Thursday, December 12, 2013

Contribution Index Case Study

Today Michi Henninger told me about an overview description and case study of the contribution index, analyzing the km4dev knowledge sharing community for development. Good overview of how to use contribution index to analyze a COIN in the making.

Tuesday, October 15, 2013

BMW i8 vs. Porsche 918 – which is cooler?

This is a simple example of Coolhunting over the internet. To start with, I would like to thank Professor Gloor for inviting me to write on Swarm Creativity blog. This was a study I did for the Collaborative Innovation Networks class in MIT Sloan, which was taught by Professor Gloor. The goal is to use variety of tools offered by Condor to compare people’s perception of products, and understand the difference between “cool” and “hot”.

Both BMW i8 and Porsche 918 are plug-in hybrid sports cars. However, the Porsche is one of the fastest and most expensive supercars ever produced, and the BMW is more practical and “affordable” (about 1/6 the price tag of 918). The question is, which is cooler?

Web Collector

The Web Collector is a good place to begin because it offers a straight forward first look into the popularity of the subjects under examination.

This is a data set that searches web pages updated between Sep 18 – 25th with the term “bmw i8” or “porsche 918” in them. Official websites were excluded to give a more objective view. Because of the limitation of searches comes with the free Google CES key, the number of websites is limited, but it still gives a fair representation of the two terms’ popularity on internet. It also allows us to see what websites are mentioning these terms and how they are interconnected. As we can see, Porsche 918 has much more mentioning and there are more links between these websites. It’s an indicator that there are more interests for 918, and the websites mentioning it are more important because they are linked to one another.

In later part of this post, we will come back to this tool to take a look at it over a longer time horizon.

Wiki Evolution

After a quick scan over the internet, we can have a deeper drill into each vehicle by looking at how they are represented in Wikipedia using the tool Wiki Evolution. Since Wikipedia requires people voluntarily take out time to update information, it gives good clue as to how much attention a certain subject is getting.

BMW i8 on Wikipedia

As we can see from the above graph, BMW i8 is connected to very positive terms such as electric vehicle, drag coefficient, sports car, government subsidy, carbon dioxide, 0-to-60. This is a desirable outcome for BMW, since the car is associated with these positive and meaningful terms. People’s perception of this car is tilted towards where BMW would want them to.

Porsche 918 on Wikipedia

Nevertheless, Porsche 918 is also linked to similar terms as i8. Moreover, 918 is linked to some of the highly positive terms that the BMW is not able to reach, such as fiber-reinforced plastic, super car, racing car, Nuburgring. Given that 918 is the fastest production car to lap the Nuburgring up to this point, this edge over i8 is not surprising.

Another point worth mentioning is that all of the models ever produced by Porsche are cross-linked on Wikipedia. It is the cloud of terms shown in the lower right part of the graph. It tells us that the fans of Porsche are devoting more time and energy into this brand. This is a sign of enthusiasm.

Twitter Fetch

Twitter Fetch gives us the mentioning of a term in any given point of time, which is useful to track the change of interest over time.

Above is a static view of 3 fetches of each term merged together. There are people mentioning both cars in their tweets. From the graph, it looks like there are similar interests in both cars.

Betweenness Centrality

To take a closer look at these two terms’ presence in twitter world, we can export the betweenness centrality into excel and form a more easily readable chart. From the above chart, it is clear that BMW i8 is much more important in twitter world on Sep 25, 2013. The production model of this car was just launched for a week, and so this result is in line with expectations.

In fact, Porsche maintained their presence on twitter fairly well, given this specific point of time when BMW is extremely hot. This in part is due to the active role various Porsche official accounts were playing.

From the above two charts of the betweenness centrality of official accounts, we can conclude that there are not only more Porsche official accounts active on twitter, but they are also making a much larger impact.

Word Cloud

Through word sentiment, we can see how people talk about these cars.

Again, people talk about these cars in a positive manner. People see BMW i8 as a “brand shaper”, and compares Porsche 918 to other super cars such as Nissan GTR and Lamborghini. The term “i8” did not come up in 918’s word cloud.

It is interesting to see how people categorize these cars. It seems like these cars are categorized according to their performance rather than been plug-ins. Nissan GTR has a similar price point as the BMW i8 and both sit 4 people, but only GTR comes up in 918’s tweets, not i8. The only reason GTR is so tightly linked to 918 is that both of them are among the fastest production cars, same for Lamborghini.

Understanding Cool and Hot across Time Span

Web Collector (one month later)

Web Collector in Mid-October revealed similar patter as the first fetch performed in Mid-September.

