In general in text analysis, a high raw count for a term inside a text does not necessarily mean that the term is more important for the text. One of the most important ways to normalize the term frequencies is to weigh a term by how often it appears not only in a text, but also in the entire corpus.
The more a word appears inside a given text and doesn't appear too much across the whole corpus, it means that it's probably important for that specific text. However, if the term appears a lot inside a text, but also appears a lot in other texts in the corpus, it's probably not important for the specific text, but for the entire corpus, and this dilutes it's predictive power.
In IR, TF-IDF is one of the most popular term-weighting schemes and it's the mathematical implementation of the idea expressed in the preceding paragraph...
