A context-aware approach based on self-organizing maps to study web-users' tendencies from their behaviour

In the context of a highly volatile web of uneven quality, the identification of content deemed valuable by end users is of paramount importance. Where page content undergoes rapid change, this issue is particularly challenging. Web browsing activity represents a unique source of context by which the value of web pages can be determined via an assessment of individual user interactions, such as scrolling, clicking, saving and so forth. Over time, this data set forms a pattern of activity which can be mined for meaning. In this paper we present an approach to web content, based on Kohonen mapping, used to generate a topological model of users' behaviour over web-pages. Each web-document can thus be represented as a semantic map built by adopting unsupervised techniques where similar users' behaviour are mapped close together, with identification of information stability emerging as a by product of the identification of similarity in user activity over content. In this model, the more similar the outputs of the map for each user who has endorsed a web-page, the more the web site is considered current or in context with changing information. We illustrate the potential application of this approach to our ongoing work in social search.