Integrated weighted majority soft voting crowdsourcing data truth value reasoning method
A reasoning method and soft voting technology, applied to instruments, character and pattern recognition, computer components, etc., can solve problems such as not considering the quality of instance feature labeling, and achieve strong implementability
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0055] see figure 1 and figure 2 , the present invention provides a flow chart of an integrated weighted majority soft voting crowdsourcing data truth reasoning method. The specific process is described in detail below.
[0056] Step 1. Convert to a new crowdsourced dataset by calculating the probability that an instance belongs to each category, then copying K-1 copies of the instance and associating the instance with a different category label k=1,2,3,...,K is used to train weak classifiers. The method removes the influence of speculative aggregated labels, improves classification accuracy, and positively affects the performance of ground-truth inference;
[0057] Step 2, using a method based on maximum likelihood estimation to aggregate weak classifiers;
[0058] Step 2.1 Obtain the confusion matrix set Π of all weak classifiers according to the statistics of step 1;
[0059] Step 2.2 obtains the new classifier prediction label according to the maximum likelihood est...
Embodiment 2
[0064] The difference from Example 1 is that we also need to consider each worker's ability to label different instances, and obtain predicted labels through the soft voting method based on worker weights, see image 3 ,Specific steps are as follows:
[0065] Step 3. Introduce different labeling capabilities of workers on different instances, and use a method based on similarity comparison to calculate worker weights;
[0066] Step 3.1 Calculate the overall quality of the worker by comparing the similarity between the worker label and the strong classification prediction label, and the related formula is as follows:
[0067]
[0068] where f(x i ) is the classifier according to the feature vector x i The predicted class label, τ j Indicates the overall quality of the jth worker, and I indicates the total number of instances;
[0069] Step 3.2 obtains the specific labeling quality of the worker by comparing the labels of the workers. If two workers have the same labeling...
Embodiment 3
[0077] The algorithm model of the integrated weighted majority soft voting crowdsourcing data truth reasoning method is as follows: Figure 4 As shown, the main steps of the algorithm are described in detail.
[0078] Input: D: crowdsourced dataset,
[0079] M: the number of weak classifiers;
[0080] output: Aggregated tags;
[0081] 1. Load the crowdsourced data set D, and divide the data set D into a training set D in a certain proportion T with test set D L ; Use resampling for each weak classifier to D T Sampling to generate subdatasets
[0082] 2. Calculate The proportion of positive and negative categories in Pr, the conversion data set is And train the weak classifier h i (x);
[0083] 3. Aggregating weak classifiers based on maximum likelihood estimation;
[0084] 4. Compare l ij with H(x i ) predicted labels get:
[0085]
[0086] 5. Compare the similarity of each worker
[0087] 6. Combination τ j ,s ij Get the reliability of the jth worke...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More