Járókelő is a platform that enables the citizens of Hungary to report and share problems they experience in their environment, like damaged trash cans in public spaces, faded pedestrian crossings or potholes. The issue, noticed by the users, is sent to the authorities that are responsible for its solution and asked to reply whether it can be fixed or not. The service, delivered by the organizations, is evaluated on the basis of the way they react to the problem. Although it is essential for each institution to get feedback on its performance, in this case it is challenging to objectively evaluate the replies and to make them measurable. Is the response of the institutions polite? Does it address the reporter of the problem personally? How quick is the answer? These are just some of the factors that influence one’s evaluation about the communication of the organizations. We developed a machine learning solution, an automatic classifier, which label each response of the organizations in a scale of 1 to 5. The more satisfied the user is, the highest the score is.
The answers of the representatives of the authorities have been evaluated by volunteers from Járókelő and annotators from Precognox to compile a training dataset. This dataset was used to teach a classifier algorithm, which selects the most valid features of classification (e.g. numbers of nouns per text, length of an answer, or presence of images).
As human beings, classification is an inevitable part of our daily lives. We all sail in the sea of information, which is made up of articles, emails and images, etc. To be able to process the tide of information, classification is needed. However, we can encounter two obstacles of automatic classification: the lack of training data, meaning the lack of labeled data, on the one hand; and the imbalanced nature of the available data, i.e. certain groups of data are overrepresented, while others are underrepresented, on the other hand.
Precognox offers a solution for automatic classification, which comprises the complex process of classification from searching data to integrating the trained algorithm into a software system.