Month: October 2019

Online radicalization is a social phenomenon that has been a popular point of discussion as of late due to recent affairs in the modern political landscape. In light of events such as the Charlottesville protest, Christchurch shooting, and overall rise in hate crimes, many have pointed to major social media platforms in playing a contributing factor. By hosting and promoting extremist fringe content through their recommender systems, platforms like Youtube are becoming fertile grounds for radical ideologues. Caleb Cain is an example of an individual who, as he describes, “had fallen down the alt-right rabbit hole.”  In a New York Times published piece, he speaks of his slow indoctrination into far-right ideology through Youtube’s own recommender system as it would suggest further dubious content for him to view.

In a report from DataSociety by Rebecca Lewis in 2018, she outlines the potential pathways to extremist ideology one might have on Youtube by producing a graph of various political pundits and their relationships through appearances on the same video. Nodes are treated as their respective YouTube channels while the edges represent videos in which both parties contributed. As the algorithm tends to suggest videos from content creators and collaborators that the user has viewed, she lays out potential links between how one could be watching videos made by mainstream political pundits like Ben Shapiro and eventually be lead to viewing content from devout white nationalists such as Richard Spencer.

The findings in this report were further built upon in a study this year titled, “Auditing Radicalization Pathways on YouTube”. Researchers found a strong overlap between the users in comment sections of videos ranging from conservative talk show hosts to alt-right figureheads. To test Youtube’s recommender system, they took a snapshot by beginning at a Youtube channel that belonged to one of their devised community groups while performing a random walk across five suggested channel links and recorded their destination. The researchers found that they would hit a channel belonging to their “Alt-Right” cluster once every five times that they ran their recommender system, meaning one out of a total twenty-five attempts. While the percentages may seem small, they can make a meaningful difference given the scale at which Youtube operates.

Having encountered recommender systems during our discussions about signed networks in lecture, I thought this would be a good opportunity to look into the applications of these systems in the real world. They are no doubt effective at engaging users given how they’re employed in virtually every social media platform in some shape or form. Whether it be for finding products you might enjoy, movies, restaurants, friends and more, these recommender systems do a great job in analyzing a network and giving users more of what already suits their tastes. They are clearly optimized for engagement given the monetary incentives involved. The more time you spend clicking that next video suggestion on Youtube, the more opportunities they have to make money from advertisers. At the cost of only optimizing for engagement, however, you have issues with these recommender systems having unintended consequences.

This brings forward the idea that with recommender systems being so ubiquitous on the web, perhaps the internet isn’t the open sea that people pictured it to be. It might seem, in certain instances, that it’s more prone to forming bubbles instead. The ways in which these recommender systems provide great convenience can be damaging in some aspects as well. It will be interesting to see how companies akin to Google choose to augment their systems, if at all, to address problems such as user radicalization in the future.

Relevant Links:

The Making of a Youtube Radical –
https://www.nytimes.com/interactive/2019/06/08/technology/youtube-radical.html

Alternative Influence: Broadcasting the Reactionary Right on Youtube –
https://datasociety.net/output/alternative-influence/

Auditing Radicalization Pathways on Youtube –
https://arxiv.org/abs/1908.08313

In Wednesday’s lecture as we discussed structure balance and how to balance theorem implies local balance and global collations, we had a couple of graphs of international relations of decades ago, which brought my attention to international relations nowadays.

The article pointed out while U.S. trade relations with China have soured, it is evidential that trade relations between China and Russia are blossoming. China and Russia are increasing their economic, political and military ties which are led by China this time. The leaders signed statements committing to “the development of strategic cooperation and comprehensive partnership” between their nations recently. In other words, they are becoming best friends. It is a good example to demonstrate what we learned in a balanced network – the enemy of the enemy is a friend. If we consider China, Russia, and the US as a triple, China and Russia seemed to have a mutual enemy which is the United States, so that they are looking forward to building a positive relationship which is to confront the sanction from their mutual enemy. It forms a balanced network structure between these three countries: China, Russia forms positive edges and they are both negative to the US.

