 clusters provides an effective way of seperating violent crime in London into 5 coherent groups.

cs.CY↗