Web 2.0

What is common between Sigmund Freud, The Dalai Lama, Karl Marx, Aristotle Onassis, Bob Marley, Albert Einstein, Marlene Dietrich, Madeleine Albright, Victor Hugo, Frédéric Chopin and Andy Garcia. Besides the fact that I am personally influenced by many of them, the one common factor is that they were all refugees during some phase of their lifetime. The term refugee is used for those who seek relief and refuge from economic, military, political, or social distress including war, famine, or civil strife. The earliest known instance of refuges was triggered by the invasion of Middle East by Sargon the Great of Mesopotamia (2270-2215 BC). Across the years, human conflicts, battles, wars and epidemics has been the root cause of human displacement. According to the UN High Commissioner for Human Rights, a refugee is "any person who: owing to a well-founded fear of being persecuted for reasons of race, religion, nationality, membership of a particular social group, or political opinion, is outside the country of his nationality, and is unable to or, owing to such fear, is unwilling to avail himself of the protection of that country". Refugees are spread around the world, with more than half in Asia and about 20 percent in Africa. United Nations High Commissioner for Refugees (UNHCR) numbers indicate that by the end of 2011, number of forcibly displaced people worldwide exceeded 42.5 million. Out of this, 15.2 million were refugees (10.4 million under UNHCR's mandate, 4.8 million Palestinian refugees), 895,000 asylum-seekers and 26.4 million internally displaced persons.

According to UN Secretary-General Ban Ki-moon, "Refugees have been deprived of their homes, but they must not be deprived of their futures."

We are interested to understand refugee movement across countries over the past decade. Since we can model origin and destination countries as nodes and people movement as edges, we use social network analysis and visualisations for our purpose. Data about refugee movement across countries during 2000 - 2011 was obtained from UNHCR.

The below network graphs and charts trace annual as well as consolidated refugee movements from 2000-2011. Our consolidated dataset had 211 nodes and 6068 edges, with the nodes representing countries (source of refugees) and edges representing cross country refugee movements. Table 1 gives details of the number of nodes and edges over the years. Figure 1 is a chart of global refugees from 2000-2011.

Table 1: Nodes and Edges

Figure 1: Global refugees from 2000-2011  

Figure 2 is a source and destination country chart of the top ten global refugee movements from 2000-2011. The largest refugee movements were from Afghanistan to Pakistan (18,790,596), which accounted for 15% of the total, followed by Afghanistan to Islamic Republic of Iran (12,423,106 - 10%), Iraq to Syrian Arab Republic (6,146,922 - 5%), Burundi to United Republic of Tanzania (4,030,134 - 3%), Vietnam to China 3,591,405 - 3%), followed by others.

Figure 2: Country-wise global refugee movements from 2000 - 2011 

We list below some key network graphs. The first set of graphs are based on aggregated data over the years 2000 - 2011 and the second set of graphs are based on annual data from the years 2000 and 2011. We have not included results for the other years for the sake of keeping this a short blog. Figures 3 & 4 are network graphs of refugee Inflows and Outflows from 2000-2011. The network graphs are based on weighted indegree as well as outdegree measures. The top ten refugee origins/destinations are also listed.   

Figure 3: Network graphs of refugee Inflows from 2000 - 2011

  Top 10: Pakistan, Islamic Rep. of Iran, Germany, Syrian Arab Rep., United Rep. of Tanzania, United States, Kenya, China, United Kingdom, Jordan

Figure 4: Network graphs of refugee Outflows from 2000 - 2011

Top 10: Afghanistan, Iraq, Somalia, Sudan, Dem. Rep. of the Congo, Burundi, Vietnam, Occupied Palestinian Territory, Angola, Myanmar

Figure 5 has network graphs of refugee flows during 2011. The network graphs are based on weighted indegree as well as outdegree measures. The top ten refugee origins/destinations are also listed. 