The Theory of Structural Balance also claims that a friend’s enemy is my enemy. If China represents a faction that antagonizes the United States, it should have negative relations to the countries which are allies of the United States. As one of the most countries associated closely with the United States, Canada had its relations to China inevitably deteriorated after the US-China trade war started. It has been revealed that China has halted its import from Canada and Canada arrested Meng, who is the chief financial officer of China’s telecom giant Huawei at the request of the United States. On the other side, Russia has been imposed sanctions by Canada for a long time ago, demonstrated Russia’s negative relation with Canada. If we consider Canada and the US represent another faction, the relations between these four countries demonstrate the Balance Theorem, that is local balance implies global coalitions.

Applying the Balance Theorem worldwide, we can clearly see how countries get partitioned into different coalitions and countries tend to build close, strong ties within their own coalitions and fewer connections to outside coalitions. International relations today look very different than decades ago since there are too many factors get involved in and form new balanced structures and coalitions. But no matter how it performs, there is an old adage that always applies: “There are no permanent friends or permanent enemies, only permanent interests. “

Relevant Links:

https://www.cnbc.com/2019/09/27/russia-and-chinas-relationship–how-deep-does-it-go.html

https://www.theglobeandmail.com/canada/british-columbia/article-canadian-us-authorities-colluded-arrest-plan-for-meng-wanzhou/

https://www.scmp.com/news/china/diplomacy/article/3027516/chinese-premier-li-keqiang-seeks-manage-expectations-moscow

https://www.armscontrol.org/act/2019-10/news/russia-china-criticize-us-missile-test

https://www.international.gc.ca/world-monde/international_relations-relations_internationales/sanctions/current-actuelles.aspx?lang=eng

Who knew that growing up and making friends and becoming who we are was not totally a random process. Was it possible that all our choices of the people we hang out with or not hang out with were based on homophily, a tendency in people to connect with others similar to themselves? As a child, I do not remember choosing friends based on their similar interests to me, but on how nice they were. So when did homophily start to affect my choices for friends or those who would remain as acquaintances or even strangers?

According to class lectures, homophily does not produce random uniform connections and is present if there are fewer connections between traits than those that are found if the connections were random. Connections between people may be built and developed based on their similar interests, but connections must also be formed as well as not formed based on other factors. As homophily is responsible for connecting people of similar interests it can also be responsible for the segregation that exists in today’s society. Whether it is segregation based on gender, cultural, social status, or one of the many types that people may have experienced, homophily may have played a part in it.

According to the scientific report “Structural transition in social networks: The role of homophily”, one of the aims of the paper was to show that segregation was due to homophily.  The report mentioned that human traits can remain rigid as well can also evolve. The authors of the study chose to keep the human traits rigid to show that social ties affecting homophily are responsible for the segregation that exists.  The report is very scientific with equations and graphs, but I shall mainly discuss the results that are of interest. The report mentioned that although humans have many traits, there are not really any particular traits that are important, that stands out when creating ties. The authors propose that the overlapping of communities of traits is a more realistic model of how society functions. Also that this connection of traits between nodes can become enhanced when the authors took into consideration the mechanism of local attachments, where there is an attachment emotionally by a person to a place.  They also state that homophily and local attachment when combine together shows an increase in segregation.

According to the study, these egocentric networks showed that as in segregation as seen in figure (a) the nodes tend to connect to similar nodes with nearly identical traits with no connection to other communities. When there is a slight connection about 5% between the communities a pattern as shown in figure (b) is seen where communities are still clustered together. In figure (c), it can be seen that as the communities overlap, the nodes become less like their neighbouring nodes as compared to figure (a) where they were similar.  In figure (d), as the features from different communities become more overlapped, the similarities between features of neighbouring nodes decrease. The study also mentioned that a decrease in the similarity of neighbouring features is also due to an increase in the ego’s diversity which is an important feature for a decrease in segregation.

If according to this study, homophily and local attachments together can increase segregation then according to the egocentric networks above it is possible to decrease it.  It is likely that there are some people who would like to make more friends outside of their comfort zones. According to the study, the authors implied that to prevent the effect of segregation in our society people should support and promote cultural diversity.  The report did not mention how to promote cultural diversity, but for me who is now aware of homophily’s existence and living in a country that embraces cultural diversity, I believe that cultural segregation can be tackled.

