Detecting Vulnerable regions in metropolitan cities

Introduction

The problem is to handle the growing violence rate by estimating the probability of the upcoming violence, especially in metropolitan cities.

Why is the problem important?

This is important since if by doing so, we could somehow able to stop even 10-15% of upcoming threat then it can have a vast effect.

Who will benefit :

Police can analyze data in real time and may increase patrolling if required.

Based on available data, police can effectively maintain law and order in vulnerable areas.

Our strategy

For this we chose the social media platform twitter

1) First of all we collected tweets with geo tagged locations for the last 7 days for 4 citites hyderabad, mumbai, kolkata and delhi

2) But only 2% of total tweets have geo tagged locations.

So what we have done is that, we made a dictionary of areas of these cities from maps of india and find the location if it is mentioned in the tweet

like My bag is stolen from CP Delhi.

Third thing is that twitter provide user location field like i m from delhi but tweeting from hyderabad then also that location is useful somtimes.

After getting locations, We performed sentiment analysis using Crowdflower dataset as training dataset and our collected tweets as test data, in which there are 3 categories, harmful, non harmful and normal.

4) After classifying all tweets, we plotted the heatmap using Google maps api to show red regions as per harmfulness of tweets

Why location is important to us?

To find vulnerable areas our first priority is Geo-tagged location. If there is a tagged location in tweet then we are done with the location if not we choose alternatives like location mentioned in tweet, or user location. How the hell is user location is important? Suppose your home is in Delhi but you are tweeting from Hyderabad, about a riot in a your home location in Delhi. So if user will search some kind of tweet, with a specific keyword, user location might be useful to analyse a tweet.