Analiza wydźwięku polskojęzycznych opinii konsumenckich ukierunkowanych na cechy produktu
Feature-based sentiment analysis of opinions in Polish

Summary/Abstract: The drawback of the kNN method is its sensitivities on irrelevant variables. K. Kira and L.A. Rendell and I. Kononenko have proposed a filter which evaluates variable importance using local information. A practical problem of the use of their algorithm is the choice of parameters (the number of iterations, the number of nearest neighbors and the threshold). In this paper we empirically verify the algorithm using real data and artificially generated variables without discrimination power. Consumer reviews are a special type of text documents due to their content – their main purpose is not to provide objective information, but to show a subjective attitude of its authors to the product or its components. The attitude presented in the opinion is called overtones. Opinion may refer to a product as a whole or its components. The aim of the paper is to present the authors’ method for automatic evaluation of features-concentrated opinions overtones. This task is realized by analyzing the words in the direct neighbourhood of the product’s characteristics found in the text. Sentiments of distinguished product’s components identified on the basis of opinion can be assigned to the appropriate parts of the product description tree and then processed according to the purpose of analysis.

• Details
• Contents