“Know yourself”: social media mining-projects at ITMO University
Social networks are not only a timekiller and a source of news, but also a field for a variety of research. The fact that the scientists and students of ITMO University have learned to learn from our posts on Facebook and Instagram, we will tell below.
/ Photo perzon seo CC-BY
This project was carried out by a group of undergraduates of the Department of Computer Technologies at ITMO University during an internship at the National University of Singapore. The task of the project is to create a system that would help companies better understand the needs of their audience, given that this audience is talking about itself in social networks.
')
Example: a company wants to understand what unites people who are satisfied (or, on the contrary, dissatisfied) with its product. The development of ITMO University students makes it possible to create a portrait of such a [satisfied / dissatisfied] customer based on data from social networks, and to propose a suitable solution: someone may be interested in a discount, and someone may need to offer a different product.
How it works: the algorithm proposed by the ITMO University team analyzes the open data that the user of social networks leaves: geolocation tags, public messages, photos and signatures to them. This data is converted into a set of parameters (average tweet size, the most frequent objects in the photo, the distribution of check-ins, etc.), which is then transmitted to the neural network.
The result of its work depends on the task: in some cases it is necessary to evaluate the emotional coloring of the messages, in others - to make a more complete picture of the chosen type of user and their tastes based on the results of data analysis from several social networks. In this case, it is not about working with individual users, but about the formation of sets of attributes that will characterize entire groups of people.
An earlier development of undergraduates at ITMO University, who worked on a consumer profiling project, is a system that guesses a person’s marital status from Twitter, Instagram and Foursquare. According to the researchers, a combination of three social networks provides prediction accuracy up to 86%, while according to information from only one social network, the algorithm guesses marital status with an accuracy of 17% lower.
True, much in this case depends on who and how declares itself in social networks - for example, Donald Trump's Twitter “informs” that its owner is a bachelor. The researchers suggested that the result is due to the fact that the main work on maintaining the Twitter account is not performed by the president himself, but by his PR manager.
By the way, at the end of this study, ITMO University programmers made a presentation at the AAAI Conference on Artificial Intelligence (AAAI-2017), which brings together scientists involved in the creation and training of AI.
Another example of the study, which is based on data from social networks, is a system for analyzing emotions, developed by employees of the Scientific-Research Institute of High-Tech Computer Technologies (NII NKT) of ITMO University.
How it works: the facial expressions in the photographs are compared with the eight basic emotions - it is anger, neglect, neutrality, sadness, surprise, happiness and sadness. The basis of the algorithm of recognition of emotions also are convolutional neural networks. To improve the accuracy of the algorithm, researchers can use the results of the evaluation of the photos “manually” by the users themselves.
At the same time, the algorithm recognizes not only the “picture”, but also the emotional mood of text messages: in the initial study, data from social networks Facebook, Twitter and Vkontakte were used, the visual and text component of the posts was evaluated comprehensively. In addition, researchers can set "clarifying conditions" - for example, evaluate materials with reference to a specific location.
The result of the work, as well as in the case of profiling accounts, depends on the current research task: for example, the first launch of the system in the Sirius children's camp allowed us to create an innovative stand - according to the materials of users of social networks, he showed what kind of emotions people have object or place on the map within Sochi, Krasnaya Polyana and Adler.
Another example of the use of the algorithm is the ability to recognize the hot spots at football matches (what other developments of the ITMO University were used in preparing for the Confederations Cup and the upcoming 2018 World Cup, we told here ).
In this case, the researchers limited themselves to the Instagram social network and analyzed the photos that fans published during matches. It turned out that in the photo from the matches, during which the fights were recorded, the algorithm is twice as likely to find "evil fans" (compared to matches, during which there were no fights).
Another development of the Scientific and Research Institute of NKT using data from social networks is a system for searching for “social drafts” and evaluating public opinion. This project allows, in accordance with a given strategy, to analyze publicly available materials from users of Vkontakte, Twitter, Instagram, Live Journal, identify the most important informative elements, subject and emotional mood of the message and mark the post in accordance with the specified characteristics, including the number of likes, reposts, citations , comments, etc.
This approach allows you to build a graph of links according to the chosen criterion and predict the distribution of information on such a graph. The system is able to distinguish real users from bots and can evaluate the user's “social weight” - including whether it is the opinion leader in a particular social group.
Social networks can not only tell about the points of tension in society and the existing problems. No less often, users share with their friends what they inspire, interest, or just smile. Therefore, for the visitors of the VK Fest and Geek Picnic festivals, ITMO University often prepares a demonstration of its developments in a non-classical - entertainment format.
For example, a study on the analysis of emotions was demonstrated at VK Fest last year - visitors who came to the 300th Anniversary Park at that time could see the “interactive picture” of the festival: the most interesting places, the most popular photo sites, etc. All information naturally changed in accordance with what Vkontakte users told about the festival on their pages.
Richard Dawkins. Photo of ITMO University
And for the Geek Picnic-2017 festival, ITMO University prepared another example of the analysis of social networks - memokub with the most recognizable memes: from “Wat” to the well-known frame from “The Great Gatsby”. An autograph session of the festival's headliner, evolutionary biologist and popularizer of science Richard Dawkins took place next to him.
By the way, at the Geek Picnic festival, ITMO University brought not only the Memokub, but also many of its developments - from the work system of a smart home to functional food. About what else the visitors of the Geek Picnic remember the University’s interactive zone - read here .
/ Photo perzon seo CC-BY“What the buyer wants”: profiling in social networks
This project was carried out by a group of undergraduates of the Department of Computer Technologies at ITMO University during an internship at the National University of Singapore. The task of the project is to create a system that would help companies better understand the needs of their audience, given that this audience is talking about itself in social networks.
