A Lexicon-based Approach for Sentiment Classification of Amazon Books Reviews in Italian Language

摘要

We present a system aimed at the automatic classification of the sentiment orientation expressed into book reviews written in Italian language. The system we have developed is found on a lexicon-based approach and uses NLP techniques in order to take into account the linguistic relation between terms in the analyzed texts. The classification of a review is based on the average sentiment strenght of its sentences, while the classification of each sentence is obtained through a parsing process inspecting, for each term, a window of previous items to detect particular combinations of elements giving inversions or variations of polarity. The score of a single word depends on all the associated meanings considering also semantically related concepts as synonyms and hyperonims. Concepts associated to words are extracted from a proper stratification of linguistic resources that we adopt to solve the problems of lack of an opinion lexicon specifically tailored on the Italian language. The system has been prototyped by using Python language and it has been tested on a dataset of reviews crawled from Amazon.it, the Italian Amazon website. Experiments show that the proposed system is able to automatically classify both positive and negative reviews, with an average accuracy of above 82%.