Crowdsourcing label speculation method and system based on graph neural network

Abstract

The invention discloses a crowdsourcing label speculation method and system based on a graph neural network, and the method comprises the following steps: (1) carrying out the data processing of a crowdsourcing label, and obtaining the initial features of a labeling person and a task; (2) constructing a tagging personnel-task heterogeneous graph, a tagging personnel isomorphic graph and a task isomorphic graph for the task allocation situation of the tagging personnel; (3) inputting the annotation personnel-task heterogeneous graph, the annotation personnel isomorphic graph and the task isomorphic graph into a graph neural network to obtain embedded features of task nodes; and (4) inputting the obtained embedded features of the task nodes into a prediction layer to obtain the probability that the task belongs to each label, and regarding the label with the maximum probability as the correct label of the task. High-accuracy crowdsourcing label speculation is achieved through the graph neural network, a large amount of available machine learning training data can be generated, people are helped to train an algorithm model, and the competitiveness of the AI field is improved.

Description

technical field [0001] The present invention relates to the field of label inference in crowdsourcing mode, in particular to a method and system for crowdsourcing label inference based on a graph neural network. Background technique [0002] Machine learning, especially supervised learning, has been widely used in computer vision, natural language processing and other fields. Since supervised learning requires a large number of samples with known correct labels to train the model, the traditional method is to review the samples by domain experts and label them correctly. This method is usually expensive and time-consuming, and cannot meet the growing demand for labeled data. [0003] Nowadays, crowdsourcing (Crowdsourcing) has become one of the most important tools for obtaining data labels due to its low cost and high efficiency. With the help of online platforms such as Amazon Mechanical Turk (AMT) and CrowdFlower, people can easily obtain and utilize crowdsourcing resour...

AI Technical Summary

Problems solved by technology

However, these models often require people to carefully design complex generation processes and inference algorithms for them, and are not suitable for large-scale data sets

There are also some deep learning models that try to learn both classifier models and label aggregation models, but these models require additional task features and rarely consider the two-way interaction between annotators and tasks

Images