The process of recommendation starts with Embedding (title, abstract, category, etc.) the papers in advance. The Embedding method we use is combination of DNN and probability model. We offer high accuracy based on item-recommendation by learning from over 1 million papers.
Our AI distinguish between trend and survey from set of papers. By selecting more recent paper, you ca n fine tune AI that recommend trending paper surrounding your research field. If you select papers evenly in time, AI that recommend papers focusing on your detailed research fields.
Our AI supports to recommend each chronicle, because GPUs are accelerating calculation. So you can set multiple recommendation by creating more than one chronicle which segment detailed research fields.
By devising the GPU, and optimizing calculation method, we are able to provide recommendation as an API that can be updated in real time. No loss of computational accuracy and no compression of the model.