Learning about and participating in the languages and activities of the different cultures can increase the number of connections between nodes of a variety of traits thus decreasing some evidence of homophily and thus cultural segregation.

Relevent Link

• https://www.nature.com/articles/s41598-019-40990-z

Reference List

Murase, Y., Jo, H.-H., Török, J., Kertész, J., & Kaski, K. (2019, March 13). Structural transition in social networks: The role of homophily. Retrieved October 4, 2019, from https://www.nature.com/articles/s41598-019-40990-z.

The University of Notre Dame and Northwestern university undertook a study in which the social and communication networks of 700 past graduates were analysed. They focused specifically on 3 aspects: network centrality, gender homophily and strength of ties.

• 

• 

• 

Their research revealed that more than 75% of female graduates who had close, same sex friendship groups rose to top roles in their workplace. Essentially, the women who had high network centrality and a close, female inner circle (strong ties with two or three other females) were more likely to be found in higher roles than those less central, and with mainly male close friends. Males with high network centrality were also more likely to be in top-level positions, but unlike women, gender was not found to contribute significantly to this, neither did the strength of these ties.

This research explores a number of themes we have discussed in class. We can first look at the idea of flow of information. Why was high centrality a significant factor in both men and women? Perhaps it has something to do with how we find out about these roles. The more people we know and connect with, the more likely we are to find out about these opportunities.

“It’s not what you know, it’s who you know.”

-Unknown

But why does having close female friends indicate an increased likelihood of high-ranking positions in women? We discussed Granovetter’s idea that structurally embedded edges (close ties within a community) are redundant in terms of access to information. We also discussed the potential pitfalls in homophily, in that it provides us a restricted view of the world if most of our information comes from other already like-minded people. So why does it appear to be playing to our advantage in this case? It appears another factor is at play here, one seemingly unique to women.

I find it fascinating the amount of information about a person that can be found from network analysis. We’ve all heard the saying ‘show me your friends and I’ll tell you who you are’, but network analysis really brings a whole new dimension to that. Now, I don’t know about any of you reading this, but I’d certainly like a high-ranking position. Research is showing my job placement is likely to be 2.5 times greater than those without a large network and close female friends, so I’m off to Facebook to send friend requests to a few hundred people and interact heavily with two to three other females. Forbes list, I’m coming for you.

Side note: of course, all these things need to be taken with a pinch of salt. Yes, correlation doesn’t equal causation, but exploring these correlations to reveal potentially underlying causation is a whole other (but equally fascinating) thing.

---

Relevant Links

• https://www.instylemag.com.au/female-friendships-successful-career-study
• http://www.develop-project.eu/news/strength-of-social-ties-at-workplace
• https://aksakalli.github.io/2017/07/17/network-centrality-measures-and-their-visualization.html
• https://www.researchgate.net/figure/Homophily-social-network-structure-and-causal-mechanisms-A-depiction-of-three-network_fig1_303504899

As one of the most violent cities in the USA, Kansas City realized that it needed to take a different approach to solving it’s crime problem. So, under new government, the Kansas City No Violence Alliance (aka KC NoVA) was created. KC NoVA decided to approach the problem from a network perspective and made the following assumptions:

1. People with friends who have committed crimes, are more likely to commit crimes
2. The closer you are to violence, the more likely you are to be a victim of it
3. Violence is concentrated among groups of people

The first and third assumptions in particular highlight Granovetter’s model of networks, with violent crime primarily occurring in these highly connected clusters of people.

Using these assumptions, their first order of business was to graph the relationships between the gang members.

The connections between people are determined based off of a multitude of different metrics such as traffic stops, arrests, informants information, street intelligence, and more. Using real world information in combination with analysis of the graph, they were able to determine the key players in the gang.

Taking into account whether the individual was on probation or parole, and their betweenness on the graph, police selected two people from each group to reach out to. They warned the individual about violence, and offered targeted support for education, employment, anger management, etc. This method allowed the city to efficiently allocate it’s limited law enforcement resources, and was successful, leading to the number of homicides dropping by more than a quarter next year.

What stood out to me about this article was the proactive approach to crime reduction. Rather than simply reacting to crime, the city was getting out in front of it, not only saving the lives of the would be victims, but also that of the perpetrator. It is truly a win-win scenario and highlights the practical implications of some of the more nebulous concepts of graph theory.

