Linking social, semantic and sentiment analyses to support modeling transit customers’ satisfaction: Towards formal study of opinion dynamics
place - north america, place - urban, planning - surveys, planning - methods, ridership - attitudes, technology - passenger information
Transit ridership, Social media, Sentiment analysis, Opinion dynamics, Public policy
Modeling the opinion dynamics of transit system riders is key to understanding their needs, motivations and sensitivities for service changes. This can make planning and operations processes to match user profile, hence promoting more transit usage with positive results on sustainable development. This research work investigated the triangulation of social (how users are related to each other) and semantic (what issues/topics are on user minds) network analysis as well as sentiment analysis (how do they feel about these topics) of social media interactions for supporting better underrating of transit user opinions and levels of satisfaction. We used data from the Twitter account of TransLink (Vancouver transit agency); tracked them over 11 month period; compared them to similar data from two other Canadian cities; and, in so doing, contrasted the results in days with disruptions and days with normal operations. Our research work developed a methodology that can be guide agencies in this regards. First, the social network should be qualified as a viable community (small world), not just a random graph (with sparsely connected nodes). It should be clustered into sub-communities based on their connectivity and the issues they discuss. Second, we used a lexicon for customer satisfaction to detect topics in tweets. Linking topics to sentiments (user view of them) can be used as indicator for their opinion. Studying the variations of topics and sentiment across sub-communities, we can observe the origination of ideas. By tracking the evolution of the social network, we could study the potential impacts of social interaction (possibly learning) on opinion dynamics. In the specific case of Vancouver data, sentiment was negative, indicating a lower levels of satisfaction across all topics. However, this negativity should be taken in context—many social media interactions (especially Twitter) have negative sentiment. Interestingly, analysis of sentiment levels in days with disruption, particularly those related to public safety incidents, showed lower levels of negative attitude. This contrast could be an indirect measure of trust: while customers may not be satisfied with services, they trust that the agency is doing its best to protect them. Analysis of customer social networks showed that the sub-network of the most influential players closely matched of the full network in terms of topics and sentiment. Analysis of sub-communities shows that the network is not a venue for disseminating information by the agency. Many sub-groups, especially the ones we call medium-size groups (defined in our work as having between a 1000 and 10,000 nodes), portray a more of inter-customer interactions. It seems this is where discussions and topics evolve. The results of our analysis is specific to Vancouver and to the data and period investigated. They should not be generalized given the evolutionary nature of social media contents and the dynamics of opinion. Our methodology, however, can be used as benchmark or a starting point to conducting similar analysis in other agencies. The interpretation, as always, will remain context-sensitive (influenced by local conditions). Transit agencies should conduct such analysis frequently; experiment with adjusting its steps and the proposed algorithms to their local conditions; and establish a mechanism for the frequency of such analysis. More importantly, they should always triangulate the results of such analysis against findings from other sources of data such as community meetings and customer surveys.
Permission to publish the abstract has been given by Elsevier, copyright remains with them.
El-Diraby, T., Shalaby, A., & Hosseini, M. (2019). Linking social, semantic and sentiment analyses to support modeling transit customers’ satisfaction: Towards formal study of opinion dynamics. Sustainable Cities and Society, Vol. 49, article 101578.