The apparatus according to the present invention stores examples constituted by an article, a summary of the article and an evaluation of the summary in a solution data storage unit. Groups of sets of solutions and features are extracted from the stored examples and what kind of solution is the most straightforward for which case is learned using machine learning techniques, with the results of this learning being stored in the learning results data storage unit. Sets of features are then extracted from the inputted summary and what kind of solution is the most easily attained in the case of the extracted set of features is extrapolated and outputted as an evaluation.