تحليل احساسات براي استخراج جنبه محصول

دانشگاه ولي عصر(عج) رفسنجان

كارشناسي ارشد

The internet has affected the lives of most people as they are able to communicate more while the communications, has changed the way they approach the decision making process. Hence, there is a genuine need to process the vast amount of opinions that are available on the web. Sentiment Analysis is done in three levels, namely document, sentence and aspect, the most difficult of which is arguably aspect analysis. The aspect sentiment analysis itself is comprised of two parts, which are aspect extraction and polarity detection respectively. To accomplish the tasks, four different groups of approaches are used. These approaches include topic, language structure, sequential and machine learning models. We utilize two sets of opinions from two sources for this research. Data from the first source is in English language and involves laptops and restaurant opinions. The second set of data, is comprised of opinions from Digikala website, which includes both formal and non formal opinions. We aim at proposing a new deep learning model titled “ram” to determine the polarity of the aspects that are extracted from these sources. The most influential parts of the proposed model are the output conditional random field, the word embedding, part of speech and the syntactic dependency. The proposed approach shows significant performances for the Persian language. Combining the aspect extraction and polarity detection approaches is a topic for future research.

56040

دانشگاه وليعصر رفسنجان

علي شكيبا

علوم كامپيوتر

سيد مجتبي صباغ