Crowdsourcing active learning method and device based on labeler reliability time sequence modeling

Abstract

The invention discloses a crowdsourcing active learning method and device based on annotator reliability time sequence modeling. The method comprises the following steps of initializing a prediction model and establishing a reliability time sequence model, a sample selection strategy, a crowdsourcing annotator selection strategy, a sample distribution strategy, model updating and iterative operation. According to the method, crowdsourcing learning, active learning and time sequence modeling are combined, and on the basis of crowdsourcing labeler reliability time sequence modeling, real-time tracking is carried out on the change of the labeler reliability, and the crowdsourcing labeler with the highest reliability is screened out; the better crowdsourcing annotators are continuously updated through iterative operation, and the prediction model is continuously optimized by samples with high information degree, so that the crowdsourcing annotation cost is reduced, the negative influence of the annotators with low reliability in learning is reduced, and the quality of crowdsourcing labels and the performance of the prediction model are effectively improved.

Description

Technical field [0001] The present invention involves machine learning and people in the field of circuit computing technology, especially a crowdsourcing active learning method and device based on the sequential models based on the reliability of the bidders. Background technique [0002] The traditional way to rely on field experts to marked data and model training in machine learning cannot meet the needs of rapid iteration and update of business models. The emergence of crowdsourcing models provides a feasible way for this issue.Crowdfunding data is fast, low cost, and has the potential to meet the needs of practical application.Although the crowdsourcing mode provides a fast and low -cost effective way for machine learning data, its inherent openness, dynamics, uncertainty, and cost restrictions are still full of challenges to use crowdsses injecting data for machine learning.The open features of crowdsourcing marked determine the diversity of the data requirements for custo...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a crowdsourcing active learning method and device based on time-series modeling of annotator reliability, so as to solve the problem that existing active learning methods cannot track each Crowdsourcing the reliability of the annotator at the time of annotation, so as to learn a higher performance prediction model

Images