SentiHawkes: a sentiment-aware Hawkes point process to model service quality of public transport using Twitter data

Responsive management of public transport nodes relies on constant monitoring of service quality. Social media content provides a unique opportunity to detect and monitor events impacting service quality in these nodes, as well as predicting future occurrences of such events. However, the confined geographic area of transport nodes exacerbates the sparsity of available feeds, raising two major challenges: limited observations—leading to biased models—and the asynchronous nature of observations—impeding the detection of causal patterns. Thus, this paper proposes a framework based on a multivariate Hawkes point process and sentiment analysis. The multivariate Hawkes point process allows effective modelling of events without making them discrete, hence it is less affected by data sparsity compared to time series models while enabling the prediction of how certain events can trigger future events. Besides, the extracted sentiments from social media feeds provide additional knowledge about passengers’ perception and thus, are used in our approach to strengthening the model. Experiments on a real-world dataset demonstrate the effectiveness of the model in identifying causal relations over the public transport nodes. They also show the efficacy of the proposed solution in predicting events over the limited context compared to state-of-the-art approaches.