Multi-label classification for identifying L labels of a document can be transformed to an L binary classification problem. In this approach, we pass a document into the L binary classifiers, where each are trained for identifying one of the L classes. The output of the L classifiers is merged to produce a vector of class labels to which the document belongs. Even for simple models such as decision trees, SVMs can be used for the binary classifiers.

Though this is the simplest approach, the disadvantage is that it assumes the categories are independent of each other, which may not always be the case. For example, in our Yelp review example, the ambiance and service categories may be correlated. We will now look at other approaches for handling this.