Service: Support system for classification of citizen inquiries or complaints

Responsible organisation: Barcelona City Council (Local Government)

Support system for the classification of incidences/complains from citizens in Barcelona. IRIS is the service that allows citizens to report incidents or send complaints to the Barcelona City Council for resolution. Through IRIS, citizens can send information and queries, as well as complaints and suggestions for improvement. In this process, the citizen who reports the incident has to classify it using a tree of topics offered through the computer application. This classification is important because it is used to direct the incident directly to the responsible department, thus speeding up the response process. Errors in the thematic classification cause inadequate responses and delays in the resolution of incidents, thus affecting the quality of the service offered. As part of the IRIS service update project, a module called MARIO has been developed, based on machine learning algorithms -Machine Learning- and natural language processing (one of the technologies included in the IA) to simplify the incident classification process for citizens. From the analysis of the free text describing the incident, MARIO suggests to the citizen the most likely categories where the incident fits because he/she chooses the most appropriate category. MARIO, which is currently being tested, minimizes the error rate in the initial classification of incidents by greatly reducing manual reclassification processes. Previously, 50% of the communications made needed to be reassigned; now, with MARIO, a hit rate of over 85% is being achieved. MARIO has been developed using open source tools -Open Source- such as Python, Scikit-*learn and *Pandas, among others. The algorithm has required a training that has been carried out using a dataset with the topics and questions asked by the citizen previously validated to ensure that they have been correctly classified. This pilot has been used by the Municipal Institute of Informatics to evaluate the techniques and processes of text classification with the following conclusions: Often the integration of AI techniques into existing processes has to be accompanied by a restructuring of the data. In the case of IRIS, the wide range of classification options makes the task difficult for a human but makes *IA inefficient. Clustering of categories is recommended - It is important to have a quality dataset. - In order to ensure good system efficiency, good data preprocessing including cleaning and standardization of this data is necessary. - Data anonymization must be taken into account. - For good training, it is necessary to have a representative volume of input data that is homogeneous in all categories.

Additional information

Source Open Innovation Regione Lombardia
Web site https://atencioenlinia.ajuntament.barcelona.cat
Start/end date 2021 -
Still active?

Related cases