Figure 5: Network graphs of refugee flows during 2011

Figure 6: Network graphs of refugee flows during 2000 

Posted by Dr. Jai Ganesh at 5:49 AM | Permalink | Comments (0) | TrackBacks (0)

According to musicologist Jean-Jacques Nattiez (Musicologie générale et semiology, 1987), the border between music and noise is always culturally defined and composer John Cage thought that any sound can be music and has said "There is no noise, only sound." Music has been an important part of many cultures and there are several hundred popular music genres and subgenres. Needless to say, the origin and style of many musical genres and subgenres can be traced to other genres and subgenres. Let us explore further the various musical genres and their stylistic origins. The below graphs trace the various genres and their stylistic origins according to data from Dbpedia/Wikipedia. The following are some caveats with regard to the data and methodology:

• The data in Wikipedia is not complete and the completeness is limited by the topics and updates made by the users. For example, several popular Indian genres such as Ghazal, Qawwali, Bhajan, Lavni, Kirtan, Thumri, Tarana, Kajri, Dhrupad etc., lack rich RDF representations. Similar scenario exists for several African, Asian and Latin American genres. This is a case for creating rich Wikipedia entries with regard to these music genres.
• The data is spread across musical genres over several generations and we do not have the specific timeline of the stylistic origin of the genres. Therefore the genre which is influenced by a stylistic origin during a certain time period could in turn be the stylistic origin of another genre during another time period. For example, the stylistic origin of Electro can be traced to Funk and Funk in turn owes its origin to Jazz.
  There could be stylistic origins and genres which could be closely related, but not captured in Wikipedia.
• We explored only the musical genres listed in the English language section of Wikipedia, which in turn could have been limited by the extent of English language penetration and usage in various countries.
• Another limiting factor for global representation in Wikipedia could be the penetration of Internet and computers.

A musical genre could be associated with more than one stylistic origin. For example, the stylistic origins of Hip Hop can be traced to Rhythm and blues, Reggae, Funk, Disco, Dancehall, Scat singing etc. Our graph takes in the musical genres and stylistic genres as nodes and the edges represent their relationships. Our data has 1058 nodes and 2767 edges. Let us go over to the graphs and results for various centrality measures. 

Overview Graph of Music Genres and their Stylistic Origins

Authority 

Betweeness

Closeness 

Degree 

Hub

Indegree

Outdegree
 

Observations & Conclusions

Wikipedia entries with respect to Musical genres and their stylistic origins follow power-law degree distribution and exhibit characteristics of Scale Free Networks. In other words, the nodes are not evenly connected and have several highly connected nodes which govern the property of the network. Past research by Albert-László Barabási and colleagues found that World Wide Web links have scale free network characteristics. Similar results have been proposed for biological networks, and social networks. It will be interesting to see if similar characteristics can be found for other Wikipedia entries. 

Degree Distribution

We observed 30 prominent clusters and 17 smaller clusters (2 or 3 entities).

The table below lists the most important nodes of the network for the selected centrality measures (in-degree and out-degree centralities). In Degree Centrality of a node are the connections that the node receives from other nodes. For example, in our result, the top five Indegree centrality measures belong to Girl_group, Dark_psytrance, Breakcore, Boy_band, New_Age and Chillwave. In other words, these genres draw their origins from a large number of other musical genres. Out Degree Centrality of a node are the connections that the node sends to other nodes and the top five in our study are Hip_hop, Punk_rock, Jazz, Pop and Funk, which means that these musical genres have resulted in the origin of the largest number of other genres. Closeness Centrality measures indicate the influence of several regional as well as country specific genres and their origins such as  Kansas City Jazz, Music of India, France, Folk music of England etc. 

Posted by Dr. Jai Ganesh at 7:51 AM | Permalink | Comments (0) | TrackBacks (0)

Board of directors consists of a group of elected members who oversee the activities of a company. They are representatives of the shareholders and play an important role in decisions on major company issues including dividends, mergers and acquisitions, senior appointments, compensation, etc. Board of directors can be modelled as a two mode network and their membership in multiple company boards can be modelled as an interconnected network with the board memberships acting as the tie. One of the means of representing the strength of the tie is the number of board memberships.

In this research, we model the board membership of the top 100 most valuable companies in India. We collected the list of Top 100 Companies in India by Market Capitalization listed on BSE. We then looked up their board memberships and analysed the network. The networks below shows board membership of three organisations in our sample. The red nodes are the companies and the black nodes are members of the board.

We grouped the sample of 100 companies and saw the emergence of 46 clusters. The figures below give details of the clusters as well as their interconnections through board memberships. The links show board membership ties between the members.

