Center for Complex Networks and Systems Research

As adaptive peer network systems becoming an increasingly important development in Web search technology, in this research, an alternative model for peer based Web search is introduced to address the scale problem of centralized search engines. Queries are first matched against the local engine, and then routed to neighbor peers to obtain more results. Initially the network has a random topology (like Gnutella) and queries are routed randomly as in the flood model. However, the protocol includes a learning algorithm by which each peer uses the results of its interactions with its neighbors to refine a model of the other peers. This model is used to dynamically route queries according to the predicted match with other peers’ knowledge. The network topology is thus modified on the fly based on learned contexts and current information needs.

This week at NaN I’ll be running through a very preliminary version of my thesis talk, “Structural Mining of Large-Scale Behavioral Data from the Internet,” which is all about the things you can discover using network flow data and Web clicks. The big things that I’ll be looking to get as feedback have to do with organization and pruning — I’m not terribly confident of the order in which I present the material, and I have a LOT more things to say than time to say it in, so I can use some suggestions on what to keep and what to just point to the actual document for.

(And, yes, there will be cookies.)

Online documents provide a rich information resource for aiding the generation of concept-map-based knowledge models, but analyzing resources to select concepts and links is a time consuming task.  This work focuses on harnessing the information in unstructured text documents using text mining algorithms to generate preliminary concept maps automatically.  These maps can be used to assist human users on question answering tasks or automatic document classification.

I will talk about some work related to to the problem of predicting the popularity of online content, and some initial results from my experiments in this area. More in detail, I’ll overview work by Leskovec et al and Huberman et al on modeling and predicting growth, then outline the results of two initial experiments.

“The problem with Wikipedia is that it only works in practice. In theory, it can never work.”  — Zeroeth Law of Wikipedia

One of the most important social and intellectual phenomena of the 21st century, the collaboratively-edited online encyclopedia Wikipedia is vexing in its ability to produce informative articles on a multitude of subjects.  Leveraging graph theoretic techniques to measure the degree to which latent connections between articles are present in the Wikipedia corpus we demonstrate that the collaborative editing process produces, over time, an increasingly logically-connected information artifact. Moreover, using the public-domain 1911 Encyclopedia Britannica as a benchmark corpus for the single-author-article paradigm, we demonstrate that Wikipedia contains a growing core of mature articles which exhibit a degree of logical connectedness significantly surpassing that found in the Encyclopedia Britannica. Taken in conjunction with an understanding of Wikipedia’s accuracy and topical coverage, this conclusion paints a rich portrait of the strengths and weaknesses of both collaboratively- and single-author-edited encyclopedias.

I’m presenting my pre-defense content of my research.  There will also be cakes!

Dear NaNers,

I will be using Tuesday (3/24) as a practice talk for AIRWEB 2009. Hope everyone had a nice spring break!

Title: Social Spam Detection

Abstract:
The popularity of social bookmarking sites has made them prime targets for spammers. Many of these systems require an administrator’s time and energy to manually filter or remove spam. Here we discuss the motivations ofsocial spam, and present a study of automatic detection of spammers in a social tagging system. We identify and analyze six distinct features that address various properties of social spam, finding that each of these features provides for a helpful signal to discriminate spammers from legitimate users. These features are then used in various machine learning
algorithms for classification, achieving over 98% accuracy in detecting social spammers with 2% false positives. These promising results provide a new baseline for future efforts on social spam. We make our dataset publicly
available to the research community.

Regards,
Ben

Ciro Cattuto and I co-chair the social linking track at HT08, the ACM Conference on Hypertext and Hypermedia, which will take place in Pittsburgh in June. Our track received many quality submissions and the technical program is shaping up to be very interesting. Bernardo Huberman will also surely deliver an exciting keynote. Registration is now open!

P.S. Congratulations to Ben, Heather, Justin, and Mike for getting their papers accepted!

1. Web click traffic network (by Fil) at Complex Networks: Dynamics and Topology Interplay
2. GiveALink (by Ben) at Enhancing Social Interaction: Recommendation Systems, Reputation, P2P, Trust and Social Networks
3. 6S (by Fil) at Enhancing Social Interaction: Recommendation Systems, Reputation, P2P, Trust and Social Networks
4. •Social Search (by Fil) at Social Websites – Complex Dynamics and Structure