BWA 2013 - 2013 Behavioral Web Analytics AAAI 2013 Fall Symposium

Web intelligence has sought to understand and predict Web behavior in order to improve a specific exogenous outcome such as the checkout of a shopping cart or improvement in meaningful search results. More interest is given now to understanding those behaviors themselves. For instance, it is of interest to understand individual and collective Web behaviors to further model and support knowledge extraction in applications such as cybersecurity, recommendation systems and human computation systems. This proposed symposium on behavioral Web analytics aims to bring together artificial intelligence researchers from the machine learning & planning community, the cognitive science community, the social computing community and the human-centered computing community to develop a more meaningful understanding of individual and group behavior in online environments.
Topics
Characterization of Web behavior: Web behavior has been characterized by the URLs visited as well as the tempo of those visits encapsulated in "clickstream" data. This data representation simplifies users' true online behavior. What are the other "actions" (e.g. scrolling, mouse dynamics) that could be relevant for characterizing Web behavior which could be captured in a browser? What are the indicators of social behavior on the Web? How should we capture and represent social behaviors such as commenting on or “liking” online material?
Categorization of Web behavior: Our online activity seems to be only our own. It is possible however to identify common online patterns of behavior. Some obvious common online behavior types include shopping and information seeking. What are the fundamental patterns of Web behavior? What are some unexpected or not so common patterns of Web behavior? How should we classify Web pages to reflect patterns of Web behavior?
Modeling Web behavior: Advances in Web technologies coupled with the prevalence of handheld and mobile devices in our daily lives have led to an explosive growth of technologically-enabled interactions between individuals and the Web. These interactions have not only created an enormous amount of multimedia data but also enable studying social and socio-cognitive phenomena. We can infer knowledge/intelligence from the online content and learn about individual and collective behavior from technology-enabled interactions. Findings from these studies could have implications in studying trust, situational awareness, social movements, flash-mobs, and disaster management, among others. Modeling Web behavior entails several questions, such as how to model and analyze data from the Web in a robust way? How can trust be modeled? How can behavioral changes be captured? How can we develop computational models that are feasible, scalable, and accurate to study these phenomena?
Authentification/Identification: Can Web behavior be used to uniquely identify someone as a signature or does authentication/identification entail a unique composition of primitive Web behaviors? Does the content of the Web page visited matter? Does the order of those visits matter? Can the tempo of clicks substitutes for keystroke authentication? What are the learning, planning, and cognitive methodologies used in constructing Web behavior profiles?
Inferring intent from Web behavior: How to determine intent from the observation of Web behavior? For example, how can "bullying" be determined through the observation of Web behavior? Consequences of Web behaviors are often indirect, so a problem is often how to distinguish "good natured" behavior from malicious intent, or casual browsing from goal-driven behavior. What AI tools and techniques are applicable here? Text mining or sentiment analysis alone do not always help in determining intent from Web behavior. However, intelligently leveraging information about composite Web behavior patterns, such as uploading a Youtube video, writing a blog post, posting a tweet, or cross-linking between different media channels could all be very helpful in inferring intent. How can one determine the composition of Web behaviors of interest?
Assessment of Web behavior: Because of its social aspects, assessment of Web behavior has mainly been measured through reputation scores. Are there other metrics for measuring influence that would apply? Can the assessment of Web behavior be reduced to a score comparable to a credit score to measure trustworthiness? In the absence of absolute ground truth about users' trustworthiness, how can the proposed models be objectively evaluated?