As any parent or educator will likely tell you, the behavior of teens closely linked in a social network will often display many similarities: teens who smoke or drink, for instance, are often friends with other teens who smoke or drink. By establishing and tracking the spread of these behaviors scientifically, we can gain a greater understanding of the mechanisms at work and perhaps harness them to help spread healthy behaviors instead of unhealthy ones.

When we look at two teens who share a common behavior pattern (healthy or unhealthy), we must ask ourselves: Did they become friends because of their similar behavior (selection), did their behavior become similar as a result of being friends (influence), or was there some third factor at work which influenced them both separately (confounding factors)? One simple way to attempt to answer this question is through a longitudinal study, where data is collected for the same group of subjects at multiple times. By looking at the co-evolution of the social network and the behavior network, we can parse out the role that each of these factors plays. Here, I wanted to briefly discuss a few studies which have employed social network analysis and longitudinal data collection to gain a better understanding of how unhealthy behaviors can spread amongst teens.

The first is a study by Ennett and Bauman (PDF) which examines smoking as a function of position in the social network.(1) Specifically, they name three classes of social network patterns: cliques (a small group of at least three adolescents whose primary friendships are with each other), liaisons (adolescents who maintain multiple friendships without being in a particular friendship clique), and isolates (adolescents who have relatively few friendships with others.) Past research showing that cliques tend to share smoking behaviors leads many people to the assumption that smoking is a primarily peer group phenomenon.  However, after looking at data from 1,092 students collected across 5 schools over 1 year (from the start of 9th grade to the start of 10th grade), the authors found that smoking was far more common among isolates (17-40% across schools) than among clique members (4-16%).  Additionally, within the 87 cliques identified, they found that smokers tended to associate in the same cliques, with the majority of cliques composed entirely or almost entirely of non-smokers. In looking at the roles played by influence and selection, they indicate that both processes contributed equally to similarity in smoking behaviors for clique members, though they do not discuss how they performed their analysis.

The next paper described a 2009 study by Mercken, et al. (PDF) which utilizes a “stochastic actor-based model” to help in separating the roles of influence and selection “by simultaneously representing changes in friendship network structure and changes in smoking behavior among adolescents.” (2) In this study, they interviewed 1326 subjects from 11 Finnish schools 4 times over the course of 30 months (starting at the beginning of 7th grade). Each time, they asked about their friendship ties, their smoking behavior, that of their families, and their alcohol consumption. They found that adolescents who smoked more had a tendency to choose friends who likewise scored high on smoking behavior. Adolescents who smoke less than one cigarette per week were most likely to make friends with classmates who don’t smoke at all, while the most attractive potential friends for those who smoke one or more cigarettes per week were those who smoked at the highest rate. The authors did not report findings regarding the data collected on alcohol consumption, which hints at the fact that the patterns of spread may be different for different behaviors.

A natural question at this point is: are all friends created equal? Most of us growing up had a “best” friend in addition to our peer group. A 1997 study by Urberg, et al. (PDF) attempts to parse out the influence of close friends vs. that of one’s peer group as it pertains to both cigarette smoking and alcohol use.(3) In this study, they collected data from 1,028 Mid-western school-children in the 6th, 8th, and 10th grades; data was collected in two waves, once in the Fall and once in the Spring, and included assessments of friendship ties as well as cigarette and alcohol use. Interestingly, they found that it was smoking behavior of the peer group and not the close friend which predicted a transition into cigarette use, while it was the drinking behavior of the close friend and not the peer group which predicted a transition into alcohol use. They also found that those who have tried cigarettes or alcohol are more likely to know current users than those who have not (echoing the Mercken findings above).

