Search

Disease diagnosis and treatment have been an important support for aquatic animal health in aquaculture. A knowledge graph can be an effective way to express and apply the knowledge on the aquatic animal disease diagnosis and treatment. Among them, the named entity recognition has been the key component to construct the knowledge graph of aquatic animal diseases, particularly on the polysemy and entity nesting. However, the low recognition accuracy of named entities has posed a great challenge to the diagnosis and treatment of aquatic animal diseases. In this study, a diagnosis and treatment of aquatic animal diseases named entity recognition was proposed using BERT+CaBiLSTM+CRF (Bidirectional Encoder Representations from Transformers+Cascade-Bi-directional Long Short-Term Memory+Conditional Random Field). Firstly, the feature of the BERT model contained the position vector information. The polysemy was effectively improved to distinguish the different meanings that were expressed by entities in different contexts. Secondly, the CaBiLSTM model was designed for the nested named entity recognition using “hierarchical thinking”. The reason was that the inner entity in the nested entity of aquatic medicine greatly contributed to the recognition of the outer entity. First of all, the BiLSTM+CRF model was used to identify the inner entities that appeared frequently, and then the dimension reduction of the identified inner entity feature matrix was connected outer entity feature matrix to retain the complete inner entity feature information. After that, the BiLSTM+CRF model was used for the outer entity recognition to improve the discrimination of outer entities for the accurate recognition of outer entities. Finally, a comparative experiment was designed to verify the effectiveness of the proposed recognition. The test results show that the accuracy, recall, and F1 value of the named entity recognition task in the aquatic medicine using the BERT+CaBiLSTM+CRF model reached 93.07%, 92.85%, and 92.96%, respectively. The entity structure features were outstanding in terms of specific entity categories, due to the five types of non-nested entities, such as aquatic animal names, drug names, disease names, disease sites, and pathogens. For example, most aquatic animal names contained the radicals, such as “worm” and “fish”. The radicalsand drug names were mostly composed of chemical elements, while the disease names were mostly ended with the word “disease”, indicating a higher recognition accuracy than that in the nested entities. But in view of the outstanding nested structure of entities, the model performed better to identify the nested named entities, such as the clinical symptoms using the named entity recognition model integrating the BERT and CaBiLSTM designed by the “hierarchical idea”. Higher recognition was achieved than before. The recognition accuracy, recall, and F1 value increased by 12.31, 12.76, and 12.53 percentage points, respectively. Therefore, the model can be expected to effectively improve the accuracy of entity recognition caused by ambiguity and entity nesting in the task of diagnosis and treatment of aquatic animal diseases named entity recognition. The finding can provide the potential support to construct the fisheries field knowledge graph, further promote the healthy aquaculture projects. [ABSTRACT FROM AUTHOR]

