Data to Provide Relevant Information

We are always seeking for relevant and useful information. However, the information provided to us generally does not conform to our information needs, mainly manifested from two aspects: One, when we search for some items, the number of candidate items is so large that targeting the most relevant items is difficult. Two, when our goal is to obtain some high-level and global type of information, the only information available is of raw and local type. To address the above two problems in providing relevant information, I propose to exploit and mine large-scale of heterogenous user-generated data, including social media messages, call detail records, photo tags, etc. On one side, those data contain rich information about users’ interests, opinions and behaviors, which can be further explored to provide contextual knowledge for content ranking and recommendation in the microscopic level (i.e. information filtering). On the other side, the local user-to-user interactions can be leveraged and aggregated to generate new indicators via statistical methods, to infer human behavior and phenomena in the macroscopic level (i.e. information transformation). My overall research framework is illustrated in Figure 1.