Finally, because I couldn’t close a post on social networks and health without a Christakis/Fowler study, I wanted to mention a study from March of this year (2010) from Mednick, Christakis, and Fowler (PDF), which examined the interaction of two separate behaviors–low sleep and drug use–within a social network. (4) Looking at a sample of 8,349 adolescents from the ADD Health Data Corpus, they presented a number of interesting findings. First, they found that an individual’s behavior is correlated with the behavior of others in their network up to 4 degrees away. In the case of sleep, an individual was 29% more likely to sleep 7 hours or less if they had a friend who sleeps less than 7 hours; a friend of a friend correlated with a 17% increase, all the way down to a 5% increase for the friend of a friend of a friend of a friend. In the case of marijuana use, a direct connection to a user resulted in a 190% increase in the likelihood of use, while a 4th-degree connection still correlated with a 11% increase in use. Another interesting finding was that individuals central in the network were more likely to sleep less, with a two standard-deviation increase in centrality increasing the probability of sleeping 7 hours or less by 13% (controlling for other factors). Finally, they report on the interrelation between these behaviors, claiming that having a friend who slept 7 hours or less actually correlated with a 19% increase in smoking marijuana.

These 4 papers served as a useful introduction to both the methods of social network analysis and some of the interesting findings as they pertain to health behaviors and teens (across a number of behaviors – sleep, smoking, drinking, drugs). It is important for educators and health professionals to have an understanding of the social mechanisms as these will likely be a critical factor in preventing unhealthy behaviors from spreading amongst teens and persisting in their lives. Perhaps these same mechanisms can be used to spread positive behaviors such as exercise and civic-mindedness. In addition, it will be interesting to see how methods like these can be applied on a larger scale to Twitter or Facebook-sized social network corpora to track the spread of behaviors, ideas, diseases, and more across entire states or countries.

References:

1. Ennett, S.T. and Bauman, K.E. (2000). Adolescent social networks: Friendship cliques, social isolates, and drug use risk. In Hansen, W.B., et al. (eds) Improving prevention effectiveness. Tanglewood Research, Inc. Greensboro, NC.
2. Mercken, L., et al. (2009). Dynamics of adolescent friendship networks and smoking behavior. Social Networks.
3. Urberg, K.A., Değirmencioğlu, S.M., and Pilgrim, C. (1997) Close Friend and Group Influence on Adolescent Cigarette Smoking and Alcohol Use. Developmental Psychology, vol 33(5), pp. 834-844.
4. Mednick, S.C., Christakis, N.A., Fowler J.H. (2010) The Spread of Sleep Loss Influences Drug Use in Adolescent Social Networks. PLoS ONE vol 5(3): e9775.

Also, I want to thank Sarita Yardi and Vladimir Barash for directing me towards some of these papers.

The paper, by Aardvark’s Damon Horowitz and Stanford’s Sep Kamvar, focuses mostly on the architecture of the Aardvark system, from the external representations with which users interact to the internal ranking algorithms on which the system runs. Below, I present a short summary of what they report, focusing on the elements I found most interesting:

The Basic Model: Aardvark’s scoring function is similar to PageRank in that both utilize two primary, but somewhat independently considered components: relevance and quality.

• Relevance in the Aardvark model pertains to the probability that a particular user i can answer the given question q based on the identified topics contained in t.
• Quality in the Aardvark model pertains to the overall probability that a user i can return a satisfactory answer to another user j, regardless of the question.

Indexing Topics: Aardvark computes the relevance score by calculating a distribution of knowledge over topics known by the user using the following sources (keyword-y sounding italicized terms are for convenience only and are not used in the paper):

• Explicit Prompting at sign-up for three “starter” topics about which the user has expertise.
• Social Prompting of a user’s friends to provide topics about which they trust the user’s opinion.
• Structured Parsing of the online profile pages connected to Aardvark by the user (e.g. “Interests” on a Facebook profile).
• Unstructured Parsing of the users’ online homepage, blog, or status updates using a linear SVM to extract overall subject area and a named entity extractor to extract more specific topics.

