Social and Organizational Search

Peter Dodds, Collective Dynamics Group, Columbia University

Download Presentation

Tonight we will talk about Social Search. The main question with Social Search is, "Can people pass messages between distant individuals using their existing social connections?" Yes, apparently because of the "Small World" phenomenon or "Six Degrees of Seperation".

Click on image for enlargement

Milgram's original experiment on this topic was in the 1960's. In his search, he began with a target person in Boston who was a stock broker. There were 296 senders from Boston to Omaha. In the end, 20% of the senders reached their target. The results were an average chain of senders that was 6.5.

If we look at this problem closer, we can see two significant characteristics of the small world network: short paths and people who are good at finding the paths. Connected random networks have short average path lengths. They scale with the size of the system. Social networks are not random. We need "clustering". In a non-random network there are long distances (an average of ten steps) between point "a" and "b". But if we began to add some randomness and regularity to the networks we can narrow the number of steps in path down to 3.

But are the short cuts findable and accessible? No. Nodes cannot find each other quickly with any local search method. How can we navigate? Jon Kleinberg wanted to look more into navigating a small world network. The small world network allowed things to vary like local search algorithms and network structures. He started to find short cuts by looking at networks and adding local links. He found that the 'greedy' algorithms work best. The best network you can have is one where you have long range links that are connecting at different lengths.

Another type of network is to have hubs that can also search. But hubs in social networks are limited. Some hubs know a tremendous amount of people for these searches to work at this level. The hubs begin to create distributed networks. Hubs are a contentious issue. They are not absolutely vital when links are easy to make.

If we do not have hubs or a latice how do you search efficiently? Which friend is closer to the target? One solution is to incorporate "identity" into the experiment. Identify is formed by attributes such as geographic location, type of employment, religion or recreation. Groups are formed by people with at least one attribute in common.

Six propositions about social networks

Proposition 1: Individuals have identities and belong to various groups that reflect these identities.

Proposition 2: Individuals break down the world into a heirarchy of categories.

Proposition 3: Individuals are more likely to know each other the closer they are within a hierarchy.

Proposition 4: Each part of your identity corresponds to the identity hierarchy.

Proposition 5: "Social distance" is the minimum distance between two nodes in all hierarchies.

Proposition 6: Individuals know the identity of themselves, their friends and the target. Individuals can estimate the social distance between friends and the target.

Conclusion

In conclusion, bare networks are not enough. The paths are findable if nodes understand how the network is formed. The importance of identity is very strong.

Applications

With this knowledge you can improve social network methods, construct peer-to-peer networks and create searchable information databases.

Recent Experiment

We recently ran an experient concurrently that involved 60,000 participants in 166 countries. There were 18 targets in 13 countries including a professor at an Ivy League University, an archival inspector in Estonia, a technology consultant in India, a policeman in Australia, and a veteran in Norwegian army.

At each stage, we had 37% participation with a probability of a chain of length of 10 getting through to the target. 384 chains made it through which gave us a result of 1.6% of success.

In the experiment, we found out that motivation, incentive, and perception matter. If the target seems reachable, the participant is more likely to participate. For our experiment, we had a median of five to seven. The complete chains were four.You can compare our results of 5-7 chains versus Milgram's results of 7-9.

It is a small world if you think about it. The successful chains disproportionately used weak ties, professional relationships, and target's work. The chains disproportionately avoided hubs, friends, and target's locations. In Milgram's work, he saw that there were key funnels of people who fed information to the target.

Current Experiments

We have two more experiments that we are operating right now. The first one is Small World Experiment II where you can nominate your own target. The second is an expert search experiment called the People Finder Project.