Adaptive Information Filtering seeks a solution to the problem of information overload through a tailored representation of the user's interests, called user profile, which constantly adapts to changes in them. Evolutionary Algorithms have been proposed as a solution to the problem of profile adaptation, but the relevant attempts have not produced successful real world applications. In this paper, we argue that Adaptive Information Filtering is a complex and dynamic problem not easily addressed with Genetic Algorithms and Memetic Algorithms that adopt weighted keyword vector for profile representation. We discuss the theoretical issues and provide experimental evidence showing that such an approach suffers due to the large number of dimensions in the underlying vector space. The genetic operators cannot randomly produce the right combinations of keyword weights, given the very large number of possible combinations. With the current work, we wish to reanimate the interest in Evolutionary Information Filtering. Profile adaptation is a challenging problem with no established solution and biologically inspired solutions have still an important role to play in solving it. This however requires experimental methodologies that reflect the complexity and dynamics of the problem. This paper is part of ongoing research work in this direction.

The Vector Space Model has been and to a great extent still is the de facto choice for profile representation in content-based Information Filtering. However, user profiles represented as weighted keyword vectors have inherent dimensionality problems. As the number of profile keywords increases, the vector representation becomes ambiguous, due to the exponential increase in the volume of the vector space and in the number of possible keyword combinations. We argue that the complexity and dynamics of Information Filtering require user profile representations which are resilient and resistant to this "curse of dimensionality". A user profile has to be able to incorporate many features and to adapt to a variety of interest changes. We propose an alternative, network-based profile representation that meets these challenging requirements. Experiments show that the network profile representation can more effectively capture additional information about a user's interests and thus achieve significant performance improvements over a vector-based representation comprising the same weighted keywords.

Term dependence is a natural consequence of language use. Its successful representation has been a long standing goal for Information Retrieval research. We present a methodology for the construction of a concept hierarchy that takes into account the three basic dimensions of term dependence. We also introduce a document evaluation function that allows the use of the concept hierarchy as a user profile for Information Filtering. Initial experimental results indicate that this is a promising approach for incorporating term dependence in the way documents are filtered.