The Conscious City
Big Data Fed Agent-Based Model

Master ︎ Contextual Computational Design
Supervisor ︎ Andrea Kondziela
Tools ︎ Python, Grasshopper
Place ︎ London, England
Year ︎ 2019

While a smart city focus on improving efficiency of services, a conscious city focus on improving the urban experience by applying new technology and behavioral insight. Studies show that life satisfaction is related to people’s personality and the places they live in, which suggests that in public spaces we could also identify clustering of people with similar personalities. With today’s constant stream of social media data, it is possible to paint an accurate profile of the citizens, which are the hotspots where they meet and also where they are not meeting.

TCC is a platform concept to assist both users to experience better their city and governance to respond to their needs. Using big data, it finds hotspots and similar unused places that could be also compatible with each personality.

An app connects the platform to the user, that receives suggestions of where to go according to their analysed profile and is encouraged to interact with other people on the way. Governance can redirect people to unused spots through the app, and an ABM model simulates how this outcome could impact the city.

Define bounding box around project site to stream tweets
For the ABM Model, 57.859 tweets were collected between October and November 2019 inside a radius of 5 km around the Isle of Dogs in London.

Delete tweets that are outside of the bounding box and without precise coordinates
In average, 1% of all tweets are geo-tagged with a precise point location. The rest come with a bounding box around it, causing the API to send tweets that are not exactly inside the box defined for streaming.

Delete duplicated users
The goal of the data collection is to gather users who frequent the area, so when the same user tweets from different locations only one location is kept.

Stream user’s timeline and delete those who don’t have enough material for analysis
The API used to analyse the user’s personality needs at least 100 words. After collecting the last 500 tweets from the user’s timeline, if they don’t reach the word minimum, they are deleted. At the end of the cleaning process 823 users were left.

User Analysis 
Python Connection with Twitter API

The app Personality Insights from the IBM Watson API analyses text and returns a wide range of personality traits from the user.

For this model, focused on improving the mood of the users, two traits were chosen as characteristics for the agents: depression and openness, with values going from 0 to 1.

Depression
Users who score HIGH are sad ︎
Users who score LOW are happy ︎

Openness
Users who score HIGH are Curious ︎
Users who score LOW are Cautious ︎

Agent Based Model and Simulation
Grasshopper Python

The goal of the simulation is to increase the mood of the agents. They interact when close to each other and their state changes depending on their type and on the value of the other’s state.

Curious agents adapt their mood going half way towards the other agent’s value.

Cautious agents on the other hand adapt only 1/10 of the distance between their state and the other’s.