Data Mining: Practical Machine Learning Tools and ... - LIDeCC

8.6 FURTHER READING 361to contain attributes that are apparently highly predictive but neverthelessirrelevant, and specialized statistical tests are needed to compare alternativehypotheses. A third is that the iterative, improvement-driven development stylethat characterizes data mining applications fails. It is impossible in principle tocreate a fixed training-and-testing corpus for an interactive problem such asprogramming by demonstration because each improvement in the agent altersthe test data by affecting how users react to it. A fourth is that existing applicationprograms provide limited access to application and user data: often the rawmaterial on which successful operation depends is inaccessible, buried deepwithin the application program.Data mining is already widely used at work. Text mining is starting to bringthe techniques in this book into our own lives, as we read our email and surfthe Web. As for the future, it will be stranger than we can imagine. The spreadingcomputing infrastructure will offer untold opportunities for learning. Datamining will be there, behind the scenes, playing a role that will turn out to befoundational.8.6 Further readingThere is a substantial volume of literature that treats the topic of massivedatasets, and we can only point to a few references here. Fayyad and Smith(1995) describe the application of data mining to voluminous data from scientificexperiments. Shafer et al. (1996) describe a parallel version of a top-downdecision tree inducer. A sequential decision tree algorithm for massive diskresidentdatasets has been developed by Mehta et al. (1996). The technique ofapplying any algorithm to a large dataset by splitting it into smaller chunks andbagging or boosting the result is described by Breiman (1999); Frank et al.(2002) explain the related pruning and selection scheme.Despite its importance, little seems to have been written about the generalproblem of incorporating metadata into practical data mining. A scheme forencoding domain knowledge into propositional rules and its use for bothdeduction and induction has been investigated by Giraud-Carrier (1996). Therelated area of inductive logic programming, which deals with knowledge representedby first-order logic rules, is covered by Bergadano and Gunetti (1996).Text mining is an emerging area, and there are few comprehensive surveys ofthe area as a whole: Witten (2004) provides one. A large number of feature selectionand machine learning techniques have been applied to text categorization(Sebastiani 2002). Martin (1995) describes applications of document clusteringto information retrieval. Cavnar and Trenkle (1994) show how to use n-gramprofiles to ascertain with high accuracy the language in which a document iswritten. The use of support vector machines for authorship ascription is