Relevant Links

• KC NoVA Website
• Using SNA to reduce violence in Kansas City: Past Present and Future
• Fight Crime with Social Network Analysis

Recently in one of my Social and Information Network lectures, we discussed the idea of Structural Balance. This was essentially the idea that if a network had signed edges (positive or negative), then we could classify the network as stable depending on the patterns every triangle that was made with the signed edges made.

Since the usual example that was used for our networks was social networks, it was easy to see this representation as a social network of people, where the signed edges either meant two people liked or hated each other. In that case, a triangle with only one positive edge would represent two friends who both hate the person, and a triangle with only one negative edge could represent a love triangle, where two people are competing against each other for the love of the other person.

When I pointed the love triangle case out in class, I got a few laughs. The professor ended up stating that the love triangle case is also not stable. Although it is strange and possibly funny, I do find it interesting that we could use graph theory, and other course content that we learnt in class, to analyze the different structures that may come up when analyzing relationships in a drama for example. Maybe it’s possible to analyze a drama episode by episode, mapping out the relationship network to predict interpersonal conflict between the different characters depending on the stability of the triangles. Since love triangles are a common plot device, maybe you could even use this to predict what happens next episode, if the negative and positive edges indicate the intensity of the relationship between two nodes.

I then decided to search around for articles around this concept and found this one. It’s not super recent (2015), but here they trained a neural network on the TV show “Friends”, (the scenes and the subtitles). From that, they were able to generate this affinity chart.

Admittedly, I’m not 100% sure of how they generate this chart, but it’s neat that tracking this is even possible. I wonder if this can also be used to predict how countries interact with each other. I did find another paper that measures international relations by the use of social media, but given the use of bots on social media, I decided not to look further into it.

Networks and their analysis is one of the core topic of this course for network analysis can prove useful and powerful in various fields such as biology, psychology, computer sciences, social sciences, etc. During the lectures, we saw how network analysis on Facebook data (or any other social media platform) to can help us gain knowledge about friendships and characteristics of social circles on the platforms. We also saw how networks and strong/weak ties can provide us with meaningful insights on job searching/hunting. These are just few of the many uses and applications of the network analysis. Through this blog, I would like to share an interesting application of network analysis which I came across. It is “Combining network analysis with epidemiological data to inform risk-based surveillance: Application to hepatitis E virus (HEV) in pigs.”

This interesting research was conducted by Morgane Salines, Mathieu Andraud and Nicolas Roseab. The goal of the research was to “evaluate the impact of animal movements on pathogen prevalence in farms and assess the risk of local areas being exposed to diseases due to incoming movements (the transportation)” and they combined Network Analysis with epidemiological data to achieve the same. They collected data consisting of pig moments from farms, slaughterhouses, etc. for the period of a year (January – December 2013) from the French Ministry of Agriculture.

            Now comes the interesting part of how they designed or defined a network or a graph. They modelled the data into directed networks which were aggregated on the data for the entire year. The nodes of the network represented the pig holdings, i.e. the farms, slaughterhouses, trade plants, etc, while the movements (all the transportations) between them were represented as directed links. They then calculated various centrality measures such as in-degree, out-degree, betweenness, etc for 178 farms. The graph below shows a sample network where U stands for Unloading pigs and L for loading pigs.

After the preliminary work was finished, the researches then conducted “a univariable analysis to assess the statistical link between each explanatory variable (i.e. the farms’ centrality metrics) and the outcome (i.e. the unbiased within-farm HEV seroprevalence).” PS – HEV seroprevalence was defined as the “number of HEV-seropositive pigs in relation to the total number of pigs sampled in the farm.” After this analysis, they successfully plotted the risk indicator for each of the French departments which looked like this.

Thus, it can be concluded that simple network (and statistical) analysis along with epidemiological data can effectively help the surveillance programs by indicating the risk and thus helping prevent diseases. The network they modeled was very simple and was something that we learnt in the C46 class.

REFERENCE –

MorganeSalinesab, MathieuAndraudab, & NicolasRoseab. (2017, November 20). Combining network analysis with epidemiological data to inform risk-based surveillance: Application to hepatitis E virus (HEV) in pigs. Retrieved from https://www.sciencedirect.com/science/article/pii/S0167587717305937.