There are more websites talking about these cars, but Porsche 918 holds its advantage in both number of websites and connection between websites.

From this Web Fetch Betweenness chart, it is more revealing that web pages mentioning Porsche 918 are the most important ones. The term “Porsche 918” out weights “BMW i8” significantly, and out of the top 8 nodes, 6 of which are related to 918.

Twitter Fetch (one month later)

From a quick glance, it looks like Porsche 918 and BMW i8 maintained their comparative position in twitter from one month ago, when BMW just officially launched the production model i8. However, the betweenness chart tells a different story.

As the effect of the i8 official launch wears off, Porsche 918 over takes the BMW in betweenness.

Hot vs. Cool

The above cross-time comparisons are good examples of the difference between hot and cool. BMW i8 was hot because it got a lot of media coverage with the launch, but Porsche 918 was cool because it has a prolonged impact across different internet platforms.

It echoes with what Professor Gloor says: hot is easy, cool is hard but lasts longer.

  • BMW i8 was newly launched, and so it was hotter on twitter in Mid-September
  • One month later, Porsche 918 takes over the lead on twitter
  • On Wikipedia and other websites where the fetch gives more weight over a longer time span, Porsche 918 has always had an advantage
  • All of the above findings proves that BMW i8 is hot, and Porsche 918 is cool

Monday, October 07, 2013

Coolhunting the German Elections

In preparation for the German parliament (“Bundestag”) elections, I analyzed the social media landscape using Condor and tried to make educated predictions on the election results. In this blog post I will outline the methods and concepts used during Coolhunting.

Political Landscape in Germany

In total, 34 political parties ran for parliament. In my opinion, seven of them had a fair chance to overcome the 5% barrier, which prevents parties with a low number of votes from being represented in the German parliament.


Elections offer a wide range of interesting Coolhunting topics. First and foremost: Prediction. When looking at the political landscape and the different social media policies of each party, it is quite obvious that one cannot predict the detailed outcomes of an election when just looking at social media. To give a viable example, the German Pirates – a rather small party consisting mainly of tech-savvy young (relative to other parties) people accounted for more than 40% of all hash tag mentions during the last election in 2009, whereas larger parties such as SPD only got around 18% (1).


When comparing the hash tag mentions of the different parties, it becomes obvious that predicting the whole election while only using social media data is not feasible. I therefore decided to focus on the proportion of votes between two parties which are comparable regarding social media policy and profiles. For this reason I decided to take a closer look at SPD and CDU/CSU, which have both been continuously in parliament since the establishment of the German Federal Republic in 1945 and have comparable social media policies.


For Coolhunting I used an Amazon EC2 Instance running two Condor 2.6 installations with distinct Twitter accounts: one fetching tweets for “SPD” and one fetching tweets for “CDU”. Fetching tweets started 9/19 around 10pm and ended 9/22 12pm EDT catching around 5000 tweets, equally split between the two parties. The algorithm only fetched tweets in the German language that were geocoded to Germany and had retweets, which kept the amount of collected tweets low, but the quality high. 

The figure above shows the fetched Twitter network and the betweenness centrality for both search terms. When using the betweenness centrality for calculation, the ration of potential votes between SPD : CDU is 34% : 66%.


Wikipedia fetching was done right after the elections were finished and reflects a potential vote ratio for SPD : CDU of 39.24% : 60.76%.


For fetching web content I used Google CSE right after the elections had finished, leading to a prediction a ratio of votes for SPD : CDU of 41.06% : 58.94%.


When comparing the election results to other predictions by German newspapers, a representative poll done by “Zeit” newspaper one week before the election turned out to be most accurate. A simple hashtag count for #SPD and #CDU done by “Bild” newspaper instead failed completely.

The best coolhunting accuracy was achieved using Wikipedia suggesting that the swarm is more representative than the crowd in this case. However Twitter proved to be the best indicator for current mood swings as the system is far more reactive than Wikipedia or a Google search.

Saturday, October 05, 2013

Analyzing Psychological Concepts on Twitter and the Web

by Ernesto Strazza

This post is based on a class project done for the 2013 COINs Seminar

What do the Web and Twitter tell us about mental conditions and problems? How are the basic ideas related, and what is their context?  What are some key institutions and organizations? 

The results shows that "self esteem" and "depression" tend to be away from each other. Causes and effects of related mental conditions seem to be closer to "depression" and isolated from "self esteem." With the addition of the concept "drug", the relation between "self esteem" and "depression" becomes stronger and more connected. 

