This article addresses the problem of detecting crisis‐related messages on social media, in order to improve the situational awareness of emergency services. Previous work focused on developing machine‐learning classifiers restricted to specific disasters, such as storms or wildfires. We investigate for the first time methods to detect such messages where the type of the crisis is not known in advance, that is, the data are highly heterogeneous. Data heterogeneity causes significant difficulties for learning algorithms to generalize and accurately label incoming data. Our main contributions are as follows. First, we evaluate the extent of this problem in the context of disaster management, finding that the performance of traditional learners drops by up to 40% when trained and tested on heterogeneous data vis‐á‐vis homogeneous data. Then, in order to overcome data heterogeneity, we propose a new ensemble learning method, and found this to perform on a par with the Gradient Boosting and AdaBoost ensemble learners. The methods are studied on a benchmark data set comprising 26 disaster events and four classification problems: detection of relevant messages, informative messages, eyewitness reports, and topical classification of messages. Finally, in a case study, we evaluate the proposed methods on a real‐world data set to assess its practical value.

Organizations increasingly rely on digital tools, such as social media, to harvest public opinion on a variety of issues ranging from brand reputations to political debates. One area in which digital tools have great potential is in consultation on design of public spaces. Public organizations are expected to consult with users of a public space before, during and after occupancy. Digital tools could increase the effectiveness of this consultation because: (i) they are available 24 h and do not rely on the presence of researchers and (ii) data are collected and analysed using automated methods, allowing faster decision-making. Despite these potential advantages, little is understood about how digital and non-digital consultations compare in practice. This paper reports on a live trial with a prototype digital consultation tool called VoiceYourView, which was co-designed with stakeholders and used to canvass opinion on the refurbishment of a major metropolitan library. Over 6 weeks, we used a variety of methods to collect comment from 600 users. VoiceYourView solicits opinion in unprompted form—it does not ask specific questions as in a survey—so users can say anything they like at any time. VoiceYourView carries out real-time analysis of data collected and uses public display screens to summarize comments-to-date in situ. Our research methods focused on observational studies conducted ‘in the wild’: this allowed us to capture user interactions with VoiceYourView in a realistic setting. Our findings indicate that VoiceYourView encouraged more positive feedback than non-digital methods. We also find that positive comments in general contain less actionable information than negative comments, and unprompted comments have similar actionable content to prompted comments. We suggest that this pattern in soliciting more (traditionally difficult to obtain) positive comments with VoiceYourView versus traditional survey methods implies that digital consultation tools of this type can complement existing channels.

This paper reports on VoiceYourView, a kind of intelligent kiosk, which uses speech recognition and natural language processing to gather the public's creative input on the public space designs. Over a six week period, VoiceYourView was deployed in a public space and 2000 design critiques were collected from 600 people. The paper shows that people are capable of providing creative input on their environment using unstructured speech or text and that a good proportion of these comments are actionable. The paper also investigates the use of public displays to auto-summarize comments left by the public so far. Although there is anecdotal evidence that this encourages participation, an experiment found that filtering comments (e.g., to display only positive responses) had no effect on what people had to say.