A deep learning methodology for automatic extraction and discovery of technical intelligence

Abstract It is imperative and arduous to acquire product and business intelligence of global technical market. In this paper, a deep learning methodology is proposed to automatically extract and discover vital technical information from large-scale news dataset. More specifically, six kinds of technical elements are first defined to provide the concrete syntax information. Next, the CRF-BiLSTM approach is used to automatically extract technical entities, in which a conditional random field (CRF) layer is added on top of bidirectional long short-term memory (BiLSTM) layer. Then, three indicators including timeliness, influence and innovativeness are designed to evaluate the value of intelligence comprehensively. Finally, as a case study, technical news on three military-related websites is utilized to illustrate the efficiency and effectiveness of the foregoing methodology with the result of 80.82 (F-score) in comparison to four other models. In more detail, data on unmanned systems are extracted to summarize the state-of-the-art, and track up-to-the-minute innovations and developments in this field.