How to “model the future” at ITMO University: from predicting crowd behavior to analyzing opinions in social networks
Is it possible to predict the behavior of the crowd? Scientists from the Institute of High-Tech Computer Technologies (NII NKT) at ITMO University undertook to solve this problem. They created a system that simulates variants of the development of events in crowded places, whether it be a stadium during the World Football Championship or holy places during the period of mass pilgrimage.
The model is built on the basis of crowd characteristics, such as social structure, and external factors, such as weather conditions or a political situation. Also set the parameters of the territory where the action takes place. As a result, scientists see a clear picture of the behavior of people in given conditions. It looks like this:
Important characteristics of the crowd are the density and speed of movement. The system clearly shows the places of increased density, where tensions increase and emergency situations occur: crush and conflicts. By varying the parameters of the area of the event is determined by its optimal configuration for the upcoming event.
')
Using the software system, you can recreate any territory and external conditions, and then place in this environment any group of people. This opens up tremendous research opportunities. You can simulate an earthquake in Japan during a holiday, when there are thousands of people on the streets. Nobody forbids the researcher to replace the Japanese with the British, and the earthquake - to the loss of Manchester United.
But in order for the behavior of a virtual crowd to be reliable, it is necessary to study the logic of human behavior in simulated situations. To do this, conducted "field" studies in crowded places. Research methods are different. The main ones are: observation, photo and video filming, surveys and the use of special sensors (GPS trackers, accelerometers, proximity sensors).
Sergey Ivanov, Head of the International Laboratory "City Informatics" of ITMO University, explained that surveillance cameras at different points in the territory would be enough for the basic work of the models. By highlighting the flow of people from the video series, you can adjust the model to the real situation and make short-term forecasts.
How to evaluate the mood of the crowd, not being in it? Help social networks. After the images with the necessary hashtags or geo-referencing arrive on the server, the model allocates faces to them, according to which machine learning algorithms determine the mood of people.
Proximity sensors — devices that “see” each other at a given distance and fix it — are not less promising. If such a device is tied to a specific person, you can restore the graph of his movements, as well as determine how other factors (distance to relatives or friends) affect their movement in a crowd. However, the use of sensors can be problematic.
“The main disadvantage of this approach is the complex infrastructure preparation of experiments, the need for volunteers to participate in the distribution / collection of special devices and the inability to use the results obtained on-line,” says Ivanov. The scientist adds that the game, nevertheless, is worth the candle - observing individual people allows you to get a deeper knowledge of the nature of the movement of the crowd.
Methods for preparing input data — that is, information about the event area — range from examining photographs to reconstructing three-dimensional objects from a point cloud obtained using drones and scanning tools.
A field study using these tools was recently conducted by ITMO staff together with colleagues from other countries at the Kumbh Mela religious festival, which takes place every 3-4 years in India. During the month, researchers observed a pilgrimage of about 70 million people to the holy places of Hinduism.
Scientists managed to identify stereotyped actions for the participants of the festival and their interrelations. Scenarios included “dead ends”, “obstacles” and “through passages”, as well as crowd behavior in extreme situations, such as heavy rain.
Behavior options were considered in various external conditions. In India, the most important of these was the air temperature, which sometimes exceeded 40 degrees.
A similar study was conducted at the St. Petersburg VK fest this year. The innovation consisted in analyzing photos of the event, published by users, and identifying the most popular areas of the festival based on geo-tags. Most of all people were attracted by the “Drive” zone (an exhibition of tuned cars), the art and food zones were almost keeping up with it.
When the field stage is completed, the information is processed on the computer and then transformed into a model. That's just the usual PC giant amount of modeling work can not do, because each element of the crowd and the external environment is presented in the form of equations and sets of rules. To recreate the typical situation of a mass gathering of people, it is necessary to perform millions of computational operations, so scientists work with the Lomonosov-2 and Lobachevsky supercomputers , one of the most powerful in the world. The first is at Moscow State University, and the second is at Nizhny Novgorod State University.
According to Ivanov, the calculations are seriously complicated by the chaotic movement of human masses, which is reflected in the Monte Carlo modeling approach. “On the one hand, the movement of an individual in a crowd is determined by a physical model that is deterministic, however, the generation of desires (changing the point of interest, choosing the desired speed, unexpected stops, and so on) is a result of a discrete choice that is formalized in terms of random processes”, - says the head of "City Informatics". To obtain reliable results, virtual experiments must be repeated many times with different inputs.
Similarly, scientists have already predicted the possible behavior of fans heading to the subway from the Zenit Arena after the match.
The simulation revealed places with potentially the greatest crowd pressure. To protect the fans leaving the stadium after the match, you can adjust the flow by changing the configuration of the fence or exits from the stadium.
Algorithms developed by scientists can predict not only the behavior of people in a crowd. They make it possible to determine the pathways for the spread of viral diseases, draw up a scheme for anesthesia in the city, and even evaluate public opinion — that is, simulate all the processes that are associated with contacts between people and their mobility.
