We are creating some awesome events for you. Kindly bear with us.

GovTech and MSO Develop OneService Chatbot to Improve Municipal Services

Image credits: www.mnd.gov.sg/mso

In line with Singapore’s Municipal Services Office’s (MSO) vision to continuously improve their services and push the technical boundaries for gathering municipal issues through other means, MSO has worked with GovTech to develop the OneService chatbot.

The tech enables citizens to easily lodge a case and provide additional, crucial information about a complaint through commonly used social messaging apps, such as WhatsApp and Telegram. Powered by machine learning, the chatbot would be able to:

  • Automatically identify the nature of the complaint and classify it into the appropriate category (killer litter, illegal parking, etc),
  • Extract the relevant details of the incident that needs attention (location, address, landmark, when it occurred, etc), fill in the feedback template, and get the user to verify and provide additional details.
  • Identify the correct government agency that should take action (NEA, NParks, LTA, etc) and pass the case on.

The Moments of Life Division’s Virtual Intelligent Chat Assistant (VICA) Team and the Data Science and Artificial Intelligence Division’s GovText team worked together to develop a chatbot capable of having a conversation. With the OneService app in operation since 2015, MSO has collected a substantial amount of feedback from the public. Every time a case is investigated, an officer will tag a case with its corresponding case type, and this information is stored in the database. Since the feedback submitted through the chatbot will be like that submitted through the app, the agency can use the app’s data to train the chatbot’s case type categoriser.

Essentially, this means feeding both the feedback text and case type of each case to the categoriser so that it learns to associate certain words and patterns in the text with its corresponding case type. Drawing from its experience, the case type categoriser will look for relevant words and patterns to help it make the best guess of what the correct case type should be, given just the feedback text.

Armed with over 160,000 cases from two years’ worth of OneService app data, the team tried out different techniques in Natural Language Processing (the field of getting computers to understand human language) and managed to build a categoriser that can predict the correct case type correctly 80% of the time. Next, the agency had to extract the key case details and pre-fill the case form for the user. This is trickier because, unlike the case type, the platform had no existing labelled keywords as the MSO staff did not need to label the keywords in their work process.

Hence, the team set up an annotation framework and got their MSO colleagues to help us annotate the words within the feedback text with labels representing the types of important information required to resolve a case such as the dates and times the incidents occurred, landmarks, and addresses.

At this stage, the tech’s implementors were able to automatically identify the nature of the complaint, extract the relevant details, fill up the feedback template, and prompt the user to add the missing information. Now that the case has been successfully filed by the user, it’s time to find the correct agency to deal with it.

Municipal services are overseen by multiple agencies, so it may not always be straightforward nor simple for the OneService Chatbot to activate the right process. For this, in addition to the feedback text and the case type (automatically tagged and then verified by the user), the team also used the images and the geolocation submitted by the user.

Geolocations and images have an important role in identifying the right agency because some types of cases can be handled by more than one agency based on just the case description alone. Hence, the agency assigned would depend on which agency’s land an incident took place in or is nearest to. For example, if a Tree Pruning case is reported within a housing estate, the nearest town council will be assigned to handle the case. However, if a similar case happens in a park (e.g., West Coast Park), NParks will be assigned to handle the case instead.

Combining these new data points, GovTech and MOS were able to correctly direct cases to the right agency 85% of the time. After making some tweaks based on feedback received from a completed trial with a small segment of the public, the OneService Chatbot has been “beta” launched and is available on WhatsApp and Telegram from July 2021.

Send this to a friend