| المستخلص: |
During the recent years, microblogging and social media have become very popular where millions of people post short text about different things. Topics range from personal life and work, to current events, news, and interesting observations and political thoughts. Education institutes become aware of the benefits engaging in such technology, and many instructors use social media in teaching courses they offer. Courses adopting social media in the learning process allow students to discuss with each other and with their teacher different topics and express their opinions on various aspects of these topics. The huge amount and variety of opinions generated out of these discussions create new opportunities for assessing teaching courses. Manual methods for analyzing opinions in these huge amount of data are infeasible. Sentiment analysis is a research field that focuses on automatically identifying the subjectivity and the polarity (e.g. positive or negative) of a given text on an entity or a topic. It is a classification problem, where learning algorithms are used. Most of previous works focus on using supervised algorithms, however such algorithms are very expensive since we need to manually annotate a large amount of data for training the classifiers, in addition it is domain dependent (e.g. products, movies, politics, etc.). Besides, certain characteristics of social media content introduce challenges in their analysis. Informal English blended with abbreviations, slangs and context specific terms; lacking in sufficient context and regularities and delivered with an indifferent approaches to grammar and spelling, all at the top of these characteristics. Most of previous works on sentiment analysis tackle domains such as economic, products, movie reviews, and political domain. There is a paucity of literature in the education domain. Our research is a contribution to this field. In particular, we propose a sentiment analysis prototype for microblogs posted in learning activities. The prototype automatically classifies microblogs of learning activities into positive and negative with less costs in terms of learning requirements. Our approach aims to achieve this objective using a novel combination of features extraction and engineering methods, and using a semi-supervised sentiment classification model based on label propagation algorithm. The results returned of the experiments we conducted to evaluate the model were competitive to existing works. The F- measures of our approach using different datasets ≈ 80 %.
|
