Rather than go into the details of all of these different classifiers, we will focus
our attention on how sentiment classifiers work. As with our previous examples,
let us consider how a person would identify the sentiment of some piece of text.
For a majority of cases, we use vocabulary clues in order to determine the sentiment.
For example, a positive digital camera review would likely contain words
such as “great”, “nice”, and “amazing”. On the other hand, negative reviews would
contain words such as “awful”, “terrible”, and “bad”. This suggests one possible solution
to the problem, where we build two lists. The first list will contain words
that are indicative of positive sentiment, and another list will contain words indicative
of negative sentiment. Then, given a piece of text, we could simply count
the number of positive words and the number of negative words. If there are more
positive words, then assign the text a positive sentiment label. Otherwise, label it
as having negative sentiment. Even though this approach is perfectly reasonable, it
turns out that people are not very good at creating lists of words that indicate pos-