Explainable Multilingual Civic Complaint Resolution System

Abstract

Civic complaint management plays a crucial role in smarturbangovernance,yetmanyexistingplatformsfallshortin multilingual support and explainability. This paper presents an ExplainableMultilingualCivicComplaintResolutionSystemthat processes complaints in English, Hindi, and Hinglish and routes them to three key municipal departments: Sanitation, Water Supply,andTransportation.ThearchitectureusesMuRIL-based embeddings for multilingual category classification and employs XGBoosttopredicturgency.Toenhancetransparency,SHAPexplanationsareprovidedforbothcategoryandurgencydecisions, offering human-interpretable outputs for citizens and officials. A balanced dataset of 15,177 instances, derived from real-world complaints translated using MarianMT and indic-transliteration, demonstratesrobustperformanceandinterpretableresults, highlighting the promise of explainable multilingual complaint resolution in civic applications.

Introduction

This work proposes an AI-based multilingual civic complaint resolution system designed to improve how urban local bodies handle large volumes of citizen complaints. It addresses key challenges such as delayed manual processing, multilingual inputs (English, Hindi, Hinglish), and lack of transparency in AI systems.

Key Features

• Automatic classification of complaints into three departments: Sanitation, Water Supply, and Transportation
• Urgency prediction (Critical, High, Medium, Low) for prioritization
• Explainable AI (SHAP) to provide transparent decision-making
• Role-based web system (citizens, officials, administrators) implemented using Streamlit

Methodology

• Uses MuRIL (transformer-based model) for multilingual text understanding and category classification
• Uses XGBoost for urgency prediction by combining text embeddings with structured features (keywords, time, population impact)
• Applies SHAP to explain both classification and urgency decisions
• Incorporates geospatial tagging for location-based routing

Dataset

• Based on real complaints from the I Change My City platform
• Expanded into English, Hindi, and Hinglish using translation and transliteration
• Final dataset: 15,177 complaints, balanced across categories and languages

Results

• High accuracy across languages:
  • English: 96%, Hindi: 96%, Hinglish: 94%
• Overall classification performance: ~97% accuracy, outperforming traditional models
• Urgency prediction achieved ~99% accuracy
• Minor errors occurred due to overlapping keywords between categories

System Implementation

• Built with Python, Streamlit, PyTorch, XGBoost, SHAP
• Uses SQLite database and role-based access control
• Ensures security through encryption, validation, and session management

Conclusion

Thispaperpresentedanexplainablemultilingualciviccomplaint resolution system designed to improve transparency, efficiency, and fairness in municipal grievance handling. The proposed framework integrates MuRIL-based semantic embeddings for multilingual category classification with an XG-Boost model for structured urgency prediction. By combining contextual language representations with domain- specific structured features, the system achieves reliable automated routing and priority estimation across English, Hindi, and Hinglish complaints. Inaddition,theincorporationofadedicatedlocationcapture module enhances the practical applicability of the system by enabling geospatial tagging of complaints through interactive mapselectionorstructuredmanualentry.Theinclusionofspatial metadata supports location-aware routing, regional prioritization, and future geospatial analytics, thereby strengthening the operational value of the framework. A key contribution of this work lies in the integration of SHAP-basedexplainabilityforbothcategoryclassificationand urgencyprediction.Theinclusionofinterpretableexplanations enhances trust among citizens and administrative officials by clearlyindicatingthetextualandstructuredfactorsinfluencing each decision. Experimental evaluation demonstrates that the proposed architecture achieves strong classification accuracy while maintaining robustness across multilingual inputs. The modular design further ensures scalability and adaptability for real-world deployment. Futureworkwillfocusonextendingthesystemtoward real municipal integration through API-based complaint ingestion and live dashboard synchronization. Additional improvements include advanced geospatial clustering of complaints for hotspot detection, incorporation of real-time traffic and environmental signals for improved urgency modeling, and feedback-driven retraining mechanisms to continuously refine model performance. Expanding language coverage to include additional regional Indian languages and exploring transformer-based multilingual fine-tuning strategies also represent promising research directions. Finally, large-scale field validation in collaboration with municipal authorities will be pursued to evaluate long-term system impact and operational effectiveness.

