Machine learning for child concern

Important results from the project

Strängnäs municipality has developed the language model ADA: an early version of operational support for the process of reporting concerns for children. The goal is that social secretaries in a new case receive eventual signals of violence and conflict, which ADA detects both in the currrent and previous reports. The model´s capacity is demonstrated, further training is necessary, and information security aspects need to be solved. The project highlights the importance data preparedness, the benefits of hybrid solutions of which AI is a part, and a need for broader collaboration.

Expected long term effects

- An early version of NLP model ADA (XLM-R) is developed, which to 70% identifies signs of two categories of violence and conflict - Valuable lessons regarding NLP use in public administration: data preparedness, annotation, infrastructural and information security challenges mm - Strängnäs municipality continues the project with own funding until December 2021, contributes with lessons learned in networks coordinated by SKR and AI Sweden, within the collaboration "4 Mälarstäder" and as a participant in AI Sweden´s project Digital Verkstad 2021 (if funding is granted)

Approach and implementation

- The initially obvious AI technology (deep learning) was not the more appropriate: the project changed tracks to Natural language processing - A broad cooperation and support in the public sector (coordinated by AI Sweden and SKR) enabled the project - A rigorous focus on the business needs is required already in the phase data preparation -Some issues of national concern are not resolved (information security etc.) -A national coordination in data preparedness work is needed, Strängnäs hopes to contribute to it as a participant in AI Sweden´s project "Digital workshop 2021"