Neural Networks: An Overview - Neural Networks: Methodology and Applications - page 57

Fig. 1.37. Neural estimate of the election probability as a function of the total
campaign expenses (data for the 1993 elections)
of being elected. Figure 1.37 shows the probability of election as a function of
the total expense.
That application, although in the area of classification, is somewhat dif-
ferent from the previous two applications: in the latter, the classifier was
intended to assign an existing pattern to a class, while, with high probability,
the actual class of the unknown pattern would never be known with absolute
certainty. In the present application, the situation is different, since the class
of each pattern (candidate running for election) will be known unambiguously
immediately after the election. This application falls in the class of forecasting
by simulation: in order to optimize the probability of success, the candidate
can estimate his success probability as a function of the strategy of expenses
that he uses, and derive from those results the strategy that is most suitable
to his situation.
In the next chapters, some sections will be devoted to forecasting by sim-
ulation: it will be shown that it is an area of excellence of neural networks.
1.4.5 An Application in Data Mining: Information Filtering
The rapid increase of the volume of available information, especially by elec-
tronic means, makes it mandatory to design and implement e cient informa-
tion filtering tools, which allow the user to access relevant information only.
Since such tools will address the needs of professionals, they must be reliable
and user-friendly. The user can access relevant information either by being
provided with full texts by the machine (text search), or by being provided
text excerpts or answers to questions (information extraction).
Text categorization, also known as filtering, consists in finding, in a text
corpus (e.g., of press releases, or of Web pages), the texts that are relevant
to a predefined topic. The user can thus be provided with information that is
important for his professional duties. In a machine-learning based system, the
user does not express his topic of interest through a query, but by providing a