Report on the TREC 2003 Experiment: Genomic Track

After retrieval, this simple strategy yield to an important improvement of the average precision: from 17.02 up to 35.80. Task II. Our approach is based on argumentative structuring, i.e. classification of textual segments into argumentative classes. We see the task as a question-answering task using always the very same question. We take advantage of a classifier likely to predict the argumentative class of any textual segment with high accuracy (F-score = 85%). We observe that when not taken from the title, GeneRIFs are found -ranked by decreasing order- in: 1) CONCLUSION, 2) PURPOSE, 3) RESULTS, 4) METHODS. After sentence splitting, sentences are classified and ranked according to these four categories. On that basis, a second ranking is made based on the similarity with the title (45% of GeneRIFs; Dice baseline = 50.47%). Then, we compute a combined score for each of these features, setting a Dice-like threshold to decide whether we use the title or the best scored sentence as Gene...

• Type: Conference
• Year: 2003
• Where: TREC
• Authors: Patrick Ruch, Gilles Cohen, Frédéric Ehrler, Henning Müller, Giovanni Coray, Hatem Ghorbel, Vincenzo Pallotta, Christine Chichester
• Keywords: Argumentative, Argumentative Structuring, Textual Segment, TREC 2003, TREC 2008