next up previous
Next: Data Collection Up: MLA_Projects Previous: Estimating Body Fat

Predicting Text Relevance from Eye Movements

For the international challenge Inferring Relevance from Eye Movements, test subjects had to read a question on a computer screen and got to see ten possible answers to the question afterwards. One of the answers was correct, four answers were somehow relevant to the question, but not correct, while five sentences were completely unrelated to the question. During the experiment, the eye movements of the test persons were tracked. The task is to predict only from the eye movements, which of the 10 sentences was correct, relevant or irrelevant. More information about the challenge can be found at http://www.cis.hut.fi/eyechallenge2005.

At http://www.igi.tugraz.at/lehre/MLA/WS05/eyemovements.zip you can find already preprocessed datasets and a description of the data. You can use two of the sets (the training and the validation set) to train a good classifier for this task. Compare different algorithms, like e.g. Boosting and Bagging with a variety of weak learners and parameter settings. Evaluate them on the test set without using any test data for training. The best accuracy results for this data until now were obtained by our institute and are around 72% on the test set. Try to optimize your classifiers and see if you can do better. Document the exact values of the parameters for the algorithms that you used.


next up previous
Next: Data Collection Up: MLA_Projects Previous: Estimating Body Fat
Pfeiffer Michael 2005-12-20