- HOME
- Reports
- Early detection of spread of infection
- Infection Simulation based on Proposed Human Behavior Models derived from SNS and Press Data #20
Infection Simulation based on Proposed Human Behavior Models derived from SNS and Press Data #20
- Date
- 2021.11.30
- Researcher
- Satoshi Kurihara
- Organization
- Faculty of Science and Technology, Keio University
/
-
Faculty of Science and Technology, Keio University
Introduction of the Omicron Variant
Preparing for implementation so that it can be submitted by December 7 (Tue.)
(Settings for implementation)
The variant entered Japan in about five people on December 1
Omicron is 1.5 times more infectious than the Delta variant
The speed of replacement is twice that of the Delta variant
With regard to vaccine effects
The infection protection effect is 50%, 75% and 100% of that for the Delta variant
We Want to Know Signs from SNS: Waves

We Want to Know Signs from SNS: Increases

About the Multiagent Infection Simulator We Have Built
Satoshi Kurihara, Takumi Sugiura, Ryoya Watanabe, Naoki Wakabayashi
Daiki Kishimoto, Takashi Kawamura, Fumito Ihara and Naoto Yoshida
Faculty of Science and Technology, Keio University
Graduate School of Science and Technology, Keio University
Center of Advanced Research for Human-AI Symbiosis Society, Keio University
Even if You Have a Physical Simulation ...
Simulations you often hear about
Tsunami simulation, fire simulation, etc.
▼
Prediction is difficult even with simulations that follow the laws of physics
The global environment is complex. Various factors affect each other in complex ways.
Even if the individual laws are known, it becomes a complicated movement overall.
▼
Complex systems
There are also situations in which accurate prediction is possible if there is an enormous amount of past data.
The performance of ML has been especially high in recent years.
However,
Are Google predictions of infected people accurate?
Abundant data is not enough ➡ Abundant data cannot be collected in the first place
Short-term decisions such as stock trading can be made using AI (low-yield stability) *Same as the latest weather forecast
Long-term economic trend prediction is difficult

Simulations Using People as the Unit
Social simulation: We want to understand, predict and respond to social phenomena by reproducing people’s movements
Simulating society = the sum of simulating human behavior
An evacuation simulation is a simulation of human movement for the simple purpose of escape
→ Although they are difficult, simulation accuracy is high because of limited space and time.

Simulation for prediction of the number of people newly infected with COVID-19 and the number of people with severe illness
* This is different to simulation of the process whereby the virus infects a person and illness occurs <- That is close to physical simulation and is still easy to predict.
Infection occurs through person-to-person contact ➡ The simulation of human movement and contact is important
How should respectively different individuals be modeled?
AI simulation, social simulation, and multi-agent simulation
Agent: Mini autonomous AI that models individuals
Mini autonomous AI simulates the movement of society by moving and exchanging information like real society.
Prediction of Phenomena on a Society Level Using People as the Unit .....
Simulation by realizing an accurate, ultra-precise digital twin is unrealistic
➡ How concisely can we include only important elements in the simulation?
➡ How do we model to grasp the nature of complex societies, and understand and predict reality?
Things that are often pointed out < a tendency to be arbitrary > < assumptions are included > → yuck!!
Solution: Find common knowledge by conducting various simulations.
➡ The Cabinet Secretariat’s AI & Simulation Project may be the first successful example of a simulation initiative.
Focus on common behavioral characteristics of people: Basically follows power law
Imagine the distance of movement each day of the year before the COVID-19 pandemic.
The majority would be travel from home to work with an occasional long business trip
Even if you look at movement in one day,
the majority would be at home or travel between desks, cafeterias, toilets and meeting rooms at work.
Occasionally, you would move to another department or a large meeting room some distance away
This kind of distribution, in which the overwhelming majority is small and very few are large is called a power-law distribution.
The way people move is basically a power law
The cause of infections is contact between people
→We model people’s movement by power law.

Design of Human Behavioral Models
• The frequency of a person’s risky behavior is a power law
→Everybody engages in risky behavior occasionally.
→Very few people engage in risky behavior frequently.

● Anxiety coefficient: High in situations where there is a threat to the unknown and drops sharply when the threat is gone (close to animal instinct).
● Behavioral coefficient: The younger a person is, the larger this becomes (socioeconomic desire for action) *This increases as any anxiety factors disappear.
● Restrictive policies: The degree to which actions are restrained by policies such as declarations of a state of emergency and legislation on epidemic prevention measures.
● Economic policies: The degree to which actions are stimulated by policies like the Go To campaign.

Human Behavioral Network: Infection Network is a Small World Network
Small-world network:
A network in which communities (dense networks) are connected to each other
*Infections spread rapidly due to infections straddling communities.
Simulation with model case
We start the simulation from infections occurring in Tokyo and Osaka and there are no infected people in Naha or on remote islands.
* This is a scenario in which infections will occur in Naha and on remote islands as a result of high-risk behavior due to the movement of infected people from other areas.

Even infrequent travel is associated with high-risk behavior, and invites an infection epidemic in uninfected areas and pressure on health care.
Reducing the frequency of movement with high-risk behavior can be expected to reduce the level of an infection epidemic in areas with vulnerable medical systems.

