Adaptive Information Filtering - Text Mining: Classification, Clustering, and Applications - page 194

community. Measures that consider novelty or properties of a document have
also been proposed by researchers (65).
8.3 Standard Retrieval Models and Filtering Approaches
In this section, we first review some existing information retrieval models
since most of them have been adapted, or can be adapted, for the information
filtering task. Then we review three common filtering approaches for learning
user profiles from explicit user feedback.
We introduce these existing approaches and their drawbacks here, so that
the readers can get a better understanding of the common practices in adap-
tive filtering. This section also provides the context and motivation of the
research work described in the following sections. As there is a large amount
of literature about standard retrieval models and filtering approaches, we will
only review them concisely. For more detail about these models, the readers
are referred to other papers or topics.
8.3.1 Existing Retrieval Models
Information filtering has a long history dating back to the 1970s. It was
created as a subfield of the more general information retrieval field, which
was originally established to solve the ad hoc retrieval task. 3 For this reason,
early work tended to view filtering and retrieval as “two sides of the same
coin” (9). The duality argument is based on the assumptions that documents
and queries are interchangeable. This dual view has been questioned (49) (12)
by challenging the interchangeability of documents and queries due to their
asymmetries of representation, ranking, evaluation, iteration, history, and
statistics. However, the influence of retrieval models on filtering is still large,
because the retrieval models were comparatively well studied and the two tasks
share many common issues, such as how to handle words and tokens, how to
represent a document, how to represent a user query, how to understand
relevance, and how to use relevance feedback. So it is worthwhile to look at
various models used in IR and how relevance feedback is used in these models.
In the last several decades, many different retrieval models have been de-
veloped to solve the ad hoc retrieval task. In general, there are three major
classes of IR models:
3 Historically, information retrieval was first used to refer to the ad hoc retrieval task, and
then was expanded to refer to the broader information seeking scenario that includes filter-
ing, text classification, question answering, and more.