A High Accurate Multiple Classifier System for Entity Resolution Using Resampling and Ensemble Selection

Algorithm 1

Resampling algorithm.

Input:

(1) the dataset to be resampled:

% each instance is the field similarity vector of a

record pair, and the class label indicates whether the

corresponding record pair is match or non-match

(2) the ratio of resampling:

Initialization:

(3) , ,

Splitting the dataset into ambiguous and normal:

(4) get the instance number

(5) Split dataset into duplicate data and distinct

data

(6) Average each instance to get record similarity

vector and

(7) Calculate expectation and variance of and as

, and , respectively

(8) Calculate lower bound LB of as

and the upper bound UB of as UB =

(9) for

(10) If   % is the similarity of the

th instance

(11)     % ambiguous instances

(12) Else

(13)       % normal instances

Resampling:

(14) For

(15) Randomly select an instance from

(16)

(17) For

(18) Randomly select an instance from

(19)

(20) Order in random order

Output:

(21) the resampled dataset: