Note in connection to the 2nd exercise of the 2nd part of the course project
23 novembro 2017, 13:59 • Bruno Emanuel Da Graça Martins
In the second exercise for the second part of the course project (i.e., "improving the graph-ranking method"), there is a typo in one the recommendations/suggestions that is provided.
The following sentence:
Edge weights based on a probabilistic model similar to a naïve Bayes classifier, which assigns a probability to each sentence conditioned on the entire document.
Should instead read:
Non-uniform prior weights based on a probabilistic model similar to a naïve Bayes classifier, which assigns a probability to each sentence conditioned on the entire document.
In what regards the implementation of this specific recommendation, students should attend to the class lecture introducing "classic information retrieval models," specifically the part that introduced "probabilistic language models for information retrieval," and that established a parallel between the naïve Bayes classifier and this particular information retrieval model.
It is also possible to implement an edge weighing strategy with basis on this general idea (i.e., estimate the probability of one sentence conditioned on the other), but the actual description that was included in the PDF with the exercises for the course project had indeed a typo.