For example, ITMO University is working on modeling the incidence of influenza. The infection process can be illustrated both for confined spaces, such as trains, and for the whole city. In the second case, the task is complicated many times, since it is necessary to take into account all the peculiarities of the city’s life, to reflect all social interactions, while for the train it is enough to describe the stereotypical behavior of passengers, determined by a small number of rules.
Another object of study are social networks. Studying the Vkontakte users posts or comments on the video on YouTube, you can find out the opinion that people share about power, social problems, products and their manufacturers. In particular, the Scientific and Research Institute of NKT of ITMO University studies messages from social networks Vkontakte, Twitter, Instagram and Live Journal. At the same time, scientists can not only analyze opinions and determine “social drafts”, but also distinguish bots from real people, as well as assess the impact on the network and public opinion of statements made by popular bloggers and famous personalities.
At the heart of these technologies is the concept of crawling. In accordance with a given strategy, the program scans the resources of social networks and enters them into a database for further analysis. Each resource, up to individual posts, is determined by a set of characteristics, such as the number of likes, reposts, external citations and comments. All these data are combined by a graph of relationships, on the basis of which resources are ranked according to a given criterion. According to Sergey Ivanov, the approach allows not only to model the dissemination of information on such a graph, but also to form strategies to counter the "manipulation of the network community."
This technology has a wide range of applications. So, in particular, the crawling and analysis of social networks is actively used to study social processes in e-government technologies (eGovernance).
NII NKT is rapidly developing , and the results of its research are demonstrated at international scientific conferences, including the recent ICCS'2016 in San Diego, USA. The institute attracts government "megagrants" and foreign experts.
The scientific leader of the Laboratory for Advanced Computational Technologies at the Scientific and Research Institute of NKT is Peter Sloots, a professor at the University of Amsterdam. Also, foreigners work in the laboratory remotely: in Singapore, Poland, Holland - all of them are attracted by Russian developments and the opportunity to participate in promising projects of the Research and Development Institute of NKT.
In 2006, only four employees worked at the Institute. After 10 years, their number has grown to 100, and the institute continues to search for talents. For example, if you are a bachelor and / or a master of applied mathematics and computer science or a related field (STEM - Science, Technology, Engineering and Mathematics), in particular, in the direction of "ICT" and "programming", with English not lower than Upper Intermediate, also:
then you have the opportunity to join the research institutes of the ITMO ITMO , which are not limited to studying the behavior of the crowd. The Institute plans to apply developments in various fields - from biomedicine and urban planning to forecasting and preventing natural and man-made disasters.
From chaos to model
The model is built on the basis of crowd characteristics, such as social structure, and external factors, such as weather conditions or a political situation. Also set the parameters of the territory where the action takes place. As a result, scientists see a clear picture of the behavior of people in given conditions. It looks like this:
Important characteristics of the crowd are the density and speed of movement. The system clearly shows the places of increased density, where tensions increase and emergency situations occur: crush and conflicts. By varying the parameters of the area of the event is determined by its optimal configuration for the upcoming event.
')
Using the software system, you can recreate any territory and external conditions, and then place in this environment any group of people. This opens up tremendous research opportunities. You can simulate an earthquake in Japan during a holiday, when there are thousands of people on the streets. Nobody forbids the researcher to replace the Japanese with the British, and the earthquake - to the loss of Manchester United.
But in order for the behavior of a virtual crowd to be reliable, it is necessary to study the logic of human behavior in simulated situations. To do this, conducted "field" studies in crowded places. Research methods are different. The main ones are: observation, photo and video filming, surveys and the use of special sensors (GPS trackers, accelerometers, proximity sensors).
Sergey Ivanov, Head of the International Laboratory "City Informatics" of ITMO University, explained that surveillance cameras at different points in the territory would be enough for the basic work of the models. By highlighting the flow of people from the video series, you can adjust the model to the real situation and make short-term forecasts.
How to evaluate the mood of the crowd, not being in it? Help social networks. After the images with the necessary hashtags or geo-referencing arrive on the server, the model allocates faces to them, according to which machine learning algorithms determine the mood of people.
Proximity sensors — devices that “see” each other at a given distance and fix it — are not less promising. If such a device is tied to a specific person, you can restore the graph of his movements, as well as determine how other factors (distance to relatives or friends) affect their movement in a crowd. However, the use of sensors can be problematic.
“The main disadvantage of this approach is the complex infrastructure preparation of experiments, the need for volunteers to participate in the distribution / collection of special devices and the inability to use the results obtained on-line,” says Ivanov. The scientist adds that the game, nevertheless, is worth the candle - observing individual people allows you to get a deeper knowledge of the nature of the movement of the crowd.
Methods for preparing input data — that is, information about the event area — range from examining photographs to reconstructing three-dimensional objects from a point cloud obtained using drones and scanning tools.
A field study using these tools was recently conducted by ITMO staff together with colleagues from other countries at the Kumbh Mela religious festival, which takes place every 3-4 years in India. During the month, researchers observed a pilgrimage of about 70 million people to the holy places of Hinduism.
