Year: 2024
Type: article
Abstract: The traditional methods of classifying texts consider values from agricultural texts as characters that could lead to a reduction in the semantic meaning while being lost.Most of such numerical features, often necessary in agriculturally related applications, carry specific information that the traditional techniques are unable to fully capture.In this regard, it is proposed that a novel text classification method based on the concept of multivariate feature dynamic fusion be undertaken.The method utilizes multiple windows of Convolutional Neural Networks (CNNs) to extract local semantic information from agricultural texts at different levels.CNNs are very effective in capturing many aspects of the text, from fine-grained details to broader context.However, the crucial numerical features that contain significant semantic value are also extracted manually through an artificial method to create a numerical feature vector.This vector would therefore encapsulate the numerical details of the agricultural text, potentially essential for correct classification.Attention mechanism: This would be proposed to dynamically fuse multiple semantic features derived from both textual and numerical data.The attention mechanism enables the model to focus more on relevant features, thereby having richer and deeper semantic expression for the agricultural text.By integrating the model with textual and numerical information, it can classify agricultural texts better than using only one kind of information.This enhances greatly the ability to mine deep latent semantic features from text and to represent these features so that the accuracy and performance in a text classification task will be even better.This approach incorporates the dynamic fusion of features in attention mechanisms that allow maximum utilization of textual and numerical information.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: gold
OpenAlex ID: W4406039110