Predicted and actual number of positive cases as of August 3 (reproduction of fifth wave)

The factor is the synergistic effect of vaccinations and the human movement network structure
* Early acquisition of herd immunity (which is merely dynamic)
* The quantity of flows of people very likely demonstrated a restraining effect.
Why Only Japan (Currently)?
・The butterfly effect (Does a butterfly flapping its wings in Brazil cause a tornado in Texas?)
In complex systems, small differences can lead to large changes (that cannot be felt at the individual level)
There is actually no big difference between Japan and overseas.
Why only Japan? ➡ It is merely dynamic equilibrium
The goodness of Japanese morals (masks, behavioral changes, etc.), which may be a very small factor, has created a big difference (X factor)
*The proportion of unvaccinated people among people traveling between cities (long distance travel) < the overwhelming proportion of people engaging in moderate behavior
Preventing an infection epidemic at the travel destination (a very small number of self-restraints generating a significant effect)
It stems from the question posed by the meteorologist Edward Lorenz “Do very small disturbances like a butterfly flapping its wings affect the weather in distant places?” and, if that is true, his suggestion that accurate long-term predictions are fundamentally difficult unless observational errors can be eliminated, which emerged from research on numerical weather prediction [4] [5]. - wikipedia
Latest Simulation Settings (for Tokyo)
・The final vaccination rate for the first and second doses was set to 80%
(Final vaccination rate for each generation: “- 14” 0%, “15 - 39” 80%, “40 - 64” 80%, “65 -” 95%)
・Third vaccination doses started in December at 90% the pace of second doses
・Vaccine passports began in November, restricting the behavior of people who have not had a second dose
With regard to effects, simulations was implemented with three scenarios setting self-restraint to 0%, 20% and 30%
・Antibody cocktails administered from December so that the number of new patients with severe illness is kept to ten people a day
Effect in preventing increased severity of illness set to 80% for patients administered antibody cocktails
・The rates of increased severity of illness are calculated as shown in the table on the right

Tokyo: Number of Newly Infected People

Third Dose Vaccination Rate by Age Group in Tokyo

Tokyo: Number of Severely Ill Patients and Number of Antibody Cocktails/Medications Required

Summary
Social simulation is a necessary and essential function to maintain and develop society
→This is natural if you think about it carefully
People: Free energy principle/AI: world model
→Prediction based on simulation is the fundamental principle
The world has a fractal structure -> a society, a group of people, has a necessarily similar dynamic
The level of social simulation is still low
→ Scale on the time and space axes is small and scattered
→Prospects for multi-scale social simulation
Mobile Spatial Statistics (Population Change)
Kabukicho, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Shibuya Center Gai, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Ikebukuro, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Harajuku, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Ueno Ameyoko, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
South side of Shinagawa Station, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Tokyo Station, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Odaiba/Tokyo Teleport, Tokyo (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Yokohama Station, Kanagawa (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Kawasaki Station, Kanagawa (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Hiyoshi Station, Kanagawa (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Kitashinchi, Osaka (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Kawaramachi, Kyoto (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Nakasu Kawabata, Fukuoka (Number of people in the area by time axis)

Mobile Spatial Statistics (Population Change)
Susukino Station, Hokkaido (Number of people in the area by time axis)

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Aggregation Definition
•Comparison with the same week (Monday-Sunday) of 2019
•The subject restaurants are 10,000 where Toreta was introduced before January 2019
•Because there was a typhoon on the weekend of Week 41 (second week of October) in 2019 in the Kanto region, the results may be a little high.
Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
National average
*Periods of the declaration of a state of emergency or pre-emergency measures in Tokyo

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
By number of customers
*Periods of the declaration of a state of emergency or pre-emergency measures in Tokyo

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
By time
*Periods of the declaration of a state of emergency or pre-emergency measures in Tokyo

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
By timing of reservation
*Periods of the declaration of a state of emergency or pre-emergency measures in Tokyo

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
By restaurant size
*Periods of the declaration of a state of emergency or pre-emergency measures in Tokyo

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
By frequency of visits
*Periods of the declaration of a state of emergency or pre-emergency measures in Tokyo

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Hokkaido
*Periods of the declaration of a state of emergency or pre-emergency measures in Hokkaido

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Tohoku
*Periods of the declaration of a state of emergency or pre-emergency measures in Miyagi Prefecture

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Kanto
*Periods of the declaration of a state of emergency or pre-emergency measures in Tokyo

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Hokuriku
*Periods of the declaration of a state of emergency or pre-emergency measures in Ishikawa Prefecture

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Tokai
*Periods of the declaration of a state of emergency or pre-emergency measures in Aichi Prefecture

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Kinki
*Periods of the declaration of a state of emergency or pre-emergency measures in Osaka Prefecture

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Chugoku
*Periods of the declaration of a state of emergency or pre-emergency measures in Hiroshima Prefecture

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Shikoku
*Periods of the declaration of a state of emergency or pre-emergency measures in Ehime Prefecture

Data on the Number of Customers at Restaurants Using the Toreta, Inc. App
Kyushu/Okinawa
*Periods of the declaration of a state of emergency or pre-emergency measures in Fukuoka Prefecture

Related reports
Report by Satoshi Kurihara
-
Understanding and prediction of infection status based on the basic model of social atmosphere, people, and movement #16
-
Understanding and prediction of infection status based on the basic model of social atmosphere, people, and movement #15
-
Understanding and prediction of infection status based on the basic model of social atmosphere, people, and movement #14