Scientists managed to identify stereotyped actions for the participants of the festival and their interrelations. Scenarios included “dead ends”, “obstacles” and “through passages”, as well as crowd behavior in extreme situations, such as heavy rain.
Behavior options were considered in various external conditions. In India, the most important of these was the air temperature, which sometimes exceeded 40 degrees.
A similar study was conducted at the St. Petersburg VK fest this year. The innovation consisted in analyzing photos of the event, published by users, and identifying the most popular areas of the festival based on geo-tags. Most of all people were attracted by the “Drive” zone (an exhibition of tuned cars), the art and food zones were almost keeping up with it.
When the field stage is completed, the information is processed on the computer and then transformed into a model. That's just the usual PC giant amount of modeling work can not do, because each element of the crowd and the external environment is presented in the form of equations and sets of rules. To recreate the typical situation of a mass gathering of people, it is necessary to perform millions of computational operations, so scientists work with the Lomonosov-2 and Lobachevsky supercomputers , one of the most powerful in the world. The first is at Moscow State University, and the second is at Nizhny Novgorod State University.
According to Ivanov, the calculations are seriously complicated by the chaotic movement of human masses, which is reflected in the Monte Carlo modeling approach. “On the one hand, the movement of an individual in a crowd is determined by a physical model that is deterministic, however, the generation of desires (changing the point of interest, choosing the desired speed, unexpected stops, and so on) is a result of a discrete choice that is formalized in terms of random processes”, - says the head of "City Informatics". To obtain reliable results, virtual experiments must be repeated many times with different inputs.
Similarly, scientists have already predicted the possible behavior of fans heading to the subway from the Zenit Arena after the match.
The simulation revealed places with potentially the greatest crowd pressure. To protect the fans leaving the stadium after the match, you can adjust the flow by changing the configuration of the fence or exits from the stadium.
Not a single crowd
Algorithms developed by scientists can predict not only the behavior of people in a crowd. They make it possible to determine the pathways for the spread of viral diseases, draw up a scheme for anesthesia in the city, and even evaluate public opinion — that is, simulate all the processes that are associated with contacts between people and their mobility.
For example, ITMO University is working on modeling the incidence of influenza. The infection process can be illustrated both for confined spaces, such as trains, and for the whole city. In the second case, the task is complicated many times, since it is necessary to take into account all the peculiarities of the city’s life, to reflect all social interactions, while for the train it is enough to describe the stereotypical behavior of passengers, determined by a small number of rules.
Another object of study are social networks. Studying the Vkontakte users posts or comments on the video on YouTube, you can find out the opinion that people share about power, social problems, products and their manufacturers. In particular, the Scientific and Research Institute of NKT of ITMO University studies messages from social networks Vkontakte, Twitter, Instagram and Live Journal. At the same time, scientists can not only analyze opinions and determine “social drafts”, but also distinguish bots from real people, as well as assess the impact on the network and public opinion of statements made by popular bloggers and famous personalities.
At the heart of these technologies is the concept of crawling. In accordance with a given strategy, the program scans the resources of social networks and enters them into a database for further analysis. Each resource, up to individual posts, is determined by a set of characteristics, such as the number of likes, reposts, external citations and comments. All these data are combined by a graph of relationships, on the basis of which resources are ranked according to a given criterion. According to Sergey Ivanov, the approach allows not only to model the dissemination of information on such a graph, but also to form strategies to counter the "manipulation of the network community."
This technology has a wide range of applications. So, in particular, the crawling and analysis of social networks is actively used to study social processes in e-government technologies (eGovernance).
Who is behind this: NII NKT team
NII NKT is rapidly developing , and the results of its research are demonstrated at international scientific conferences, including the recent ICCS'2016 in San Diego, USA. The institute attracts government "megagrants" and foreign experts.
The scientific leader of the Laboratory for Advanced Computational Technologies at the Scientific and Research Institute of NKT is Peter Sloots, a professor at the University of Amsterdam. Also, foreigners work in the laboratory remotely: in Singapore, Poland, Holland - all of them are attracted by Russian developments and the opportunity to participate in promising projects of the Research and Development Institute of NKT.
In 2006, only four employees worked at the Institute. After 10 years, their number has grown to 100, and the institute continues to search for talents. For example, if you are a bachelor and / or a master of applied mathematics and computer science or a related field (STEM - Science, Technology, Engineering and Mathematics), in particular, in the direction of "ICT" and "programming", with English not lower than Upper Intermediate, also:
- You own standard data structures and algorithms;
- You program in C ++ / C # / Java;
- You can use modern math packages and data processing languages (Python / R / MathCAD / MatLab),
then you have the opportunity to join the research institutes of the ITMO ITMO , which are not limited to studying the behavior of the crowd. The Institute plans to apply developments in various fields - from biomedicine and urban planning to forecasting and preventing natural and man-made disasters.
Source: https://habr.com/ru/post/308694/
All Articles