The second part of the study shows that  "fame" increases the influence of the concepts "drug" and "self esteem."

- Find relations  and interaction between psychological concepts and conditions.
- Discover how  those concepts relate (link) and aggregate on the Web.
- Which associations and institutions articulate the relationships between those conditions.
- How do their significance and context generate successive connections and new relations.
- Find impressions about the same concepts from Twitter users, interpreting their emotional response.

1. Do background research to find basic definitions and associations.
2. Use Condor to identify common causal and effect relationships.
3. Use Condor to fetch Web: concepts, links and aggregations.
4. Use Condor to fetch Twitter concepts and relationships.

Formal word definitions.

C.3 Web Analysis. Static View. comparing: depression, self-esteem. Node size by Betweenness – Centrality.

C.3 Web Analysis, Word Cloud: depression, self-esteem, New words which are related: suicide, mental health.

C.3 Web Analysis. Static View. comparing: depression, self-esteem. Concepts appear through connectors: mental health, suicide, drug abuse.

C.3 Web Analysis. because of occurrence new words added in analysis: depression, self-esteem, suicide, drug use (abuse). Node size by Betweenness – Centrality. 

C.3 Web Analysis. Word Cloud: depression, self-esteem.
New words which are related: drugs, mental health, celebs.

C.3 Web Analysis. Static View. Node size by Betweenness – Centrality: depression, self-esteem, suicide, drugs, mental health.

C.3 Web Analysis. Word Cloud: depression, self-esteem, suicide, drugs, mental. New word to appear: fame

C.3 Web Analysis. Static View depression. Node size by Betweenness – Centrality. Which connectors articulate the concepts self-esteem, fame, suicide, drugs, mental 

C.3 Web Analysis. Betweenness – Centrality zones and proximities.

C.3 Twitter Analysis. Static View. Node size by Betweenness – Centrality.Word analysis response: depression, self-esteem, suicide, drugs, mental health.

C.3 Twitter Analysis. Node size by Betweenness – Centrality. Word Cloud View.
Word analysis response and association  to: depression, self-esteem, drugs, mental health, fame.

C.3 Twitter Analysis. Node size by Betweenness – Centrality. Word Cloud View.
Several words analysis.
Response to: depression, self-esteem, drugs, mental health, fame.
Appellative and emotional associations or relations to key words. 

Thursday, October 03, 2013

Coolhunting Oakley Apple Google - Wearable devices industry

by Diego Mendez

Using the software Condor and its capacity to interpret word sentiment I will explore the context and relationships between the 3 brands Apple, Google and Oakley on the Web. This approach is based on coolhunting the Web through qualitative data analysis.

Brands Selection
What do Apple, Google and Oakley have in common? The relation between the giants Apple and Google is obvious, as they both compete in the technology arena and they are both trendsetters in their particular and common markets.

How does Oakley fit into this profile? Oakley is the top innovator in the eyewear industry (e.g. they invented the curve lens); this culture of innovation has carried them to explore other potential uses of this eyewear. In particular ,Oakley launched a product called AIRWAVE last year. This product includes a particular technology that allows any snowboarder (or skier) to watch and read analytics, track information, text messages, videos etc… through a very small display embedded into the lens.
Oakley's Airwave. This is, basically, Oakley’s moving into the tech industry.

On the other hand, it is publicly known that Google has launched a device called Google Glass.
Google Glass.This is Goggle’s landing into the eyewear industry (or wearable industry).

How about Apple? Well, the market has been wondering for awhile what is their next big thing. A kind of wearable device? Is it a watch (Oakley also owns a line of watches)? Is it glasses?
Oakley, Google, and Apple have in common their aim in marketing high end technological accessories (wearable industry). Let’s see what the SWARM can tell us?

Brand names association
Twitter brings a good scenario for finding words associated with the brands. Tweets are short with a strong content.

Traditionally associated with coolness, extreme sports and innovation. Fetching tweets (sample over 1500 tweets) containing only the brand name, we could not observe a good sense of this sentiment.

Most of the tweets are referring to e-retailers advertising points of sales (these tweets include re-tweets). Perhaps that is the reason why Oakley Canada has a strong presence.
The only reference to a product is the #frogskin (this is an Oakley lifestyle line); for the moment, no sign of Airwave.

Traditionally associated with disruptive technology.

The tweets (sample of 1500) reflect a stronger presence of products or business concepts (iTunes, music, download), mainly from the music area. It is interesting to see words like amazing and kicking in as positive sentiment. The tweets are cleaner, not as many e-retailers looking to generate traffic to points of sales. This might suggest  better digital marketing management at Apple as opposed to Oakley.