The network graphs below are varying visualisations of companies and their board membership.

 
The table below lists the most important nodes of the network for the selected centrality measures (in-degree and out-degree centralities). In Degree Centrality of a node are the connections that the node receives from other nodes. For example, in our sample Dr. Omkar Goswami has the highest in-degree centrality measure as he is on the board of 6 companies in our sample set of top 100 companies in India by market capitalization listed on BSE. Out Degree Centrality of a node are the connections that the node sends to other nodes. Larsen and Toubro has the highest out-degree centrality measure as the company has 28 board members. 

This was a simple illustration of the network of key decision makers of top 100 companies in India by market capitalization listed on BSE. The network between the board members of larger sample sets could get denser.

Posted by Dr. Jai Ganesh at 4:47 AM | Permalink | Comments (0) | TrackBacks (0)

Social Media sources have played a key role in disaster reporting, relief and rescue efforts. Social media destinations such as Twitter, Flickr and Facebook were leveraged extensively to spread information during the Victorian bushfires in Australia in 2009. In this blog, we try to explore the role played by government, civic authorities, law and order, general public, celebrities, activists, journalists and media in the advent of two natural disaster situations: Hurricane Sandy and Cyclone Nilam.

The terms hurricane and typhoon are regional names used to refer to a strong tropical cyclone. Once the tropical cyclone reaches winds of at least 17 m/s, they are called a tropical storm and are assigned a name. If winds reach 33 m/s, then they are called Hurricane in North Atlantic ocean, Eastern pacific and South Pacific Ocean, Typhoon in Western pacific and Cyclone in Southern pacific and Indian ocean. In the recent past, we experienced two significant simultaneous cyclones: hurricane Sandy in the East Coast of US and Cyclone Nilam in the East Coast of India. Hurricane Sandy is the largest Atlantic hurricane on record, as well as the second costliest Atlantic hurricane only surpassed by Hurricane Katrina in 2005. Hurricane Sandy struck in late October 2012. In the United States, Sandy caused severe damage in New Jersey and New York. It has claimed more than 50 lives, left millions without power and caused over US$ 50 billion in damage in the United States. Damages to life and property are spread across Jamaica, Haiti, Dominican Republic, Puerto Rico, Cuba and The Bahamas. Cyclone Storm Nilam, which struck India in late October 2012 caused damages across Indian states such as Tamil Nadu, Andhra Pradesh, Karnataka and Odisha.

According to Semiocast, the United States has 141.8 million Twitter users, and India has over 15 million Twitter users. This depicts a healthy Twitter user base in the United States and a fast growing Twitter user base in India. We collected tweets containing the hashtag #Sandy from Oct 30th until Nov 6th. We collected about 500,036 tweets from 306,348 users. We also collected about 1500 tweets with the hashtag #Nilam during 31st Oct 2012. We classified #Sandy tweets into four different sets based on the time interval when the tweets occurred. Time Interval 1 has 180,489 Nodes and 232,578 Edges, Time Interval 2 has 60,351 Nodes and 65,842 Edges, Time Interval 3 has 31,033 Nodes and 31,039 Edges, Time Interval 4 has 14,550 Nodes and 13,350 Edges. The network graphs, tables and charts below unravels the role of social media in disaster reporting, relief and rescue efforts by government, civic authorities, law and order, general public, celebrities, activists, journalists and media.

Hurricane Sandy: Time Interval 1 

Hurricane Sandy: Time Interval 2
 

Hurricane Sandy: Time Interval 3
 

 Hurricane Sandy: Time Interval 4

 Hurricane Sandy: Distribution of most important Tweeters based on Network Centrality measures 

Meanwhile in some other part of the graph, we observed the following clusters. 

Cyclone Nilam

Conclusions

1. Sandy was 'Instagrammed': Number of #Sandy tweets originating from Instagram ranked fourth, preceded only by Tweets originating from iPhone, Web and Android and in the case of Nilam, Twitpic scored high. This is an interesting social behavior where in disaster scenarios, visual means of message propagation assume prominence. This is in contrast to our observations with regard to message propagation during the Olympic Games, where Tweets originating from Instagram scored very low.

