[Solomonov Seminar] 177. Solomonov seminar

[Marko Grobelnik]

Vabim vas na 177. Solomonov seminar, ki bo v torek 30. januarja,
ob 13:00 uri v Oranzni predavalnici IJS (sejna soba E8).

Predavala bo Annalisa Appice iz univerze v Bariju
o analizi relacijskih dreves.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Annalisa Appice, Università degli Studi di Bari
Mining Relational Model Trees

Multi-Relational Data Mining (MRDM) refers to the

process of discovering implicit, previously unknown

and potentially useful information from data scattered

in multiple tables of a relational database. MRDM is

necessary to face the substantial complexity added to

data mining tasks when properties of units of analysis

to be investigated are potentially affected by

attributes of related units of analysis eventually of

different types and naturally modeled to yield as many

tables as the number of object types. Regression is a

fundamental task in MRDM where the goal is to examine

samples of past experience with known continuous

answers (response) and generalize future cases through

an inductive process. Following the mainstream of

MRDM research, Mr-SMOTI resorts to the structural

approach in order to recursively partition data stored

in a tightly-coupled database and build a

multi-relational model tree that captures the linear

dependence between the response variable and one or

more explanatory variables of both the reference

objects and task-relevant objects. The model tree is

top-down induced by choosing, at each step, either to

partition the training space (split nodes) or to

introduce a regression variable in the linear models

to be associated with the leaves (regression nodes).

Internal regression nodes contribute to the definition

of multiple models and capture global effects, while

straight-line regressions with leaves capture only

local effects. The tight-coupling with the database

makes the knowledge on data structures (e.g., foreign

keys) available free of charge to guide the search in

the multi-relational pattern space.