Characterisation of visitors and description of their navigation behaviour using web log mining techniques

Abstract

The value of a company’s website depends on visitors performing actions that add value for the company. Those actions are called conversions. We present techniques for both characterising website visitors in terms of the conversions they make, and describing their navigation behaviour in an abstract way, with the aim of making them more amenable to interpretation. Existing web analytics techniques have not been designed to highlight the distinguishing characteristics of a class of visitors. There are no approaches for characterising classes of visitors that take into account specific business goals; further, the navigation behaviour of a visitor, let alone a class of visitors, is conveyed as a sequence of visited pages, without giving this an abstract meaning. In this thesis, we introduce a means of characterising website visitors. To find what the different segments of visitors have or do not have in common, we first separate visitor sessions in terms of conversions and then for each class we mine patterns to contrast one another. We also introduce a simplified description of visitor navigation behaviour. Our technique works by identifying subsequences of visited pages of common occurrence, called ``rules'', and then by shrinking a session replacing those rules with a symbol that is given a representative name. Further, we extended this to an entire class of visitors, creating a graph that collects the class sessions, summarising the class navigation behaviour and enabling an easier contrast of classes. Our results show that a few patterns are enough to characterise a visitor class; since each class is associated with a conversion, an expert can easily draw conclusions as to what makes two classes different from one another. Also, with our abstract representation, a session can be shrinked so that the behaviour of an entire visitor class can be depicted in a moderately small graph. Further work is concerned with incorporating information from other sales channels and completing the analysis provided by existing techniques.