2. Government and officials leveraging Social media: In the case of Hurricane Sandy, @NYGovCuomo, which is the Official Twitter account for the Governor of New York State, Mr. Andrew Cuomo consistently emerged as one of the most important entities. No such trends in the case of Cyclone Sandy where we did not see even a single tweet from a government official or utility.

3. News Media leveraging Social Media: Journalists and News Media extensively leveraged Social Media for news propagation and for news amplification. News Media sources and Journalists were the heaviest users of Social media in the events of both Sandy and Nilam, supporting the findings from previous research. Individual Journalists leveraged their Twitter follower base to spread messages faster and they invariably figured higher up in the network centrality measures as against the twitter accounts of Media houses which employ them.

4. Cyclone Nilam saw a low overall coverage in Social Media in India and surrounding affected regions.

5. Activist groups such as Anonymous etc. leveraged social media for message propagation in the case of Hurricane Sandy.

6. Celebrities in the United States were active in social media during Hurricane Sandy as against celebrities in India who chose to ignore Cyclone Nilam.

Posted by Dr. Jai Ganesh at 10:31 AM | Permalink | Comments (0) | TrackBacks (0)

The third Presidential Debate concluded just a few hours back. The two network graphs visualize the entire debate content. The table below lists the most influential keywords used by the two presidential contestants.

President Obama

 Mitt Romney

Posted by Dr. Jai Ganesh at 4:58 AM | Permalink | Comments (2) | TrackBacks (0)

Presidential Debates in the United States occur between the two main candidates of the largest parties. Critical topics are discussed in the debate and these debates have the tendency to sway general public. According to Nielsen Ratings, the second Presidential debate on the 16th Oct 2012 had an estimated viewership of 65 million viewers. This is a slight dip over the first debate which attracted viewership of over 67 million viewers. The table below lists the most influential keywords used by the two presidential contestants. The two network graphs visualize the entire debate content.

President Obama

Mitt Romney

Posted by Dr. Jai Ganesh at 6:53 AM | Permalink | Comments (1) | TrackBacks (0)

State of the Union Messages to the Congress are mandated by Article II, Section 3 of the United States Constitution which states, "He shall from time to time give to the Congress information of the state of the union, and recommend to their consideration such measures as he shall judge necessary and expedient". Since 1790 State of the Union messages have been delivered regularly at approximately 1 year intervals (Source: The American Presidency Project http://www.presidency.ucsb.edu/sou.php#axzz274c0J8EF ). The State of the Union Message is delivered near the beginning of each session of Congress and it is a report on the condition of the country and a platform for presidents to outline their legislative agenda and their priorities. They have over the years emerged as a communication between the president and the people of the United States. The messages are available at http://www.presidency.ucsb.edu/sou.php#axzz274c0J8EF. We are interested to examine what President Barack Obama and past President George Bush have been stressing in their speeches. We used Texttexture (http://textexture.com/) to visualize the State of the Union Messages by President George Bush (2005-08) and President Barack Obama (2009-2012). The network visualisations and results from Texttexure are given below. Each network has an average of 100 words (nodes) and about 1000 edges (co-occurrences). In the network graphs below, the words are the nodes and their co-occurrences are the connections between them.

Posted by Dr. Jai Ganesh at 7:24 AM | Permalink | Comments (0) | TrackBacks (0)

The Paralympic games have come a long way from the earliest event in 1960 which was held in Rome exclusively for war veterans and attracted participation of 400 athletes from 23 countries. The London Paralympic games 2012 saw participation of 4,200 athletes from 147 countries, taking part in 21 sports. The International Paralympic Committee has defined six disability categories across physical and intellectual disabilities. This are: Amputee, Cerebral Palsy, Intellectual Disability, Wheelchair, Visually Impaired and Les Autres (athletes who do not fall under the other five categories). In this blog we will explore how Paralympic games were tweeted. We collected Twitter data with the hashtag #paralympics via 140kit.

A sample of 82,560 tweets across 54,522 users was collected between Aug 30th and Sept 6th 2012. The table below lists the overall statistics of the top ten measures with regard to language, source application from where the Tweet originated, location and hashtag. The network graph was created of the entire network of users who retweeted as well as mentioned one another. In this dataset, there were 48,956 nodes and 38,959 edges. There were 42,881 retweets, 229 mentions, 546 average followers, 354 average friends and 113 average favourites. The figure below gives the network graph of the network of users created using Gephi. Each dot in the graph is a network node and corresponds to the Twitter account of a user. The links are the edges and corresponds to the users who retweeted one another or those who mentioned one another. 

