Author attribution.

Suppose that the N feature n-vectors x (1) , . . . , x (N) are word count histograms, and the labels y (1) , . . . , y (N) give the document authors (as one of 1, . . . , K). A classifier guesses which of the K authors wrote an unseen document, which is called author attribution. A least squares classifier using regression is fit to the data, resulting in the classifier

For each author (i.e., k = 1, . . . , K) we find the ten largest (most positive) entries in the n-vector ßk and the ten smallest (most negative) entries. These correspond to two sets of ten words in the dictionary, for each author. Interpret these words, briefly, in English.