Search

Microorganisms play a vital role in various ecosystems, but their complex interaction is still unclear. With the publication of a large number of microbial literatures, many experimentally verified microbial interaction is dispersed therein. Organizing them into a database or knowledge graph can facilitate the development of microbiology research. Text mining technology is able to automatically extract and integrate these microbial interactions, as well as discover implicit information in literatures. For this purpose, we manually annotate a Microbial Interaction Corpus (MICorpus) containing 1005 abstracts, which provide a useful data source for the MIE task. On this basis, we propose an automated MIE extraction system based on Max-Bi-LSTM model. The best result of the system is precision (P) 76.313%, recall (R) of 90.121%, and an F value (F) 82.476%.

Interactions among microorganisms have been the key to understand microbial communities. As an important member of microorganisms, bacteria are closely related to human diseases. Therefore, studying the interaction between bacteria plays an important role in microbiome research. There are a large number of published medical literatures that contain small-scale data about the interactions between bacteria. These literatures often record the bacteria interactions discovered by co-cultural experiments for two or more species. Mining and organizing them into databases will provide reliable support for microbiome research. Named entity recognition (NER) is an essential step of interaction extraction (IE) by automatically identifying bacterial entities in the text. In this paper, we propose a method based on language model for identifying bacteria named entities. Using the language model to learn the semantic information between words, the F1 score reaches 96.14%, which is the best performance in bacteria NER compared with the previous experimental results.

Microorganisms have been confirmed to be essential for the fundamental function of various ecosystems. The interactions among microorganisms affect the human health and environmental ecosystem. A large number of microbial interactions with experimental confidence have been reported in biomedical literature. Extracting and collating these interactions with experimental confidence into a database will create a valuable data resource. Named Entity Recognition (NER) is the premise and key to interaction extraction from literatures. Especially, bacterial named entity recognition is still a challenging task due to the specialty of bacterial names. In this paper, we propose a bacterial named entity recognition system based on a hybrid deep learning framework (HDL-CRF), which integrates two deep learning models: the bidirectional long short-term memory network and the convolutional neural network, as well as the conditional random field approach, for automatically extracting the features. Finally, we prove that this model outperforms previous methods in performance.

Object tracking is a key technology in video surveillance. Reliable tracker must be adaptive to the constantly changing object sizes. Most of the state-of-the-art methods estimate the object scales using their appearances. Those methods are vulnerable to occlusion, object deformation, illumination change and background clutter. In this paper, we propose to use the geometric context of the surveillance site as a strong clue for scale adaptation. With three reasonable assumptions on the video cameras and the surveillance sites, we deduce a simple geometric model for object scales. The parameters of this model are learned without any human intervention. Then we integrate this model into baseline trackers for robust scale adaptive object tracking. Experimental results on challenging surveillance videos indicate that our approach favorably improves the performance of single-scale baselines, and performs better or comparative to the state-of-the-art multi-scale trackers while significantly improve the speed.