Improving Word Association Measures in Repetitive Corpora with Context Similarity Weighting