Occupants' Comfort at Urban Scale: Analyzing Citizens’ Opining Using Convolutional Neural Networks

Moving more and more toward urbanization, citizens’ comfort is widely considered as a potential source for demand flexibility. In this regard, getting feedbacks and engaging occupants is recognized as a key component to ensure that urban spaces are capable of providing acceptable functionality. In this paper, we mainly focus on the ―parks‖ in the cities under the three main categories of amenities, visual and acoustics. In this way, the comfort level of the citizens is evaluated by using the citizens’ feedbacks. Accordingly, more than 10,000 reviews about different parks in Canada are gathered from the google map. On the next step, by applying the latent Dirichlet allocation (LDA), the google map reviews are classified into various topics. On the other hand, each review has a rate between 1 to 5 in google map which indicates the amounts of the importance of the issue. By applying convolutional neural networks (CNN), the reviews classified based on their polarities (positive, negative and neutral). Upon improving its accuracy, this classifier can be used for labeling other texts and reviews generated by citizens that do not have explicit rating score (such as tweets). Moreover, by analyzing reviews assigned to each topic and through the polarity of the comments, the most significant issues related to Canadian parks in terms of citizens’ comfort were detected and discussed. The result of this paper can provide a break down of comfort conditions assessment (from perspective of space end-users), to be used as an input by urban space designers.