Michael  Twardos
Projects in Data Science and Behavioral Analytics



Characterization of Online Influence Events: Using data across the most popular social media platforms from Klout's data warehouse, influence events such as likes or comments are characterized across the Klout user base. An assessment of the correlation between different social media events is proposed to implement more meaningful and less redundant measures of influence.



Recommendation Optimization These papers descibe algorithmic strategies to handle impression allocation for a content platform to optimize a given metric. The algorithms use transition proclivity for a given behavior and account for uncertainty in measurements used in optimization to determine level of risk / return.



Population Modelling based on Growth and Retention: This prediction model defines the retention and virality as specific mathematical probabilities and uses infinite series to compute long term population trends. Incorporating specific examples of retention and virality curves, I compare to the actual population changes in a selection of digital communities.



Behavioral Regularities in Financial Decisions: The likelihood of offers made to buy and sell a given option is characterized as a function of existing limit order book structure.



Popularity Dynamics in Social Media: Using data from the "most popular" items in Sporepedia, the online data source for user generated creations in the video game Spore, the transitions and overall probabilities of popularities is characterized. Overall, this study suggest ways to generate popularity algorithms to better represent recency or historical thoroughness. Presented at the ICWSM 2010.




Other Work



Constructive Solid Geometry: I built a basic mathematical foundation to digitally add and subtract (boolean operations) two arbitrary three dimensional objects. Items are triangulated and identified by a normal vector to the surface. The software identifies what triangles intersect and what new pieces are inside or outside the resulting structure.



Ocean Wave Physics This college level physics textbook provides a basic foundation to understanding and studying dynamics associated with breaking waves in the ocean. The text starts off with simple mathematical models such as the harmonic oscillator and wave equation. The reader is then introduced to more advanced mathematical topics including Navier Stokes and nonlinear equations.



Turbulent Mixing: In this research performed at Los Alamos National Lab, we used several metrics to parametrize mixing in the transition from periodic flow to turbulence. As the Reynolds number of the system increases, there is a monotonic increase in the rate of mixing and now obvious discontinuity at the onset of turbulence.



Fluctuation Dissipation Theorem in Non Equilibium Systems: For my PhD thesis we studied the dynamics of macroscopic particle systems (non-thermal) to draw parallels in statistical methods used in thermodynamics. For systems near the jamming transition, a fluctuation dissipation relationship exists across all observable timescales and a temperature can be defined.



Elasticity in Biological Membranes: Langmiur monolayers consist of amphiphilic fatty molecules self organize on the surface of water. As the available surface area is decreased these systems transition between different packing fractions and phases. Beyond the solid phase, we study the mechanical properties as these systems buckle into the 3rd dimension.



The Spore API: This restful API serves the gaming community detailed information about the usage and construction of UGC used in the game.