In today’s society, people make connections with others from all kinds of backgrounds. Whether it be for business or hobbies, having a connection to share with will provide an individual with extra insight and empowerment. Therefore, it is imperative to learn effective networking skills to expand their social circle. What Does “Mutual Friend” Mean? and Psychological tricks to make people like you immediately are articles that explores the advantages of having mutual friends to establish friendship.

These article lists out 4 areas where mutual friends help in making connections: Identification, Social Networks, Dating and Mutual Friend’s Companions. To summarize, all areas make use of mutual friends to assure the new connection that you have similar traits as your mutual friends and provide a possibility of repeated encounters.

For example, Charlie has known Bob for a long time and talks to Alice on the daily. at some point in time, Charlie will have overlapped time spent with them both. Normally, 2 strangers do not engage in conversation but since Alice and Bob have a mutual friend, they both assume that the other have likable traits and begin to put effort in getting to know each other.

Interestingly, this relates to Social Network’s Triadic closure. Since your mutual friend knows both you and the new connection well, you both are very likely to have a weak or strong connection.

Relevant Links:
Psychological tricks to make people like you immediately
https://dailytimes.com.pk/46941/psychological-tricks-to-make-people-like-you-immediately/

What Does “Mutual Friend” Mean?
https://oureverydaylife.com/mutual-friend-mean-10031289.html

The topic of false and distorted news has been a part of our history for thousands of years. The underlying motive behind the spread of false information is to sway the opinion of people to achieve some goal. In recent times, the term “fake news” has made its way into common everyday vocabulary due to its prevalence in Western culture, particularly in Western politics where certain groups defame political parties to influence the outcome of elections. The cascading effects of fake news in society is extremely relevant to our course material as it is a topic that can be examined through the flow of information networks to understand how fake news becomes viral. This blog post will discuss the following questions: How exactly does fake news spread? Is there a way for social media platforms to eliminate the spread of fake news?

Fake news is spread most effectively through social media platforms, such as Facebook, Twitter, and Instagram. The spread of fake news is typically done through bots, and “According to an estimate in 2017, there were 23 million bots on Twitter (around 8.5% of all accounts), 140 million bots on Facebook (up to 5.5% of accounts) and around 27 million bots on Instagram (8.2% of the accounts)” for a total of 190 million bots on the three platforms combined. The bots reach a large audience by flooding the platforms with false information through trending topics or hashtags, to gain publicity on their posts by being recognized by the platform’s relevance algorithm. With countless bots simultaneously relaying the same false information, naïve audiences often tend to believe propaganda and further spread the lies themselves, creating a cascading effect.

Can social media platforms prevent the spread of false information? Miriam Metzger, a UC Santa Barbara communications researcher says, “Fake news is perfect for spreadability: It’s going to be shocking, it’s going to be surprising, and it’s going to be playing on people’s emotions, and that’s a recipe for how to spread misinformation”. The natural attractiveness of fake news to the human mind means that fake news will always manage to trend in some way. Platforms such as Twitter can use false information detection algorithms to help with reducing the flow of fake news by creating true and false information models and act on their platforms accordingly. However, the issue of fake news arising in social networks will never be eliminated entirely because of the emotional responses generated from them.

Relevant links:

https://www.vox.com/science-and-health/2018/3/8/17085928/fake-news-study-mit-science

https://www.cits.ucsb.edu/fake-news/spread

https://arxiv.org/pdf/1804.08559.pdf

https://www.cits.ucsb.edu/fake-news/brief-history

Zuck: Yeah so if you ever need info about anyone at Harvard
Zuck: Just ask
Zuck: I have over 4,000 emails, pictures, addresses, SNS
[Redacted Friend’s Name]: What? How’d you manage that one?
Zuck: People just submitted it.
Zuck: I don’t know why.
Zuck: They “trust me”
Zuck: Dumb f*cks
– Mark Zuckerberg, CEO of Facebook [1]

That quote is only tangentially related to what I’m going to talk about, but I feel it is not mentioned nearly enough. Social media has been largely blamed for the divide in the United States, and the rise of Nationalism all around the world. While some argue social media giants are either incompetent or complicit in the divisions they are causing, I feel it is far more likely a natural progression of business, social networks, and human nature.