References

[1] T.ChenandC.Guestrin,“XGBoost:AScalableTreeBoostingSystem,”in Proc. 22nd ACM SIGKDD Int. Conf. Knowledge Discovery and DataMining (KDD), 2016. [2] S. M. Lundberg and S.-I. Lee, “A Unified Approach to InterpretingModel Predictions,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS),2017. [3] J.Devlin,M.-W.Chang,K.Lee,andK.Toutanova,“BERT:Pre-trainingof Deep Bidirectional Transformers for Language Understanding,” inProc. NAACL-HLT, 2019. [4] S. Kumar, A. Chauhan, A. Goyal, S. Jain, and others, “MuRIL:Multilingual Representations for Indian Languages,” arXiv preprintarXiv:2103.10730, 2021. [5] R. Gupta et al., “Mining Railway Grievances on Twitter for Efficient E-GovernanceinIndia,”inProc.IEEEInt.Conf.Innov.Technol.Commun.Eng. (ICITCE), 2023, pp. 1–6. [6] A. Kumar et al., “AI-Based Solution To Enable Ease of GrievanceLodging and Tracking for Municipal Corporations,” in Proc. AtlantisPress Int. Conf. Smart Cities, 2024, Art no. 126012595. [7] National Institute of Urban Affairs, “Municipal Grievance Redressal(MGR):FrameworkandImplementation,”NewDelhi,India,Rep.,Apr.2024. [8] Government of India, “National Urban Digital Mission: GrievanceRedressal Framework,” Ministry of Housing and Urban Affairs, NewDelhi, India, Tech. Rep., 2024. [9] Smart Cities Mission, “AI Integration Guidelines for Urban LocalBodies,” Ministry of Housing and Urban Affairs, New Delhi, India,Guideline Doc., 2024. [10] M. Khan et al., “Multilingual Sentiment Analysis for Civic FeedbackUsingIndicBERT,”inProc.IEEERegion10Conf.(TENCON),2024,pp.789–794. [11] S. Nair et al., “SHAP-Based Interpretability for Tree Ensemble Modelsin Public Service Applications,” in Proc. Int. Conf. Mach. Learn. Appl.(ICMLA), 2024, pp. 567–573. [12] V.KumarandR.Singh,“UsingXGBoostandSHAPtoExplainCitizens’DifferencesinMunicipalServiceSatisfaction,”PeerJComput.Sci.,vol.10, Art no. E1929, Jul. 2024. [13] S.SharmaandR.Patel,“SmartCivicComplaintAnalyzerUsingNaturalLanguage Processing,” Int. J. Innov. Res. Technol., vol. 18, no. 7, pp.1–10, Jul. 2025. [14] N. Sharma et al., “Explainable AI Framework for Municipal ServiceRequest Prioritization Using LIME and SHAP,” J. Ambient Intell.Humaniz. Comput., vol. 16, no. 3, pp. 1235–1248, Mar. 2025. [15] P. R. Rao et al., “Deep Learning Architecture for Multilingual CivicComplaint Routing in Indian Smart Cities,” in Proc. IEEE Int. Conf.Smart Cities (ISC2), 2025, pp. 345–351. [16] R. Patel and S. Joshi, “Hindi-English Code-Mixed Complaint Classification Using Transformer Models,” in Proc. Int. Conf. Adv. Comput.Commun. Paradigms (ICACCP), 2025, pp. 456–462. [17] A. Rao et al., “Smart City Grievance Management Using BERT andGradient Boosting Ensembles,” Int. J. Inf. Manage. Data Insights, vol.5, no. 1, Art no. 100089, Feb. 2025. [18] P. Verma and A. Das, “Automated Routing of Municipal ComplaintsUsing Deep Learning and Geospatial Analysis,” Sustain. Cities Soc.,vol. 102, Art no. 105234, Mar. 2025. [19] G. Sharma and R. Patel, “Hybrid CNN-LSTM Model for UrgencyPrediction in Municipal Complaint Systems,” Eng. Appl. Artif. Intell.,vol. 128, Art no. 108512, Feb. 2025. [20] V. Reddy et al., “Low-Resource Language Processing for Indian CivicApplications Using Cross-Lingual Transfer,” ACM Trans. Asian Low-Resour. Lang. Inf. Process., vol. 24, no. 2, pp. 1–18, Feb. 2025. [21] S. Chatterjee et al., “Continual Learning