Intelligent Raising Knowledge Computing System Using NLP and Bi-LSTM: Design and Implementation

• Main Authors: Kuo-Hao Lu, 呂國豪
• Other Authors: Hsu-Yang Kung
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/f787z7

碩士 === 國立屏東科技大學 === 資訊管理系所 === 106 === The pig raising industry has been developing over one hundred years in Taiwan. Accumulated a large number of livestock husbandry knowledge and experience. For the advent of the era of smart agriculture. These valuable animal husbandry expertise are urgently needed to be converted into a computerized intelligent management system. However, there is currently no suitable Knowledge Graph in the field of pig breeding in Taiwan. Therefore, the following technical problems need to be solved, which include: (1) How to correctly determine the segmentation of the animal husbandry vocabulary so that the computer can understand the animal husbandry statement. (2) How to effectively extract the knowledge in the livestock field and classify it into the correct animal husbandry theme. (3) How to build a knowledge Graph of livestock husbandry and show the relationship between entities and entities. This study established “Intelligent Raising Knowledge Computing System”. The proposed system is mainly based on the deep learning schemes, which are Natural Language Processing (NLP) and Bidirectional Long Short Term Memory (Bi-LSTM). To correctly determine the segmentation of the animal husbandry vocabulary, this study proposed the Accuracy Livestock Word Segmentation Scheme (ALWS). ALWS uses the NLP to translate Internet information into a computer understandable language. ALWS also adopts Directed Acyclic Graph, Dynamic Programming, and Hidden Markov Model to effectively understand the animal husbandry vocabulary and build word vectors for deep learning. To build an effective knowledge Graph of livestock husbandry, this study proposed Intelligent Knowledge Unit Construction Scheme (IKUC). IKUC uses Bidirectional Long Short Term Memory to deeply learn livestock domain word vectors. The purpose is to understand contextual semantics and build knowledge units in conjunction with Conditional Random Field (CRF). Finally the knowledge units are classified into the correct categories. This study created four categories, which are Production Management, Feeding Management, Childbirth Management, and Breeding Management. We built knowledge Graph and collected expert knowledge of livestock husbandry to provide reference for pig farmers. The performance evaluation of the proposed methods are as follows. (1) For ALWS, we built a livestock dictionary and verify dictionary performance with accuracy, recall rate, and F-Measure. ALWS was compared with the Academia Sinica dictionary, the performance of the F-Measure is increased by 6.04. (2) For IKUC scheme, which was compared to the CNN and Bi-LSTM. Experiments show that the effectiveness of the proposed IKUC method is improved by 9.33% and has the smoother learning rate. Through the implementation of the paper's system and performance verification, this study built the first domestic animal husbandry pig knowledge map demonstration system. It is expected that this proposed system can collect expert knowledge of livestock husbandry through the developed knowledge extraction technology and provide the references for Taiwanese pig producers to improve the efficiency of pig feeding quality.