Entity relation extraction is a fundamental task to detect a list of triplets, including two entities and the semantic relations between them. An overlapping relationship has caused the low performance of standard entity relation extraction in fishery in recent years. In this study, a novel entity relation extraction was proposed in fishery using the dual attention mechanism. First, a sentence classification and labeling strategy were selected to solve the difficulty in labeling overlapping relations in fishery standard texts. Second, an extraction model of standard entity relationship was established using a combination of dual attention, and Bidirectional Encoder Representations from Transformers-Bi-directional Long Short-Term Memory-Conditional Random Field (BERT-BiLSTM-CRF). Five components were included from the bottom to top: BERT layer, BiLSTM layer, word attention mechanism, sentence attention, and CRF output layer. In the pre-trained BERT model, the dual-layer two-way conversion and decoding were used to automatically learn sentence features for the vector representation. In the BiLSTM model, the context feature of the target entity was learned from the BERT output. The word and sentence level attention was utilized to optimize the weight of target words and sentences in the paragraph, including the noise removal, and the accuracy of relationship extraction. The CRF decoder was used to represent the output of the attention layer in the form of sequence tags. Finally, a comparative experiment was designed to verify the effectiveness of the model. The results show that better overall performance was achieved in the entity relationship extraction using the dual attention on the fishery standard interaction (DLOU-FSI) dataset, where the accuracy, recall, and F1 value were 92.67%, 92.31%, and 92.49%, respectively. The overlapping relations were effectively solved in the extraction of fishery standard relations. In addition, the recognition accuracy rates in the seven relation categories were all higher than others. The accuracy rate, recall rate, and F1 value were greatly improved by more than 90% in the quotation relationship, regulation, release, proposal, drafting, and centralization. Nevertheless, the recall rate of the improved model was dropped slightly in the comparison test. The reason was that there was a sparse distribution of sentence samples in the related category of comparative relations. In the DLOU-FSI corpus, each standard text of fishery contained only 0-3 comparison relation triples, less than 1% of the total number of all relational triples. As such, the comprehensive relationship features were necessary to learn for the higher recognition in the relationship extraction task. In addition, the quality of corpus was determined on the extraction of fishery standard text relations. The higher quality and the more accurate recognition of the model were achieved, as the deep learning model was trained to learn more data. There was also a critical value, after which the model cannot be greatly improved. In the case of a model adaptation, the test data can be effectively expanded to increase the number and diversity of samples for a better overall effect of relationship extraction. This finding can lay a significant foundation for the construction of fishery standard knowledge graphs. [ABSTRACT FROM AUTHOR]

Fishery information service is a vital component to realize data analysis, feature extraction, and fishing forecasting, particularly for a high comprehensive production capacity and modernized management in fishery. The commonly-used keyword matching without standard contents cannot meet the high demand for accurate service in the current information system of fishery. The standard quantitative indicators in fishery have become one of the most important tasks in the information service. Therefore, it is very necessary to accurately identify the effective standard quantitative indicators for the automatic extraction of fishery. Combining the attention mechanism and the BERT+BiLSTM+CRF (Bidirectional Encoder Representations from Transformers + Bi-directional Long Short-Term Memory + Conditional Random Field) model, this study aims to propose a highly accurate recognition method of standard quantitative indicators in fishery, further to replace the commonly-used entity recognition. The quantitative indicators were firstly divided into four types of entities: the indicator name, indicator value, unit, and qualified words for identification. This operation effectively dealt with the difficult identification of fishery standard quantitative indicator entities. It was found that the location information behaved a significant impact on the recognition of indicator names and other entities. Vector data was also utilized to improve the recognition of indicator names. Secondly, the BiLSTM model was used to learn the semantic features of long sequences in the fishery standard text quantitative indicators. The attention mechanism was then integrated to treat the long-sequence semantic dilution. Finally, all sequence tags were obtained through the CRF layer. The test results showed that the accuracy rate was 94.51%, the recall rate was 96.37%, and the F1 value was 95.43% for the fusion attention mechanism and the BERT+BiLSTM+CRF model. Compared with the fusion attention + BiLSTM + CRF (named entity recognition model), the accuracy, recall rate, and F1 value increased by 2.78, 6.73, and 4.65 percentage points, respectively. The word vectors, position vectors, and sentence features were combined for better recognition in the model. The self-attention mechanism of the BERT model was pre-trained, where a bidirectional encoder was used for the transformer layer in the BERT model, indicating a better performance on the text context memory. Compared with the BERT+BiLSTM+CRF model, the accuracy, recall, and F1 value increased by 1.62, 0.25, and 0.97 percentage points, respectively, indicating that the attention mechanism contributed to the greater weight of the target entity in the long- and short-term memory network. The features were then weighted to make the model more accurately identify quantitative indicators. The proposed model can be expected to more accurately identify the fishery standard quantitative indicators, especially the indicator names, indicator values, units, qualifiers. This investigation can provide promising data support to accurate information using standard content services. The effective fishery standard quantitative index can also offer new ideas for the identification of quantitative indicator named entities in agricultural, medical, and biological fields [ABSTRACT FROM AUTHOR]