Overall Statistics

 Network Graph 1 

 Network Graph 2  

English was the language of choice for people tweeting about the Paralympic games making up about 97% of the sample. An exact similar trend was observed in an earlier study using Olympic tweets. This was followed by other languages such as Dutch, Spanish, German, French and Swedish. Tweets arising from Twitter for BlackBerry constituted about 27% of the total tweets followed by Web based Tweets (27%), Twitter for iPhone (15%), Twitter for Android (13%) followed by the rest. This indicates that more people were tweeting about the Paralympic Games through mobile devices. Most of the tweets are arising from Europe, followed by North America. London leads the list, followed by Amsterdam, Casablanca and Hawaii. Analysing the hashtags used reveals that #teamGB is the most used followed by #paralympics, #London2012 and #inspirational. Team Great Britain seems to be the most popular team.

Posted by Dr. Jai Ganesh at 5:24 AM | Permalink | Comments (0) | TrackBacks (0)

The global appetite for medicine and medications continue to grow at an alarming pace. According to sources, the global pharmaceutical sales is growing at the rate of 6% YoY and sells about $800 billion worth  of drugs globally. Populous markets such as India and China are contributing to a significant part of this growth as these markets are growing at about 15% YoY. The United States is the single largest pharmaceutical market contributing to sales of about $320 billion annually. Prescription drug abuse is the fastest growing drug problem in the United States and in 2007, approximately 27,000 unintentional drug overdose deaths occurred in the United States, one death every 19 minutes (Centers for Disease Control and Prevention). Americans consume more than three times the per capita rate of consumption of prescription drugs than countries such as Germany. Similar scenarios exist for livestock drugs as well.

Given the above context, let us examine further what exactly constitutes the drugs we consume. The motivation for this research is to understand from a network perspective the drugs we consume and their active ingredients network. We use social network analysis methods to visualize and analyse drugs listed in the US FDA Drug Label Data. The large quantum of information available in this database gives us a glimpse of the fundamentals of drugs and their active constituents. The co-occurrence of active ingredients in thousands of drugs provides information about which active ingredients are used frequently, which of them are used together and which of them occur together most often. We analyse the network structure of drugs and their active ingredients, examine their centrality measures and other key indicators.
The data source of our analysis is the US FDA Drug Label Data provided by the Department of Health & Human Services. The agency collects this data on a daily basis and we downloaded this data on Aug 30th 2012. Our dataset had 70,904 National Drug Codes. The dataset contains data elements such as proprietary name, active ingredients, marketing application number or regulatory citation, National Drug Code, and company name. Of particular interest to us is the drug's Proprietary Name and its Ingredients as these two form the nodes of our two-mode network.

The final cleaned up data had 16,444 nodes and 32,627 edges. The network analysis and network graphs were created using Gephi. As seen from Figure 1, there were several clusters around drugs and their active ingredients. Figures 2 to 8 are close-up screenshots of various clusters.

Figure 1: Network graph of Drugs and their Active Ingredients

Figure 2: Close-up of a cluster in the Network graph

Figure 3: Cluster around Octinoxate and Titanium Oxide

Figure 4: Crowded part of the network graph 

Figure 5: Cluster around Potassium Cation

Figure 6: Cluster around Sodium Flouride

 Figure 7: Cluster around Alcohol

 Figure 8: Cluster around Triclosan

The resulting associated network statistics are given below. Table 1 lists the top 10 drugs and their active ingredients based on network centrality measures such as Degree, Closeness, Page Rank, Betweeness and Eigenvector. The measures are listed in the decreasing order of their scores. The shaded blocks are the drugs with Proprietary Names which had high scores.

Table 1: Top 10 Drugs and their Active Ingredients listed in decreasing order of network centrality measures

In our case the Eigenvector measures indicates the strength of interconnections between the strongly connected active ingredients. The highly connected nature of active ingredients such as Octinoxate, Titaniumdioxide, Octisalate, Oxybenzone, Avobenzone depicts their widespread usage in key drugs as well as their widespread customer adoption. Degree centrality in our case depicts the drugs and active ingredients which are most connected to other active ingredients.  Betweenness Centrality for us is a measure of the drugs and active ingredients which are best positioned to broker connections between other drugs and active ingredients or serve as connecting node between them.