Indexing Connections: Aardvark computes the quality score by building a set of weighted connections between users using characteristics ranging from social proximity to similarities in demographics or behavior, such as:

• Social Connections either in the form of explicitly defined “friend” connections or implicit “network” connections, such as both being part of the Stanford network.
• Demographic Similarity, which likely includes age, gender, and location based on profile information collected by Aardvark.
• Profile Similarity, which seems to include similar movies and other items which might be listed on other profiles, such as Facebook.
• Vocabulary Match, which they explain with the example of “IM Shortcuts” (i.e. I assume this means it is based on the language you use to interact with Aardvark, but I am unsure.)
• Chattiness and Verbosity Match, which relate to frequency and length of messages used when interacting with Aardvark.
• Politeness Match, which basically seems to mean whether or not say “Thanks!” or not.
• Speed Match, which is a measure of responsiveness to other users.

Analyzing Questions: While all of the other components are pre-computed, this part is computed at question time (obviously). The utilize a number of classifiers to classify the question and then a set of mappers to map the question to a set of topics, noting that “the role of the Question Analyzer…is simply to learn enough about the qeustion that it may be sent to appropriately interested and knowledgeable human answerers”. Here are the classifiers they list (with the names used in the paper):

• NonQuestionClassifier: Determines if input is a valid question.
• InappropriateQuestionClassifier: Determines if input is obscene, spam, or otherwise unsuitable for asking.
• TrivialQuestionClassifier: Determines if input is a simple factual question (examples given: “What time is it now?”, “What is the weather?”). If so, the user gets an automatically generated answer via traditional web search.
• LocationSensitiveClassifier: Determines if the question contains location information; if it does, it passes that information along to the Routing Engine

• KeywordMatchTopicMapper: Checks for string matches against user profile topics (the mapper attempts to classify meaningful vs. spurious matches).
• TaxonomyTopicMapper: Classifies question text using an SVM trained on an “annotated corpus of several million questions” (where did they find that?)
• SalientTermTopicMapper: Extracts salient phrases using a noun-phrase chunker and tf-idf and finds “semantically similar user topics”.
• UserTagTopicMapper:Utilizes tags explicitly provided by the asker or other answerers and maps them to user topics.

This description of the routing algorithm comprises the main function of the paper. After some more description of how users interact with the system, the authors provide some interesting data collected over the past several months of use (from the beta launch in March 2009 until October 2009).  Here’s a quick run-down of the more interesting facts that they presented:

• Strong User Growth: As of October 2009, they reported 90,361 user accounts, and users appear to be remaining active (in the study period, over 1/2 the users actively generated content and over 2/3 of the users passively participated).

User Growth on Aardvark (graph taken from the paper).

• Higher Query Contextualization: Aardvark queries average 18.6 words in length while the average query length reported for web search is between 2.2 and 2.9 words (citing previous comparison and characterization studies).  They further state that “98.1% of questions are unique”, though I am unsure as to how exact they are being about matching (I am sure the question “What’s a great restaurant in SF” has been asked 1000 times in different forms). In addition, they report from manual scoring of 1000 randomly selected questions that 64.7% of questions asked have a subjective element, with advice about travel, restaurants, and products being specifically popular.
• Fast, High-Quality Answers: They report that 87.7% of questions get answers and 57.2% received an answer within 10 minutes. They report that 70.4% of answers receiving feedback are rated as “good” and only 15.5% are rated as “bad”. Interestingly, they observe a notable difference in feedback on answers from users within the asker’s social network (76% rated as food) and outside the asker’s network (68% rated as good).

Questions on Aardvark (chart taken from the paper).

Overall, I really enjoyed reading this paper. After using Aardvark for over a year now, it was really interesting to get to peer inside and see how the system works, and a lot of great details were provided about the ranking engine.

One place where I feel that the authors missed the mark was in the cursory side-by-side evaluation which pitted Aardvark against Google for a set of 200 questions randomly selected from the Aardvark system. They report that 71.5% of the questions studied were answered successfully on Aardvark, while 70.5% of the questions were answered successfully on Google. This comparison seems mostly useless as the questions, having been pulled from the Aardvark system in the first place, are ones that were specifically chosen because they are better adapted to what is being called ‘social search’. This comparison left me desirous of more investigation into two main questions.

