近年來，服務型政府的概念被廣為接受，各個城市的專門處理市民疑難雜症的市民服務專線也隨之成立，此專線目的為提供一個便捷管道讓民眾反映社會中需要公權力介入解決的問題及建議，讓政府人員更能掌握城市運作情形。直至今日世界各城市皆有類似的管道供市民使用。面對一天有成千上萬則通報案件，民眾最關心的是政府處理案件的速度及效率，而延宕處理案件將導致市民抱怨。為了解決此問題，我們提出了使用Wide-Deep-Recurrent (WDR)架構的機器學習方法來預測市民服務專線的處理時間，此方法可以有效地結合時空間資訊、通報案件文字敘述、處理單位行經路徑的社會結構以及案件之間前後的相關性。在廣度學習上我們採用了XGBoost模型並提出了各種有效的提空間特徵做結合，而在深度學習上，我們同時對於文本數據及處理單位調度路線提出Embedding，另外我們也運用LSTM模型處理案件和先前案例之間的相關性。我們將此方法應用於台灣高雄市的1999市民服務專線中並與其它機器學習模型進行比較。實驗結果顯示，我們所提出的WDR架構能有效地結合所擷取出的相關特徵並能提升整體效能約30%。預測案件處理時間不僅可以讓市民了解問題解決所需的時間，更可以讓公務人員當作資源分配的參考依據，以便在將來提供更高效的服務此舉能提升市政服務滿意度，增加市民對政府的信心。

Nowadays, city governments have provided convenient ways for citizens to reflect daily affairs and opinions in society. The public service call is one of a direct way to collect citizen's needs. However, there are thousands of cases of public service calls happened during one day and citizens may care about the processing speed of handling municipal services from the government. More unfortunately, the delay in dealing with these calls will cause citizens to complain. In order to solve these issues, we propose a novel tool to predict the processing time for each call. The proposed tool can not only let citizens roughly know how long it needs to take to deal with an event, but also remind civil servants if it might take too long to process. The proposed tool can effectively combine the temporal, spatial, text information, and the correlations between unsolved events, using the Wide-Deep-Recurrent framework to predict the response time of the public service calls. The wide module in the framework focuses on making an initial prediction based on case properties. We also introduce two embedding ways to represent text description and the dispatch route in the deep model. In addition, the correlations between the current case and previous cases are handled in the recurrent model. We apply our three-in-one prediction framework on 1999 public service calls in Kaohsiung, Taiwan, and compare it with several baseline models and state-of-art machine learning models. The experimental results show that our model is much memorization and generalization. As a result, our framework could not only help citizens to estimate the cost time of queried cases from users but also help the government to understand the priority of cases in order to give much efficient service in the future. Moreover, our framework is general to be applied to other public service calls in worldwide cities.

[1] C. Alexopoulos, Z. Lachana, A. Androutsopoulou, V. Diamantopoulou, Y. Charalabidis, and M. A. Loutsaris, "How machine learning is changing e-Government," in Proceedings of the 12th International Conference on Theory and Practice of Electronic Governance, 2019, pp. 354-363.
[2] T. Chen and C. Guestrin, "XGBoost: A Scalable Tree Boosting System," presented at the Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, California, USA, 2016. [Online]. Available: https://doi.org/10.1145/2939672.2939785.
[3] H.T. Cheng et al., "Wide & Deep Learning for Recommender Systems," presented at the Proceedings of the 1st Workshop on Deep Learning for Recommender Systems, Boston, MA, USA, 2016.
[4] D. DeFazio, A. Ramesh, and A. Seetharam, "NYCER: A Non-Emergency Response Predictor for NYC using Sparse Gaussian Conditional Random Fields," in Proceedings of the 15th EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services, 2018, pp. 187-196.
[5] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, "Bert: Pre-training of deep bidirectional transformers for language understanding," arXiv preprint arXiv:1810.04805, 2018.
[6] R. Eshleman and H. Yang, ""Hey #311, Come Clean My Street!": A Spatio-temporal Sentiment Analysis of Twitter Data and 311 Civil Complaints," 2014 IEEE Fourth International Conference on Big Data and Cloud Computing, pp. 477-484, 2014.
[7] M. W. Gardner and S. Dorling, "Artificial neural networks (the multilayer perceptron)—a review of applications in the atmospheric sciences," Atmospheric environment, vol. 32, no. 14-15, pp. 2627-2636, 1998.
[8] S. Hochreiter and J. Schmidhuber, "Long short-term memory," Neural computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[9] C. Kontokosta, B. Hong, and K. Korsberg, "Equity in 311 reporting: Understanding socio-spatial differentials in the propensity to complain," arXiv preprint arXiv:1710.02452, 2017.
[10] E. Lee, S. Lee, K. S. Kim, V. H. Pham, and J. Sul, "Analysis of Public Complaints to Identify Priority Policy Areas: Evidence from a Satellite City around Seoul," Sustainability, vol. 11, no. 21, p. 6140, 2019.
[11] X. Liu, Y. Liu, and X. Li, "Exploring the Context of Locations for Personalized Location Recommendations."
[12] L. E. Nugroho, L. Lazuardi, and A. S. Prabuwono, "Emergency alert prediction for elderly based on supervised learning," in 2016 1st International Conference on Biomedical Engineering (IBIOMED), 2016: IEEE, pp. 1-6.
[13] X. Qiu, L. Zhang, Y. Ren, P. N. Suganthan, and G. Amaratunga, "Ensemble deep learning for regression and time series forecasting," in 2014 IEEE symposium on computational intelligence in ensemble learning (CIEL), 2014: IEEE, pp. 1-6.
[14] Z. Tariq, S. K. Shah, and Y. Lee, "Smart 311 request system with automatic noise detection for safe neighborhood," in 2018 IEEE International Smart Cities Conference (ISC2), 2018: IEEE, pp. 1-8.
[15] K. Wagstaff, C. Cardie, S. Rogers, and S. Schrödl, "Constrained k-means clustering with background knowledge," in Icml, 2001, vol. 1, pp. 577-584.
[16] Z. Wang, K. Fu, J. Ye, and Y. Guo, "Learning to Estimate the Travel Time
[17] Y. F. Zha and M. Veloso, "Profiling and prediction of non-emergency calls in NYC," in Workshops at the Twenty-Eighth AAAI Conference on Artificial Intelligence, 2014.