')
Example: a company wants to understand what unites people who are satisfied (or, on the contrary, dissatisfied) with its product. The development of ITMO University students makes it possible to create a portrait of such a [satisfied / dissatisfied] customer based on data from social networks, and to propose a suitable solution: someone may be interested in a discount, and someone may need to offer a different product.
How it works: the algorithm proposed by the ITMO University team analyzes the open data that the user of social networks leaves: geolocation tags, public messages, photos and signatures to them. This data is converted into a set of parameters (average tweet size, the most frequent objects in the photo, the distribution of check-ins, etc.), which is then transmitted to the neural network.
The result of its work depends on the task: in some cases it is necessary to evaluate the emotional coloring of the messages, in others - to make a more complete picture of the chosen type of user and their tastes based on the results of data analysis from several social networks. In this case, it is not about working with individual users, but about the formation of sets of attributes that will characterize entire groups of people.
What Donald Trump Twitter Talks
An earlier development of undergraduates at ITMO University, who worked on a consumer profiling project, is a system that guesses a person’s marital status from Twitter, Instagram and Foursquare. According to the researchers, a combination of three social networks provides prediction accuracy up to 86%, while according to information from only one social network, the algorithm guesses marital status with an accuracy of 17% lower.
True, much in this case depends on who and how declares itself in social networks - for example, Donald Trump's Twitter “informs” that its owner is a bachelor. The researchers suggested that the result is due to the fact that the main work on maintaining the Twitter account is not performed by the president himself, but by his PR manager.
By the way, at the end of this study, ITMO University programmers made a presentation at the AAAI Conference on Artificial Intelligence (AAAI-2017), which brings together scientists involved in the creation and training of AI.
Emotions and safety: how to feel the mood of the crowd
Another example of the study, which is based on data from social networks, is a system for analyzing emotions, developed by employees of the Scientific-Research Institute of High-Tech Computer Technologies (NII NKT) of ITMO University.
How it works: the facial expressions in the photographs are compared with the eight basic emotions - it is anger, neglect, neutrality, sadness, surprise, happiness and sadness. The basis of the algorithm of recognition of emotions also are convolutional neural networks. To improve the accuracy of the algorithm, researchers can use the results of the evaluation of the photos “manually” by the users themselves.
At the same time, the algorithm recognizes not only the “picture”, but also the emotional mood of text messages: in the initial study, data from social networks Facebook, Twitter and Vkontakte were used, the visual and text component of the posts was evaluated comprehensively. In addition, researchers can set "clarifying conditions" - for example, evaluate materials with reference to a specific location.
The result of the work, as well as in the case of profiling accounts, depends on the current research task: for example, the first launch of the system in the Sirius children's camp allowed us to create an innovative stand - according to the materials of users of social networks, he showed what kind of emotions people have object or place on the map within Sochi, Krasnaya Polyana and Adler.
Another example of the use of the algorithm is the ability to recognize the hot spots at football matches (what other developments of the ITMO University were used in preparing for the Confederations Cup and the upcoming 2018 World Cup, we told here ).
In this case, the researchers limited themselves to the Instagram social network and analyzed the photos that fans published during matches. It turned out that in the photo from the matches, during which the fights were recorded, the algorithm is twice as likely to find "evil fans" (compared to matches, during which there were no fights).
For extreme situations, in addition to fights, the use of pyrotechnics, throwing objects at the stands and the stadium field, demonstrating offensive banners, chanting intolerant chants, and more were also considered. We considered matches [FC Zenit] in the period from 2013 to 2015. It turned out about ten games, after which we took the other teams.
There were only a few matches with fights, but when we started working with the official statistics of the Russian Football Union and watched not only fights, but also other events that could threaten the health and well-being of the fans, [it turned out] for three seasons out of 700 matches with extreme events in the stands about half of the games.
- Vasily Boichuk, engineer, Research Institute
Social drafts and dissemination of information
Another development of the Scientific and Research Institute of NKT using data from social networks is a system for searching for “social drafts” and evaluating public opinion. This project allows, in accordance with a given strategy, to analyze publicly available materials from users of Vkontakte, Twitter, Instagram, Live Journal, identify the most important informative elements, subject and emotional mood of the message and mark the post in accordance with the specified characteristics, including the number of likes, reposts, citations , comments, etc.
This approach allows you to build a graph of links according to the chosen criterion and predict the distribution of information on such a graph. The system is able to distinguish real users from bots and can evaluate the user's “social weight” - including whether it is the opinion leader in a particular social group.
Punches, sketches, vains
Social networks can not only tell about the points of tension in society and the existing problems. No less often, users share with their friends what they inspire, interest, or just smile. Therefore, for the visitors of the VK Fest and Geek Picnic festivals, ITMO University often prepares a demonstration of its developments in a non-classical - entertainment format.
For example, a study on the analysis of emotions was demonstrated at VK Fest last year - visitors who came to the 300th Anniversary Park at that time could see the “interactive picture” of the festival: the most interesting places, the most popular photo sites, etc. All information naturally changed in accordance with what Vkontakte users told about the festival on their pages.
Richard Dawkins. Photo of ITMO University
And for the Geek Picnic-2017 festival, ITMO University prepared another example of the analysis of social networks - memokub with the most recognizable memes: from “Wat” to the well-known frame from “The Great Gatsby”. An autograph session of the festival's headliner, evolutionary biologist and popularizer of science Richard Dawkins took place next to him.
By the way, at the Geek Picnic festival, ITMO University brought not only the Memokub, but also many of its developments - from the work system of a smart home to functional food. About what else the visitors of the Geek Picnic remember the University’s interactive zone - read here .
Source: https://habr.com/ru/post/333962/
All Articles