Usecase Qualified

Qualified's challenge was to improve the matching algorithm for the search engine, allowing projects to be directly and automatically linked to organizations or experts. Traditional search engines look for similar terms, tags and words in both the project and organization or expert. When the word "data" appears in a project, traditional algorithms match it to an organization or expert with data in the title. The context is not understood in this process. The percentage of correct matches had to go up for a high-tech matching platform to perform optimally. In addition, there was little data available, which can be a big problem since AI models need a lot of data.

Using a neural network based on the BERT architecture, we developed an AI-powered matchmaking system that can understand project descriptions based on Natural Language Processing (NLP) and can select the best matches. This was done by using pre-trained BERT models and fine-tuning them for this specific task. This allows state-of-art NLP technology to be used, with less data. In this way AI is used to match the right expertise with technological challenges within society.

The precision of our model improved the original version by more than 30%. A qualitative analysis with Qualified.technology showed that after training, the algorithm kept increasing it's 𝐮𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝𝐢𝐧𝐠 of which companies were a match for a project and which were not. In addition, a mechanism was built which ensures that the model will learn over time: the more the platform is used, the better it gets. By implementing the matchmaking system, the recommendations become better and more accurate. In addition, Qualified.technology's platform is not dependent on ElasticSearch and Qualified has now developed an interesting piece of IP that sets it apart from competing platforms.