Behavioral Analytics & Visualization
Behavioral analytics is the study of human-generated Big Data such as those generated by mobile devices, banking transactions, and cell tower / wi-fi connectivity, where the goal is to analyze and identify behavioral patterns of individuals in space and time. Such patterns may then be used at the micro level to model and explain human behavior, or at the macro level to understand how masses of people such as inhabitants or visitors of a city or country move around and consume economic resources. The quantitative analysis of the data is usually accompanied with visualization tools and visual analysis, which make it much easier to identify patterns.
The BAV Lab, co-founded by FENS, SOM and MIT Media Lab’s Human Dynamics Group, is where we analyze data donated to us by major banks and insurance companies of Turkey, as well as data from public sources. We conduct research on such topics as customer shopping behavior and churn analysis, financial risk analysis, insurance fraud detection, data physicalization and collaborative decision making using tangible table-top displays. We rely on many quantitative techniques such as machine learning and data-mining for predictive analytics, optimization modeling, spatio-temporal modeling, and use various analytics software such as R, Weka, SAS, Matlab and ArcGIS. Feel free to visit our website for further information
Big Data Visualization
In recent years visualization of large temporal social network datasets (email, Facebook etc.) has become one of the hot topics in information visualization domain. Existing techniques rely on relational information to compute layouts as node-link diagrams and omit network metrics. Therefore, analysts employ software suits to understand the structure and the communication channels between actors of evolving social network datasets. We can summarize the requirements for visualizing temporal social network datasets as follows: The underlying method must use an appropriate visual language to combine information space and the visualization space. The technique must reflect the evolution of the network through minimizing the topological differences between adjacent images. Finally the method must highlight the important aspects of the data with interactive, topology preserving clutter reduction techniques.
Privacy Preserving Data Management
Knowledge Discovery and Data Mining has a lot of applications in personal data analysis which may contain sensitive information that is protected with law and regulations. Protecting sensitive data while performing data analysis or data publishing is crucial especially in the context of Big Data. My research includes data anonymization and privacy preserving data mining. Currently I am interested in privacy preserving data management for complex data.
Social Media Analytics/Opinion Mining
With the widespread use of social media, a lot of textual data is being generated which reflects the opinion of people towards various issues. Research in Social Media Analytics aims to leverage the large amount of information in social media (Twitter, Facebook, …) for understanding how people view a certain product (e.g. a movie or new car model) or a brand; or what is the public opinion with respect to political or social events. Opinion mining deals with the extraction of opinion of people from large scale textual data sources in general. In this context, we have established a research group (sentilab.sabanciuniv.edu) and a startup company (http://www.somatech.com.tr/). We are working on general tools and techniques for opinion mining with a special focus on Turkish. The methods combine Machine Learning, Deep Learning, Data Mining, along with Natural Language Processing (for both English and Turkish). Our start-up company SomaTech is commercializing the technologies developed in this area.