Search and analysis of user keywords in online social networks

Online social networks provide a great platform for people to come in contact, form relationships and socialize over the internet. Major online social networking platforms like Facebook, Twitter, LinkedIn have grown rapidly in the last few years. In addition to listing friends, users share a variety of information on the social networking platforms that can be broadly categorized as personal information and non-personal information. The personal information shared by users is highlighted in their user profile entries: name, contact information, geographic location, interests, etc. The non-personal information is through the sharing of traditional Web data such as internet URLs. Taken together, this provides a sizable dataset of considerable heterogeneity with user-to-user link structure and user-to-content link structure. Such a dataset provides us with an unique opportunity to analyze the relationship between the two set of link structures and build software systems that leverage this dataset. In this thesis, first, we analyze the impact of user-to-content link structure on the growth of the social network itself. We examine the effect of user similarity on friendship formation through content correlation. We develop a model to relate keywords based on their semantic relationship using natural language processing techniques. Second, we put forward a novel algorithm for decentralized search between users of a social network in a scalable and efficient manner. The algorithm is based on a linear combination of the topological distance and trust between user pairs in a social network. Finally, we present InfoSearch: a social search engine. The goal of InfoSearch is to exploit the heterogeneous dataset to improve the relevance of information search results through a social context. To do so, we employ six different ranking factors to define relevancy of result sets and quantify the impact of user contributions in search results. InfoSearch is currently in deployment with real users. An evaluation of the ranking factors available in InfoSearch is also presented through user studies.