“What makes a search engine ‘social’ in the first place?”

The distinction between social and non-social is extremely murky, something Brynn and I discovered when working on our Social Search paper. It has been argued before (one small example here) that Google’s PageRank algorithm is inherently social, as it aggregates information provided by people (links to one another) to rank results. However, it is clear that something seems categorically different between Google and what people perceive to be ‘social search’. When it comes down to it, even though everyone is excited about Google’s forays into “Social Search”, there’s nothing all that fundamentally different about Google indexing your blog and your tweets than any other documents extant on the web.

To me, it seems that the key difference is really the change in the direction of interaction. While Google takes a query (question) and compares it against traces of discussion about that question from the past (web documents), systems perceived as ‘social’ take a question and attempt to generate new answers in the future. This change in direction is what allows for the higher context that makes ‘social’ search answers so much more rich (at least for some questions.)  Perhaps we need a different word to define this phenomenon – ‘real-time search’ seems to get at it more, but has its own problems.  Perhaps something like ‘generative search’? I really don’t know.

“Why do we need a social search engine at all?”

This one seems like the best fodder for a follow-up study by Aardvark. While they do provide a rough breakdown of the types of questions asked on Aardvark (see pie chart above), I think that a comparison might have been much more interesting if they had looked at a variety of classes of user needs and had compared the relative efficacy of searching on Aardvark and a traditional search engine such as Google. It is clear that ‘social’ will work much better for some needs and much worse for others, but up to this point, people who talk about social search always seem to use the same types of examples (travel, restaurants, and products, for instance). It would be great to get a clear idea over a wide range of needs and use cases where systems such as Aardvark can provide benefits over existing tools.

Anyways, for those of you interested in ‘social search’ and search systems, I encourage you to read this paper and tell me your thoughts!

Today we’re introducing Raindrop, an exploration in messaging innovation being led by the team responsible for Thunderbird, to explore new ways to use Open Web technologies to create useful, compelling messaging experiences.

We hope to lead and spur the development of extensible applications that help users easily and enjoyably manage their conversations, notifications, and messages across a variety of online services. A central principle behind Raindrop is that messaging should be personal — we want Raindrop to be people-centric both in how we process messages, and in how we can help give people control over their personal data and experiences.

When a friend’s link from YouTube or flickr arrives, your messaging client should be able to show the video or photos near or as part of the message, rather than rudely kicking you over to a separate browser tab. Notifications from computers and mailing lists should be organized for you, not clutter your Inbox or require tedious manual filter setup. It should be easy to smoothly integrate new web services into your conversation viewer entirely using open web technologies.

The post doesn’t remains a little too vague to offer a specific vision of what they are talking about.  Essentially, it sounds like Raindrop will be some sort of aggregator for conversation on the web, delivering messages to you in an email-like format.  The “fundamental ideas” video shines the light a little bit more on the idea of intelligently culling “personal” messages (as opposed to bulk) from your various streams. (P.S. The video didn’t play correctly for me, but you can watch it in large-format at Vimeo here).

The "Second Iteration" of the Raindrop Interface

As I have obviously not yet had the opportunity to try out Raindrop, I can’t really give any sort of review of the service.  However, I think that the design principles here are interesting; with the increasing number of conversation platforms appearing on the web, the need for intelligent aggregation is growing quickly.  Even Friendfeed, the leading social web aggregator, felt unmanageable to me at times, and they weren’t even trying to deal with email!

I also enjoyed the use of “Raindrop” as a name, as it conjured up (for me) a very specific comparison with another possibly water-themed product.  While Google Wave’s approach to aggregating information is to literally inundate you with it and force you to use the search function to paddle your way out, Raindrop (in theory, at least) seems to focus on keeping messages separate, allowing you to catch a few in your hand when you need them.