Service quality monitoring in confined spaces through mining Twitter data

Abstract

Promoting public transport depends on adapting effective tools for concurrent monitoring of perceived service quality. Social media feeds, in general, provide an opportunity to ubiquitously look for service quality events, but when applied to confined geographic area such as a transport node, the sparsity of concurrent social media data leads to two major challenges. Both the limited number of social media messages-leading to biased machine-learning-and the capturing of bursty events in the study period considerably reduce the effectiveness of general event detection methods. In contrast to previous work and to face these challenges, this paper presents a hybrid solution based on a novel finetuned BERT language model and aspect-based sentiment analysis. BERT enables extracting aspects from a limited context, where traditional methods such as topic modeling and word embedding fail. Moreover, leveraging aspect-based sentiment analysis improves the sensitivity of event detection. Finally, the efficacy of event detection is further improved by proposing a statistical approach to combine frequency-based and sentiment-based solutions. Experiments on a real-world case study demonstrate that the proposed solution improves the effectiveness of event detection compared to state-of-the-art approaches.

Depression Classification From Tweets Using Small Deep Transfer Learning Language Models

“…In [28], the authors proposed a model based on BERT and DenseNet which identifies multi-model tweets containing both image and text data during the disaster. Regarding the combination of fine-tuning BERT with aspect-based sentiment analysis, a model for event detection is proposed using Twitter data in [29].…”

Using textual volunteered geographic information to model nature-based activities: A case study from Aotearoa New Zealand

“…The geospatial component that VGI often contains in an explicit or implicit form allows researchers to explore a wealth of thematic and temporal information linked to various locations. This multi-dimensional information of unprecedented granularity, not available in more traditional spatial data, provides insights into human behavioral patterns such as urban mobility, as well as into the spatio-temporal unfolding of events such as natural disasters [55,61,77]. Much of this work is framed around the concept of place, whereby various forms of textual VGI are used to extract "soft" information in the form of local place knowledge and feelings about neighborhoods, model place properties, or identify potential for leisure activities, with topic modeling being a wide-spread methodology [32,40,52,71].…”

GeoComputation 2019 special feature

“…2 It truly embodied the aims of the conference series in other ways too, featuring state-of-the-art work at the convergence of multiple disciplines, from experienced to early-career researchers, including opportunities for presenting graduate student findings.The full proceedings of the 2019 conference can be found online. 3 There you will find extended abstracts and short papers on many and varied topics, including AI, machine learning, spatial data mining, computational movement analysis, cyber GIS, open geospatial, VGI, environmental modelling, remote sensing, surfaces, hydrology, qualitative spatial reasoning, geospatial text processing, urban and social spatial analysis, geosimulation, agents, geostatistics, uncertainty, and a themed session on novel spatiotemporal paradigms. As always with this conference series, it was wonderful to see the skilled use of advanced statistical, computational, and machine learning methods used to solve challenging and real geographical problems.For many of us involved, this event was our last in-person conference, prior to the COVID-19 outbreak that forced the cancelation of so many conferences, and the adoption of on-line delivery for others.…”

“…The full proceedings of the 2019 conference can be found online. 3 There you will find extended abstracts and short papers on many and varied topics, including AI, machine learning, spatial data mining, computational movement analysis, cyber GIS, open geospatial, VGI, environmental modelling, remote sensing, surfaces, hydrology, qualitative spatial reasoning, geospatial text processing, urban and social spatial analysis, geosimulation, agents, geostatistics, uncertainty, and a themed session on novel spatiotemporal paradigms. As always with this conference series, it was wonderful to see the skilled use of advanced statistical, computational, and machine learning methods used to solve challenging and real geographical problems.…”

“…This Special Feature section showcases three full-length papers that have been developed and extended from the conference proceedings and fully peer-reviewed by JoSIS. The papers cover interesting theoretical and practical ground, including (i) the study of dissimilarity and its role in determining segregation by Michael Dorman, Tal Svoray, and Itai Kloog [2], (ii) the relationship between the road network and the wayfinding of pedestrians by Debjit Bhowmick, Stephan Winter, Mark Stevenson, and Peter Vortisch [1], and (iii) the use of social media posts to monitor the service quality of public transport by Mohammad Masoud Rahimi, Elham Naghizade, Mark Stevenson, and Stephan Winter [3].…”