Tweets (1500 sample)

I encountered expected words such as "search", "report", "analytics", "page" along with other negative sentiment words (there were very few negative words in Oakley's and Apple's analysis above). Perhaps, the most interesting finding is  that the word "GLASS" is associated with negative sentiment. Further analysis shows an association of the word "GLASS" with others such as "Privacy" or "Spy".

OBSERVATION 1: Neither of the 3 Brands has a strong association with their products Airwave (Oakley), Glass (Google), or iWatch (Apple).

If we combine the 3 data sets and ask Condor to order nodes by betweenness centrality we can identify actors with a mediating role.

  • The large nodes at the the top and bottom end are the queries "Oakley" and "Google"
  • The query "Apple" loses its relevance
  • Nodes in the middle connecting the big 2 queries are 2 identifiable users.  Their follower base is 2,292 and 127 respectively.

OBSERVATION 2: Reaching out to these users represents an opportunity to influence other followers.

I now use a web search for comparing the 3 products (Airwave, Glass, iWatch). The main reason is that websites offer more content, as opposed to tweets which are shorter.

Airwave (Oakley) vs. iWatch (Apple)
The word Airwave by itself is not strongly related with the brand Oakley; however, if we search for Airwave together with iWatch, we observe the brand Oakley. The web is relating the Airwave goggles with the possible iWatch. Below are the strongest sentiments, graph prunning and some examples of content:

Example of Snippet:
  • Dec 13, 2012 ... And in TEST 24 we check out Oakley's Airwave - the goggles that ... latest releases (a new iPhone, and maybe also an iWatch and an iPhablet?)
  • Sep 9, 2013... previously worked for Oakley on that company's Airwave heads-up goggle ... The company has since filed for trademark on the term iWatch

OBSERVATION 3: Oakley Airwave and iWatch might be complementary goods.
OBSERVATION 4: Talented people working for iWatch might have worked for Oakley Airwave.

Airwave (Oakley) vs. Glass (Google)
The sentiment of these 2 words together is positive (in general). Both products are observed as truly disruptive in their markets (words as "technology", "video", "smart", "wearable").

The snippets suggest a hint of competition.

  • Oct 26, 2012 ... Watch Oakley unveils its version of Google's Project Glass videos on CNET TV: The new Airwave goggles gives skiers an eyefull of stats
  • Mar 5, 2013 ... With the recent development of Pebble Watch, Oakley Airwave and Google Glass , we have been ushered into an era of wearable computing.
  • Mar 11, 2013 ... Since Google announced it wearable device project, Google Glass (originally ... Airwave, the digital ski goggles by Oakley which provide jump
  • Sep 4, 2013 ... In all, it's a full-on wearable technology device that's getting a bit into Google Glass territory, though it's important to note that the Airwave's

OBSERVATION 5: A pattern of new competition is being established in the wearable market.

iWatch (Apple) vs. Glass (Google)
Both products have not been launched yet. However, they both have generated big expectations (words as "rumors", "talked"; word-dates as "Sep" and "Aug").

Snippets suggest a competition between the 2 brands in the new industry (wearable devices)
  • Aug 12, 2013 ... Wearable tech is all the rage these days, with much of the talk about these devices centring on their use as entertainment and lifestyle aids.
  • May 30, 2013 ... Wearable computers are the next big thing. Reports say Apple is coming out with a watch. Analyst Gene Munster guesses it will cost $300.
  • Aug 28, 2013 ... The age of wearable computing is underway, and one by one, tech titans are fielding their futuristic devices to consumers in the hopes of

OBSERVATION 6: The Swarm considers these 2 brands to be in competition in the wearable market.
OBSERVATION 7: The Swarm, perhaps, does not see Oakley as a direct rival for the 2 big names Apple and Google.

This first analysis opened up some possible hypotheses for further development
  • Can Oakley compete in this wearable-technology market? Or due to its smaller size, does it need to partner up?
  •  Will other big brands penetrate this wearable technology market? (there have been rumors about Dell)
  •  Might Apple or Google identify the big influence-users from Airwave on-line and lure them to their own new devices?

Friday, July 12, 2013

Is Zimmerman guilty?

