Developing a Question-Answer-System for the Interactive Learning Platform goStudent
Question Answer Systems (QAS) have been the subject to research over the last decades without having ever lost importance due to the fact that similar questions are repeatetly asked over time and given answers remain significant notwithstanding. The necessity for such systems arises for different reasons, ranging from rising user satisfaction by keeping the response time in a reasonable time frame to enhancing the quality of service by providing the best answer for a given question and in order to improving the economical factor. A crucial aspect in the thriving field of QAS is the nature of the data on which the QAS operates. Unstructured data produced through human-computer interaction requires context-aware data cleansing. GoStudent being an interactive learning platform for adolescents allows pupils to ask questions which are then answered by tutors. Within its database, GoStudent hosts questions and answers whose language is aligned towards youth language.
The goal of this work is to detect question similarity by applying and evaluating various machine- and deep-learning algorithms. As the user-interaction is of chat-like manner spelling and grammatical errors are not seldom cases. Against this background great emphasis is dedicated on the data cleansing. Selecting features that capture the distinguishing properties of any question is another milestone.