Team Name

Data @ Heart

Team Members

Project Description

• Our project aims to be a service to Australians living their lives, telling their stories through social posts to help people to be more aware of the issues facing themselves. It aims at improving the safety and peace of mind of people in and about our great, vibrant urban, regional and open spaces.

• Joining BOCSAR crime data (22 years data across NSW postcodes) with layers of geolocated, time-filtered social posts, (Twitter, Instagram and Facebook) that have been publicly posted and pinned, we process these for NLP semantic and entity understanding, in order to predict and offer insight to users of our app. The information of trend predictions of crime in locations around where they are, helps better inform them of the risks that might be experienced. This is particularly important given the #MeToo movement and tragic circumstances experienced by the likes of Jill Meagher and Eurydice Dixon, as well the alcohol related “king-hit” punch crimes that ended in the tragic death of Daniel Christie.

• With our passion for data at the heart of decision making, we hope to make Australian society a better place to live, love and enjoy.

• This too, would help utilise police resources patrolling areas of predicted high crime areas based on our modelling calculations in a simple to use easy to access app on a smartphone. Our layers of geolocation mapping also blends and calculates the location of pubs and restaurants to weight the variables of incidence of crime, due to such factors of alcohol consumption. The data from AIHW has added the ability to incorporate trends and predictions from consumption of alcohol (particularly the increase in consumption of wine in pubs and restaurants) and aims to use neural network image recognition of socal (e.g.: Instagram) images into the future to dynamically model risk factors around the app user.

• The data that we have brought together is difficult to join and blend in a meaningful easy to use way, but by starting at the audience benefit of safety information, spatial awareness and the ability to make data aware choices it betters their ability to enjoy their lives, with knowledge of past and predicted future safety values in and around where they are living their lives, socialising or commuting.

• Data @ Heart team has particularly enjoyed using our knowledge to make this blended weighted decision tool for all members our community.

https://public.tableau.com/profile/tones#!/vizhome/CrimeByPostcode1995-2017WithSocialPostOverlaid/CrimeByPostcode1995-2017WithSocialPostsIncidences?publish=yes

https://public.tableau.com/profile/tones#!/vizhome/CrimeByPostcode1995-2017WithSocialPostOverlaid/CrimeByPostcode1995-2017Categories?publish=yes

https://public.tableau.com/profile/tones#!/vizhome/CrimeByPostcode1995-2017WithSocialPostOverlaid/SocialPostsofCommunityReportsPotentialWarnings?publish=yes

Data Story

• Joining BOCSAR crime data (22 years data across NSW postcodes) with layers of geolocated, time-filtered social posts, (Twitter, Instagram and Facebook) that have been publicly posted and pinned, we process these for NLP semantic and entity understanding, in order to predict and offer insight to users of our app. The crime is split into 17 major categories which would work as user personalisation filters to tailor the experience to individual concerns.

• We first gathered NSW police crime reports into a corpus of text data to train topic vectors which are then matched in closeness values to identify suitable relevant social posts, that are geolocation enabled, mapped in the area that you are currently in.

• The information of trend predictions of crime in locations around where they are, helps better inform them of the risks that might be experienced. This is particularly important given the #MeToo movement and tragic circumstances experienced by the likes of Jill Meagher and Eurydice Dixon, as well the alcohol related “king-hit” punch crimes that ended in the tragic death of Daniel Christie.

• This too, would help utilise police resources patrolling areas of predicted high crime areas based on our modelling calculations in a simple to use easy to access app on a smartphone. Our layers of geolocation mapping also blends and calculates the location of pubs and restaurants to weight the variables of incidence of crime, due to such factors of alcohol consumption. The data from AIHW has added the ability to incorporate trends and predictions from consumption of alcohol (particularly the increase in consumption of wine in pubs and restaurants) and aims to use neural network image recognition of socal (e.g.: Instagram) images into the future to dynamically model risk factors around the app user.

• We identified a particular growth recently in the occurrence of transport related crimes in the 22 years of BOCSAR data and so would plan on building into the distance weighting of risk value to the users current location, further factorisation of key train stations and transport hubs where crime incidents have occurred.

• The data that we have brought together is difficult to join and blend in a meaningful easy to use way, but by starting at the audience benefit of safety information, spatial awareness and the ability to make data aware choices it betters their ability to enjoy their lives, with knowledge of past and predicted future safety values in and around where they are living their lives, socialising or commuting.

• Data @ Heart team has particularly enjoyed using our knowledge to make this blended weighted decision tool for all members our community.

Structural Topic Model Project - Crime Data

Sarah Fawcett + Tony Nguy

07/09/2018

library(stm)
library(igraph)
library(stmCorrViz)
library(tidyverse)
library(dplyr)
library(stringr)
library(tidytext)
library(car)
library(reshape2)
library(lubridate)
library(ggpmisc)

Set working directory

setwd("~/Documents/DATA-SCIENCE/GOVHACK/DATA")

Clean Out Old Objects

rm(list = ls())

rm(Crime9517_wide)

1: INGEST (PROTOTYPE)

Crime9517wide <- read.csv("CrimePostcodeData1995_2017.csv", header=T)

Convert to Long Format

Crime9517long <- melt(Crime9517wide, id=c("Postcode","Offence.category","Subcategory"))

Crime9517longdate <- melt(Crime9517_wide, id=c("Postcode","Offence.category","Subcategory"))

2: PREDICT

Test of factor

class(Crime9517_long$variable)

Convert Data to Date If Necessary

mdy(Crime9517_long$variable)

Crime9517long$variable <- as.Date(Crime9517long$variable)

Crime9517long$variable <- as.factor(Crime9517long$variable)

Convert Postcode to Character

class(Crime9517_long$Postcode)

Crime9517long$Postcode <- as.numeric(Crime9517long$Subcategory)

Crime9517long$Postcode <- as.numeric(Crime9517long$Postcode)

Crime9517long$value <- as.numeric(Crime9517long$value)

Generalised Linear Regression Model

attach(Crime9517_long)

model.glm.crime <- glm(Crime9517long$Postcode ~ Crime9517long$value + Crime9517long$Subcategory + Crime9517long$variable)

Summary

model.lm.internet

Pedict

predict.glm(model.glm.crime, data.frame(value=30, Subcategory=71, variable=21017, type="response", interval="confidence"))

detach(Crime9517_long)

WRITE

write.csv(Crime9517longdate, file = "~/Documents/DATA-SCIENCE/GOVHACK/DATA/Crime9517longdate.csv", row.names=FALSE)