.... that's what the Twittersphere seems to think. The process in Florida, whether George Zimmerman killed teenager Trayvon Martin in cold blood is now resting with the jury. I quickly checked the Twittersphere through Condor Coolhunting, searching the most recent 2000 retweets for both "zimmerman guilty" and "zimmerman innocent". Drawing the retweet network leads to the following picture:

The size of each red circle denotes the number of followers of each twitterer, the connecting line between two circles A and B means B retweeting A. Looking at the betweenness of the circles tells that the guilty tweets carry a higher weight (0.74) compared to the innocent tweets (0.69), which means that the "guilty" tweets are retweeted more, and by more influential twitterers. The question now is: do the 6 jury members come to the same conclusion as the tweeting crowd?

Thursday, March 28, 2013

Getting a Grip on Data in Motion to Understand How We Collaborate

Recently I had a great discussion with my friends from Cisco about how to use dynamic social network analysis - initially of e-mail archives, but now extended to all types of communication archives - to improve collaboration in organizations. In particular, we use the virtual mirroring process, which I described in previous blog posts.

Thursday, February 14, 2013

Analyzing the Communication Network of the 2012 COINs Seminar

In this seminar about 50 students from the five universities MIT/SCAD/Aalto University/ University of Cologne/ University of Bamberg worked together for five months in multinational virtual project teams as COINs (Collaborative Innovation Networks). They formed ten teams ranging in size from three to six students from at least two locations, working on a project related to social media and social network analysis (2012 course Web site). Students were asked to cc all their project-related e-mail communication to a dummy mailbox. This allowed us to construct a virtual mirror of ongoing communication within and between teams. At the end of the course, each team presented their results to their classmates in a virtual meeting. Each of the ten presentations was ranked by the students in the three categories “presentation quality, content quality, and creativity”. Comparing the virtual mirror of communication with the peer and instructor ratings permitted us to identify the communication patterns leading to the most highly-ranked work output.

Qualitative Analysis
Looking at the group network below, the different teams can be clearly recognized. The communication of each team is shown in a different color, usually team members are clustered together as a COIN, with external collaborators and other students being in more peripheral positions.

Analyzing the contribution index between team members shows that members of the same team tend to show similar behavior regarding the ratio of e-mails sent to e-mails received. Clusters of dots of the same color are members of the same group, overall we find that higher-performing teams tend to communicate more actively, with more similar send/receive ratios.
The temporal social surface indicates creativity, as there is a relatively large group of high-betweenness class members which is constantly changing over time, in earlier work we found this to be a reliable predictor of creativity.

The 6 snapshots of the 10 teams’ communication networks over the 5 months show Tuckman’s four phases in the life of a team: forming, storming, norming, and performing. We see how the main instructor in the uppermost picture at right is most centrally, but how then teams start connecting in the middle row, and how they then huddle together team-by-team to focus on their work in the bottom-most pictures.

The group betweenness centrality curve as well as the absolute number of messages sent and received shown below illustrate the higher traffic in the forming, storming, and norming phase, followed by the lower traffic in the second performing phase.

The sentiment curve illustrates the same phenomenon, with higher emotionality (defined as the sum of positivity and negativity) in the forming and storming phase in the first half of the course. X-axis is always days in these pictures.

Quantitative analysis
In the second half of this analysis we investigate what communication patterns will be indicative of high-quality work. The first pattern is “oscillations in betweenness centrality curves”.

The above picture illustrates the team ranked most creative by the instructors (blue shaded lines, each line titled “Series X” is one actor's betweenness over 115 days), and the team ranked least creative (red shaded lines). As can be easily seen, the centrality of most actors in the low-ranked team hovers around the zero-lines: these actors will be peripheral in the social e-mail network shown in the first picture of this document.

As the correlations below illustrate, the instructor rating of creativity (each of the instructors at each of the five participating locations ranked the 10 presentations) correlates highly (0.83**) with oscillation in betweenness centrality. As the picture above shows, the team rated the most creative had 80 oscillations, i.e. handovers in leadership, compared to the lowest rated team with less than 40 oscillations.

A similar correlation was identified for speed of response. The faster a team’s members communicated with the lead instructor, the higher the team’s work output was rated by the other students.

The correlation between the peer rating of a team’s content with its communication balance with the main instructor (Peter) is -0.719**, i.e. the higher the communication balance, the more communication with Peter, the better is the team’s content.

The more emotional the language of a team in the e-mails they exchange, the higher is the content rated by its peers.

The positivity in the e-mails sent to the main instructor is also highly predictive of high ratings in all criteria that have been rated (Presentation, Content, Creativity).

The conclusions for high-functioning teams are therefore:
1. Pass the baton frequently: the more leadership rotates among team members, the more creative the output will be
2. Communicate rapidly with the instructor, and among team members
3. Use emotional language: praise when praise is due, but also say when something is not ok.