Now showing 1 - 2 of 2
  • Publication
    Ontology-based Knowledge Map Model for Handling Mined Knowledge in Digital Agriculture
    (University College Dublin. School of Computer Science, 2022) ;
    Farm knowledge plays an important role in digital agriculture. In which the knowledge discovered from the process of data analysis is the most diverse, flexible and dynamic for digital farming. However, the heterogeneous, diverse and dynamic knowledge also makes it difficult to use, exploit and manage for different users. This PhD research presents a novel Ontology-bAsed Knowledge map (OAK) model for representing, storing, managing and retrieving knowledge, which is discovered from popular data mining tasks in agriculture or knowledge collected from experts (farmers, agronomists, researchers, etc). OAK model includes a set of entity definitions to provide the robust and inclusive principle to model any type of knowledge. This research also provided a new definition of knowledge representation to include in my model the results of a data mining technique, including clustering, classification, and association rules mining. This makes not only easy to infer the knowledge but also to extract it from external documents, such as journal papers. Moreover, this study also proposes a new Knowledge Map framework based on the OAK model with 6 components to handle mined knowledge. To realize the OAK model, the study firstly built an Agriculture Computing Ontology (AgriComO), which contains nearly 600 classes and 14,000 instances in agriculture, computing domain and related subdomains. This ontology also includes 1,310 transformations, which are used to process and discover knowledge in data mining. Then, a prototype for a knowledge repository was built to hold up to 500 knowledge representations, which were extracted from 1,000 data mining results. A scientific agriculture entity recognition module has been built on the semantic-based deep learning approach to assist in extracting entities and re-building the knowledge representations. In addition, this study proposed a novel standard for knowledge assessment within the OAK model to verify knowledge representations before transforming them into the knowledge repository. This PhD study built an innovative Knowledge Browser as a demonstration of knowledge exploitation to support to identify knowledge by input concepts and roles. There is no approach to evaluate the whole model in knowledge management. In this research, the proposed model has been evaluated in individual components. Firstly, this project applies four ontology evaluation methodologies to validate and verify the proposed ontology model. To demonstrate the efficiency of the entity recognition module, this project built an annotated entity corpus from 20,000 agriculture news articles and over 3,500 abstracts of scientific papers in digital agriculture for training and evaluating the semantic-based deep learning approach for extracting agricultural entities and scientific entities. Finally, the project also build completes several experiments on the system to demonstrate the ability of the OAK model in knowledge management and handling mined knowledge in digital agriculture. The demonstration of the knowledge repository and knowledge browser can support data scientists and agronomists in finding mined knowledge from input concepts and their roles as well as finding similar solutions for data processing of agricultural attributes in digital agriculture. Moreover, the successful adoption of the proposed model paves the way to build an effective knowledge management system for storing and retrieving previously created knowledge, especially knowledge from data mining.
  • Publication
    Predicting Soil pH by Using Nearest Fields
    In precision agriculture (PA), soil sampling and testing op-eration is prior to planting any new crop. It is an expensive operationsince there are many soil characteristics to take into account. This papergives an overview of soil characteristics and their relationships with cropyield and soil profiling. We propose an approach for predicting soil pHbased on nearest neighbour fields. It implements spatial radius queriesand various regression techniques in data mining. We use soil dataset containing about 4,000 fields profiles to evaluate them and analyse theirrobustness. A comparative study indicates that LR, SVR, andGBRTtechniques achieved high accuracy, with the R2 values of about 0.718 and MAEvalues of 0.29. The experimental results showed that the pro-posed approach is very promising and can contribute significantly to PA.
      71Scopus© Citations 3