Design of hybrid machine learning and TF-IDF models to discard irrelevant reviews on public transport stations

Users' opinions on social media about city landmarks are valuable tools for the responsible authorities. However, if the name of a city landmark is similar to a slightly related but different item, then users may be confused and inadvertently comment on the wrong item. Public transport stations are a good example of this situation because the stations' names often refer to their location (e.g. a square, a neighbourhood, a hospital, etc.). In this paper, we use artificial intelligence models to develop a classification system that distinguishes reviews referring to the station itself from those that do not. To achieve this, we apply Natural Language Processing (NLP) techniques to numerically represent words and phrases, and artificial intelligence models to classify the text once it is numerically represented. Our experiments show that the combination of Term Frequency-Inverse Document Frequency (TF-IDF) and machine learning models, such as Support Vector Machine and Random Forest, yields the best results overall. To establish a precise setting for evaluating our system, we consider reviews on Google Maps about Madrid metro stations. However, our methodology should be easily extrapolated to other transport networks.