Q-WordNet, is a lexical resource consisting of the WordNet senses that are automatically classified by their Positive and Negative polarity. Although originally built with WordNet 3.0, four versions are offered, based on Wordnet 1.6, 1.7, 2.0 and 3.0. The one based on WordNet 2.0 is used for comparison and evaluation with respect to SentiWordNet. Versions from WordNet 1.6 and 1.7 are offered for compatibility with other lexical resources.
Polarity classification amounts to decide whether a text (sense, sentence, etc.) is associated to a positive or negative connotations. This task is becoming important for determining opinions about commercial products, on companies reputation management, brand monitoring, or to track attitudes by mining online forums, blogs, etc. Inspired by work on classification of word senses by polarity in SentiWordNet, and taking WordNet as a starting point, we create Q-WordNet, but instead of applying supervised classifiers, we decided to effectively maximize the linguistic information contained in WordNet (by human annotators). An intrinsic quantitative evaluation as a binary classification task shows good improvement with respect to SentiWordNet.
Q-WordNet is ongoing work. Although it may be used in its current form for lexical sentiment analysis tasks (like Affective Task in SemEval-07), it may also be used as a training set for supervised classifiers that could subsequently be applied for the improvement of Q-WordNet, or for the creation of alternative resources containing positive and negative polarity.
Rodrigo Agerri (ragerri at gmail.com)
GSI Group, Universidad Politécnica de Madrid (UPM)
Working at MAVIR S-505/TIC/0267