Uma medida de similaridade textual para identificação de plágio em fóruns educacionais

Abstract

With the increasing use of technology as an educational support tool, the use of Virtual Learning Environment (VLE) has increased in recent years. These environments provide several tools to improve the interaction between teachers and students, where some examples are: forum, blog, wiki, among others. These tools have great potential for generating content, which can be used to aid in the process of teaching learning. However, due to the great amount of interactions between the students and the teacher, it is difficult for the teacher to evaluate and follow up all the material that is made available by the students. A tool that stands out in relation to the generation of collaborative content is the forum. Among the possible functionalities of the forums is the question of evaluation. Many distance disciplines use forum interaction as a form of student assessment. However, with the large amount of information posted on the tool, it often becomes impractical for the teacher to manually detect plagiarism in the responses. The fundamental basis for the creation of automatic plagiarism detection systems is the creation of a measure of similarity that can measure the relationship between two texts. The similarity between texts is important in several Natural Language Processing (NLP) applications, such as retrieving information, summarizing text, extracting information, and grouping text. For example, in retrieval of information, the similarity measure is used to assign a classification score between a query and the obtained text. Various measures of similarity between texts can be found; however, in general, they are language dependent. In the case of Portuguese, few measures have been found and most use only statistical techniques, not taking into account semantic aspects of texts. In addition, there are papers in the literature to identify plagiarism in activities, scientific articles or course completion work. However, when context is educational forums the identification of plagiarism becomes even more difficult mainly because of the size of the text and by not requiring a formal language. Therefore, this paper aims to propose a measure that calculates the similarity between sentences written in Portuguese taking into account the semantics of texts. This measure was evaluated on the basis of the ASSIN workshop 2016. The proposed measure achieved better results than the first place in the competition reaching 0.70 Pearson correlation and 0.47 mean squared error. In addition to this evaluation, a case study was carried out to evaluate similarity in postings of educational forums in a discipline of Computer Science. The results were evaluated by the teachers of the discipline who confirmed the effectiveness of the tool.