| The rapid growth in volume of web texts from major social network sites like Facebook and Twitter drives us to analyze and mine the data through computational techniques. In my thesis, I mainly study social media data analysis for marketing intelligence. It has a variety of important applications, including: sentiment identification for social media texts; recommending online products based on customer reviews, which could helps users make informed purchase decisions and manufacturers adopt intelligent marketing strategies; ranking celebrities in terms of social reputation; building predictive models for audience-targeted online social advertisement based on user historical behaviors. My thesis work focused specifically on 1) social media sentiment identification, 2) using sentiment information to rank social entities1 by applying graph theories and probabilistic models, and 3) using distributed techniques and optimization methods to find targeted audiences for better improving social online advertisement.;However, due to the specialty of social media data, old learning-based sentiment classification algorithms are not applicable without modifications. In addition, making recommendations only based on texts while ignoring social network information leads to an unfair system. Therefore, my thesis shows how to incorporate syntactic, semantic, and context information from texts and network information from historical social activities to improve social sentiment identification. Based on these improved sentiment identification, I also study how to use graphical models and probability theories to rank social entities while reducing biases given large data sets. In addition, I employ optimization techniques and distributed methods to build a predictive user model for identifying targeted users for online advertisement. To summarize, my existing research provides a comprehensive set of techniques that span multiple views of data, from considering data as static texts, to importing derived network information behind the data. It also provides an automated computational framework of ranking social entities under big data.;Since social network often involve technological, social and/or economic interactions among nodes, it is necessary to use theories and models from sociology and psychology to study them. In my thesis, I also present some future work which includes understanding online communities through evidence-based social design by mining the social data. I will also study the dynamics of the network and understand the relationship among network structures, node properties, and information spreading by using some cascading models. Since the data becomes big and unstructured, finding scalable machine learning model and optimization methods also becomes my future research interests.;1A social entity is an object in the social network that allows other users to leave comments/reviews on its page. Examples are companies, organizations, individuals, or consumer products.}. |
