Using hybrid algorithmic-crowdsourcing methods for academic knowledge acquisition

Scientific literature contains a lot of meaningful objects such as Figures, Tables, Definitions, Algorithms, etc., which are called Knowledge Cells hereafter. An advanced academic search engine which could take advantage of Knowledge Cells and their various relationships to obtain more accurate search results is expected. Further, it’s expected to provide a fine-grained search regarding to Knowledge Cells for deep-level information discovery and exploration. Therefore, it is important to identify and extract the Knowledge Cells and their various relationships which are often intrinsic and implicit in articles. With the exponential growth of scientific publications, discovery and acquisition of such useful academic knowledge impose some practical challenges For example, existing algorithmic methods can hardly extend to handle diverse layouts of journals, nor to scale up to process massive documents. As crowdsourcing has become a powerful paradigm for large scale problem-solving especially for tasks that are difficult for computers but easy for human, we consider the problem of academic knowledge discovery and acquisition as a crowd-sourced database problem and show a hybrid framework to integrate the accuracy of crowdsourcing workers and the speed of automatic algorithms. In this paper, we introduce our current system implementation, a platform for academic knowledge discovery and acquisition (PANDA), as well as some interesting observations and promising future directions.