As a similarity function, we use cosine similarity weighted with TF*IDF.

We define sim(efl, ej/lx) as the cosine similarity between excerpts 63-; from topic 253- and ej/l/ from topic if.

If excerpts ejl and ej/l/ have cosine similarity

For each topic present in the human-authored article, the Oracle selects the excerpt from our full model’s candidate excerpts with the highest cosine similarity to the human-authored text.

During classification cosine similarity is measured between the feature vector of the classified document and the expanded vectors of all categories.

The first avoids any expansion, classifying documents based on cosine similarity with category names only.

We also examined another filtering score, the cosine similarity between the vectors representing the two rule sides in LSA (Latent Semantic Analysis) space (Deerwester et al., 1990).