=== It’s all about the Benjamin’s baby! ===

Thanks to draconian laws in the United States, money is an expression of freedom of speech, and companies count as people. In practice, this means Exxon Mobil, with its $30 million dollars in lobbying, gets more votes than you do (which is why the US still has a climate change denial problem) [2]. These allow the “motivated reasoners” to exist. They are people in a social network who have decided (for one reason or another) what they believe, and will not be swayed. They share information (blogs, new articles) constantly, and only share things that promote what they believe in. Since money is an expression of free speech, you can pay anyone to be a megaphone for what you want people to believe, and no one can stop you.
A less important note that is still related to money: social media companies are just that, companies. They need to push for profits, and we the people have spoken with one voice: “give it to us for free or we won’t use it.” As a result, we are the product, and Facebook sell advertisements. Social media needs to push for high user engagement, so they can sell ads that people will see. Alright that covers the business, let’s get to the fun.

=== Your taste is trash ===

As Mark is in need of a 3rd mansion, he needs the company to pad the bottom line. User engagement is the bread and butter, and social media companies have learned the golden rule of grabbing our attention: appeal to our negative emotions [3]. While news in movies will often be “local cat taken down from tree safely,” in reality, there has been a massive shift towards more negative, eye catching headlines. Because these are the article users are more reactive too, and engage most with, algorithms will push these types of news. It is just a natural progression, and not necessarily because they want to watch the world burn, although I wouldn’t put it past them.

=== It’s only natural ===

And now finally the source of this blog post:

When facts fail: Bias, polarisation and truth in social networks[4]

In this 2018 paper published by Cornell University, the researchers set out to find out why we end up so divided. They generated a random graph of “agents,” (people who send and receive information on the social network), with some set of edges connect agents to connect the whole graph. The model looks at a single, simplistic version of a statement, like “global warming is real and man made,” and each agent was assigned a 1 or -1 to start (at random), 1 representing agreement with the statement, -1 disagreement. Some percentage of the people were assigned “motivated reasoners,” they will not change their starting belief. The model then evolved the network overtime in discrete “timesteps”. At each timestep, every agent sends out a signal (1 or -1), to all their neighbours. A person’s belief is now based on a Bayesian representation of the signals they have received. The researchers let it run:

“Our results showed that in every case, motivated reasoners came to dominate the conversation, driving all other agents to fixed opinions, thus polarizing the network.”[5]

It is the only natural progression of social networks to make the truth unobtainable. If you ended up surrounded by motivated reasoners, you would believe whatever they tell you to believe. Keep in mind, real world networks are balance (either weakly or strongly), and in those networks, you already have an established web of trust with new sources or you friends. This is how you end up in an echo chamber, where you only hear opinions you agree with. The paper concludes:

“We show that this simple confirmation bias mechanism can generate permanent opinion polarisation. Furthermore, the model results in states where unbiased agents behave “as if” they were biased, due to their biased neighbours effectively functioning as gatekeepers, restricting their access to free and diverse information.”

A motivated reasoner doesn’t need to convince you, just someone in your coalition, and the network will do the rest for him. The algorithm puts coalitions into like minded groups, pushes dissimilar content to each, and thus a divide grows wider and wider. The algorithm increases engagement with more divisive content, the network owner gets richer, truth becomes perspective, and society falls apart at the seams.

Sources:

[1] Zuck-man-sam quote: https://www.businessinsider.com/well-these-new-zuckerberg-ims-wont-help-facebooks-privacy-problems-2010-5

[2] ExxonMobil wants to watch the world burn:
https://www.scientificamerican.com/article/exxon-knew-about-climate-change-almost-40-years-ago/

[3] Being sad is fun and profitable: https://www.pnas.org/content/116/38/18888

[4] When facts fail: Bias, polarisation and truth in social networks
By Orowa Sikder, Robert E. Smith, Pierpaolo Vivo, Giacomo Livan
26 Aug 2018
https://arxiv.org/abs/1808.08524

[5] Robert E. Smith Op-ed: https://www.azcentral.com/story/opinion/op-ed/2019/09/08/social-media-bias-blame-algorithms/2208612001/