Augmented Co-Creation: Using Artificial Intelligence and Neural Network Algorithms to Support Design Collaboration.
Authors: K. Brandt, B. Lonsway, S. Masiclat
Corresponding author: S. Masiclat
This research project developed a web portal to serve as a nexus between the broader technology community and social organizations (nonprofits, NGOs, social enterprises, etc.) in need of technology. The team was asked to go beyond developing a searchable code repository and understand the organizational, structural and technological challenges of connecting social sector organizations to a globally distributed community of technologists who spoke different professional languages.
In this project it was important to design a system that did not attempt singular perfect matches between problem statements and software descriptions. To accomplish this, we used Self-Organizing Map (SOM) algorithms. SOMs allow for a complete and robust representation of highly complex data, but as they are intentionally constructed to avoid computing a single “correct” result, they are considered less accurate than other neural networks based on gradient descent back propagation. We employed a Latent Semantic Indexing technique as a placeholder for a variety of NLP pre-formatting techniques, representing unstructured text inputs in a bi-gram vector form. We then used these bi-gram vectors as inputs to the SOM to perform unsupervised training on the computational neural network such that it formed a representation of the topology (geometric inter-relations) of the input data.
This research was sponsored by a grant from JP Morgan Chase.