DuOMAn

We need a reasonably large lexicon of Dutch "opinionated" words, i.e., words that serve as subjectivity clues in text (e.g., for English, excellent, wow or admire are, often, positive clues, whereas awful, dislike or shame are negative clues).

That's what we tried. We took a large publicly available English opinionated lexicon and translated the lists of "positive" and "negative" words into Dutch using Google translate. Of course, the result is quite noisy: many words are not translated, many are translated incorrectly (for our purposes).

To get cleaner Dutch word lists, we used the Dutch Wordnet. We viewed Wordnet (essentially, a network of inter-word relations) as a... network of inter-word relations (to be precise, relations between words and synsets, i.e., set of synonymous words). Each word (node in the network) was assigned a weight 1 or -1 if it occurred in the positive (resp., negative) word list we got from Google Translate; other words were assigned weight 0. Then, we ran a weighted version of PageRank on this network, using Wordnet relations as links between nodes. Finally, we checked the result for nodes with highest and lowest PageRank score. This simple method gave us surprisingly clean lists of positive and negative words. Here are the top words:

In total, we got 4466 positive words and 7091 negative. After a bugfix, all (except for 15) words in the Wordnet were assigned a score, of course. The resulting rankings still need a careful evaluation, but the lists do look good, given the simplicity of the method. Thanks to OpinionFinder for the English lexicon, and to Google for the translation service and the PageRank!