Let us next examine what these drugs are used for. We looked at various sources such as Drugs.com, Wikipedia etc. to understand what these drugs and ingredients are used for. The blocks coloured in light blue refers to veterinary medicines or active ingredients used in veterinary medicines. The blocks coloured in light green refers to proprietary names of drugs.

 Table 2: Drugs, Active Ingredients and their end consumer usage 

Most of the network measures indicate the dominance of active ingredients such as Octinoxate, Titaniumdioxide, Acetaminophen, Octisalate, Oxybenzone and Avobenzone which are used in sunscreens, lip balms, food colouring, cosmetics etc. This is followed by active ingredients which make up the generic base for drugs such as Alcohol. Other key leaders are Triclosan which is an Antibacterial and antifungal agent used in many consumer products. There is a strong presence of drugs as well as active ingredients used in veterinary medicines as seen from the high ranking of several drugs with proprietary names. The data source of our analysis is the US FDA Drug Label Data provided by the Department of Health & Human Services. Since these drugs are sold globally either in same form or as generics, it is safe to assume that similar trends exist for other markets as well.

Posted by Dr. Jai Ganesh at 8:26 AM | Permalink | Comments (0) | TrackBacks (0)

The Olympic Games is the largest sporting spectacle of the planet. The games have come a long way from the ancient Olympic Games which were held in Olympia, Greece during the 8th century BC to the 4th century AD to the ongoing London 2012 games. The London games have attracted participation from 205 countries and over 10,500 athletes. The games started on 27 July and will continue until 12 August 2012. According to The Economist, the British government's budget for the games has risen to £9.3 billion ($14.5 billion) from an initial estimate of £2.4 billion. The eleven global partners of the games are Acer, Atos, Coca-Cola, Dow, General Electric, McDonald's, Omega, Panasonic, Procter & Gamble, Samsung, Visa. In addition there are the Olympic Partners, Olympic supporters and Olympic Providers and suppliers (see http://www.guardian.co.uk/sport/datablog/2012/jul/19/london-2012-olympic-sponsors-list). Given the global nature of the games and the large sums spent by the sponsors, it will be interesting to understand and analyse the ground level interest and support towards the games. Since social media is proxy for end user behaviour, we set out to understand the world's social media behavior during the largest sporting event. We collected Twitter data with the hash tags #olympics, london olympics, londonolympics, olympics2012. The data was obtained from the Twitter data aggregator 140kit (www.140kit.com).

A sample of 43,048 tweets across 27,683 users were collected between July 29th and July 31st. The table below lists the overall statistics of the top fifteen measures with regard to language, source application from where the Tweet originated, location and hashtag. The network graph was created of the entire network of users who retweeted as well as mentioned one another. In the dataset, there were 16,444 nodes and 15,233 edges. There were 16,190 retweets, 274 mentions, 916 average followers, 368 average friends and 135 average favourites. The figure below gives the network graph of the network of users. The network graph was created using Gephi.

English was the language of choice for people tweeting about the games, making up about 97% of the sample. Spanish and French took silver and bronze respectively. Web based Tweets constituted about 23% of the total tweets followed by iPhone (20%), Mobile Web (13%), Android (12%), Blackberry (11%), followed by the rest. This indicates that more people were tweeting about the Olympic Games on the go than from home or work. Following a global trend, most users have not revealed their location and where the user location is mentioned, London leads the list. Analysing the hashtags used reveals that gymnastics is the most popular sport followed by cycling, basketball, swimming, equestrian, rowing, archery and judo. Finally Team Great Britain seems to be the most popular team. Are the Olympics really a global phenomenon when we examine it from the perspective of social media mentions? The initial results show concentration across the US, Europe and Australia. We will come back to the same analysis once more post the conclusion of the games.

Jai Ganesh

Sheera L. Gendzel, MBA Candidate, ESADE Business School 

Posted by Dr. Jai Ganesh at 10:14 AM | Permalink | Comments (1) | TrackBacks (0)