Computer-implemented method for performing hierarchical classification
Given a number of records and a number of target classes to which these records belong to, a (weakly) supervised machine learning classification method leverages known possibly dirty classification rules, efficiently and accurately learns a classification model from training data, and applies the learned model to the data records to predict their classes.
1. A method for training a hierarchical classification model using (i) a hierarchy of classes, (ii) a collection of data records, (iii) a collection of classifiers each of which predicts, as represented by a predicted score, whether a data record is a member of one of the classes in the hierarchy of classes or any of its descendants, and (iv) a collection of training data which includes data records, each of which is labeled with a class in the hierarchy of classes, the method comprising a software program executed on a computer processor configured to perform the following steps:
(a) predicting, using the collection of classifiers, whether each data record in the collection of data records is a member of a plurality of the classes in the hierarchy of classes, thereby obtaining a predicted score for a plurality of classes for each data record in the collection of data records;
(b) computing an entropy of the predicted scores for each data record in the collection of data records;
(c) selecting a weighted random sample of the data records in the collection of data records for labeling, wherein the weight used in the weighted random sample is a function of the entropy computed for each data record in the collection of data records, wherein the data records selected for labeling are high impact questions;
(d) presenting each data record selected for labeling to an operator for the operator to label with the correct class from the hierarchy of classes, the operator thereby labeling the high impact questions;
(e) combining the labeled high impact questions with the training data, thereby expanding the training data; and
(f) building, using the expanded training data, the collection of classifiers each of which predicts, as represented by a predicted score, whether a data record is a member of one of the classes in the hierarchy of classes or any of its descendants, thereby training the hierarchical classification model.
2. The method of claim 1 where the entropy computed in step (b) is the Shannon entropy computed from the probability distribution obtained by normalizing the predicted scores for the classes of an individual data record into a probability distribution of class membership.
3. The method of claim 1 wherein in step (c), additive smoothing is applied to the weighted random sample.