A Deep Self-learning Classification Framework for Incomplete Medical Patents with Multi-label
| Autor: | Qianqian Ji, Mingsheng Shang, Xianbo He, Xiaoyu Shi, Luo Mengzhen, Weiguo Tao |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Thesaurus (information retrieval)
Vocabulary Information retrieval Computer science media_common.quotation_subject ComputingMilieux_LEGALASPECTSOFCOMPUTING 02 engineering and technology Construct (python library) Task (project management) ComputingMethodologies_PATTERNRECOGNITION 020204 information systems 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Precision and recall Function (engineering) Classifier (UML) media_common |
| Zdroj: | Advances in Natural Computation, Fuzzy Systems and Knowledge Discovery ISBN: 9783030325909 ICNC-FSKD |
| DOI: | 10.1007/978-3-030-32591-6_61 |
| Popis: | The classification of medical patents play an important role for pharmaceutical company, since medical patens with well labeled can significantly accelerate the process of new drug research. The previous studies using machine learning methods focus on classification the medical patents with single label. However, the classification of medical patents is a multi-label task, and the available data are always incomplete with losing the part information of patents. In this paper, we propose a deep self-learning classification framework that can deal with the incomplete medical patterns with multi-label issue. It consists of a text processor and patent classifier. For the text processor, a professional medical text thesaurus is built via GloVe method, which can learn more specialized vocabulary. For the patent classifier, we adopt a bidirectional long short term memory (Bi-LSTM) model to construct our patent classifier, which can learn the hidden knowledge from medical patents and associate with appropriate labels to medical patents automatically. Furthermore, an advanced focal loss function is design to further improve the classification accuracy. Experiments on the Thomson Reuters dataset demonstrate that our proposed method outperform the other existing methods in terms of precision and recall, when dealing with incomplete medical patents with multi-label issue. ... |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|