Approach for Evolving MunicipalComplaintCategories,”Neurocomputing,vol.525,pp.123–135,Mar. 2025. [22] H. Chen and L. Wang, “Federated Learning for Privacy-PreservingComplaint Classification in Smart Cities,” IEEE Trans. Ind. Informat.,vol. 21, no. 4, pp. 4567–4576, Apr. 2025. [23] K. Singh and P. Gupta, “NLP-Based Categorization of MultilingualCitizen Complaints for Smart Governance,” in Proc. IEEE Int. Conf.Comput. Commun. Autom. (ICCCA), 2025, pp. 234–240. [24] K. Iyer and R. Mehta, “Real-Time Multilingual Chatbot Deploymentfor Urban Grievance Redressal,” in Proc. Int. Conf. Intell. Syst. Control(ICISC), 2025, pp. 321–327. [25] D. Kumar et al., “Graph Neural Networks for Complaint EscalationPrediction in Civic Systems,” in Proc. IEEE Int. Conf. Data Mining(ICDM), 2025, pp. 890–897. [26] R. Sen et al., “Zero-Shot Classification of Civic Complaints Across IndianLanguages,”inProc.Conf.EmpiricalMethodsNat.Lang.Process.(EMNLP), 2025, pp. 6789–6798. [27] A.BoseandP.Roy,“ActiveLearningStrategiesforEfficientAnnotationof Multilingual Civic Datasets,” in Proc. IEEE Int. Conf. Artif. Intell.Appl. (ICAIA), 2025, pp. 123–130. [28] M. Reddy et al., “Multilingual NLP Framework for Indian CivicComplaintCategorization,”inProc.IEEEInt.Conf.Comput.Intell.DataEng. (ICCIDE), 2025, pp. 112–119. [29] A. Rajkumar, S. Yuvasini, and others, “A Zero-Shot LLM Frameworkfor Multimodal Grievance Classification, Urgency Scoring, and AbuseDetectioninCivicFeedbackSystems,”Sci.Rep.,vol.15,Artno.32079,Dec. 2025. [30] R. Desai and P. Nair, “Nagarik Connect - Integrated Citizen GrievanceSystem Using NLP Classification,” Int. J. Front. Multidiscip. Res., vol.6, no. 41, pp. 107–115, Jun. 2025. [31] V. Patel et al., “AI-Powered Petition Analysis and Grievance Management System,” Int. J. Creat. Res. Thoughts, vol. 25, no. 4, pp. 794–802,2025. [32] A.Mishraetal.,“SMARTCITYCOMPLAINTANALYZERUsing NLP and Machine Learning,” Int. Res. J. Modernization Eng. Technol.Sci., vol. 7, no. 8, pp. 128–135, 2025. [33] T. Bangalore Municipal Corp., “Integrated Complaint ManagementSystem: Technical Architecture,” Bengaluru, India, Rep., 2025. [34] Maharashtra State Data Bank, “Pune Smart City Complaint ResolutionAnalytics,” Pune, India, Dataset Rep., 2025. [35] J. Lee et al., “Temporal Analysis of Urban Service Requests UsingLSTM Networks,” IEEE Trans. Smart Cities, vol. 2, no. 1, pp. 45–56,Jan. 2025. [36] S. Borade et al., “Driven Multilingual Chatbots for Low-ResourceLanguages with a Focus on Municipal Services,” Int. J. Comput. Appl.,vol. 187, no. 77, pp. 1–8, 2026. [37] S. K. Jain et al., “An Explainable Spatio-Temporal Decision SupportSystemforUrbanComplaintManagement,”J.SNATI,vol.12,no.1,pp.45–58,Jan.2026. [38] S. Gupta and N. Agarwal, “Vision-Language Models for MultimodalCivic Complaint Analysis,” in Proc. IEEE/CVF Winter Conf. Appl.Comput. Vis. (WACV), 2026, pp. 2345–2352. [39] Urban India Analytics, “Benchmarking Municipal Grievance SystemsAcross 50 Smart Cities,” Mumbai, India, Rep., Jan. 2026. [40] JanaagrahaCentreforCitizenshipandDemocracy,“IChangeMyCityDataset,”OpenCityDataPortal.Available: https://data.opencity.in/dataset/i-change-my-city-data/resource/a60abf5c3a15-4967-af32-c3074248580f

Copyright

Copyright © 2026 Gunasree KM, Pragadeeshwari A, Sabarisri K, Sahithya K, Dr. D. Nithya. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.