Aeterna Labs’ own Ph.D. students Arezoo, Anton, and head of R&D Mona have developed an automatic news story discovery model called PromptStream. Prompting asks a language model for information. The novelty of our researchers’ work is to adapt this technique so that it has the capacity to be applied to a continuous stream of data. Hence the name, PromptStream.
PromptStream can be applied to any data stream (such as Aeterna's steadily incoming flow of news articles). It works by automatically clustering similar articles together, resulting in a cloud of clusters or groups of news stories, current discussions and events.
What’s innovative about this is that the model works "unsupervised," meaning we don't need any expensive labelled data. This also means that the model can adapt to new trending topics which are very common within news cycles, and of interest to our clients. For example, say a relatively unknown politician is taking heat, we won't need to manually update the model with example articles or new keywords specific to the story to find and categorise articles that pertain to that discussion.
This week, Anton and Mona are presenting their findings at the prestigious LREC-COLING, The 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation in Turin.
We’re working hard to create state of the art systems that get your campaigns in the best possible context, when buyers are ready to buy. Maybe a story about a local politician getting sacked (or coming close to it) might not be a motivator for people to buy much of anything, but what it will be getting is a lot of traffic. We also know every product has its own ideal contextual placement. Reach out to us today or take a look through our catalogue to get more ideas about how to tailor your campaign. We’re looking forward to hearing from you!