We have built and implemented a classification based on the IAB Audience 1.1 taxonomy. Contrary to its original purpose, we apply it contextually. This means that we use the content of the article to identify the reader’s potential and likely Purchase Intents, Interests, and Demographics. In this way, we are not tied to user tracking or cookies in any way.
Journey Towards a Cost-Effective Classification Model
Developing this classification model that finds the potential characteristics of the reader has not been simple. In many cases, there is no single correct answer. However, it is possible to find better and worse answers. To achieve maximum cost-efficiency and quality in production and also high-volume usage, we have combined manual work, capabilities of generative AI, and our own ontology in an iterative process. When using only generative AI, the achieved results were not sufficient for this creativity-demanding task relative to the costs, and the outcomes were narrow and somewhat bland. Therefore, we have also utilized manual work that we have already done for many years, our ontology, and our other service based on it for identifying similar articles, which has helped us create suitable noise for the training data and generate a vast amount of it. The basic idea of our original ontology is to identify the topics that the content covers. By combining this information carefully with manually and generatively produced training data, we have built a cost-effective AI model that identifies the reader’s potential characteristics in the best possible way, not just what is directly stated in the text. One crucial factor in this ongoing project is also the significant computational power of the LUMI supercomputer, which made it possible to handle large amounts of data and turn the information of the data into a machine learning model.
Manageability and Cost-Effectiveness
Significant benefits for the client are manageability and cost-effectiveness. Targeting advertisements and understanding the structure of the produced content gain a new dimension with contextual identification of the reader’s characteristics. There is no longer a need to consider what the potential buyer of a product is interested in contextually, as the essential information is directly included in the AI model in Interests, Purchase intents and Demographics. We have ensured that our classification model can be customized based on preferences and offers our clients the best possible benefit cost-effectively. The development and adaptation of the model to a changing environment has been an essential part of the development work.
If you want, you can try it out here to see how it works: https://neuwo.ai/api-demo/
