Annotating and collecting data-centric AI quality metrics considering user preferences
A method, computer program, and computer system are provided for collecting and annotating data based on user preference. Unlabeled data corresponding to one or more entries within a dataset is received. Pseudo-labeled data is generated based on the unlabeled data. Based on one or more quality metrics, each entry from among the pseudo-labeled data is determining to be included within a final dataset. A user is prompted for annotations corresponding to entries of the pseudo-labeled data included within the final dataset. A determination is made as to whether additional data is needed based on comparing the final dataset to the one or more quality metrics, and the additional information is collected if the final dataset does not meet the quality metrics.
1. A method of collecting and annotating data based on user preference, executable by a processor, comprising:
receiving unlabeled data corresponding to one or more entries within a dataset;
generating pseudo-labeled data based on the unlabeled data and based on unsupervised clustering of the unlabeled data;
receiving user-provided annotations corresponding to entries of the pseudo-labeled data;
re-clustering the unlabeled data based on the user-provided annotations;
regenerating the pseudo-labeled data based on the re-clustering;
determining, based on one or more quality metrics, whether each entry from among the pseudo-labeled data is to be included within a final dataset, wherein the one or more entries are selected by two or more hidden layers corresponding to, at least, an input layer and an output, wherein the input layer corresponds to a feature map of input samples and the output layer corresponds to a probability of considering each entry for inclusion in the final dataset;
upon determining that user preferences are not met, providing a recommendation to a user to collect additional data, wherein the recommendation includes providing samples to the user from labeled data that contribute most to the one or more quality metrics; and
receiving additional user-provided annotations corresponding to the samples.
2. The method of claim 1 , wherein the unsupervised clustering is performed by an unsupervised machine learning model.
3. The method of claim 1 , wherein based on the user-provided annotations, at least one of the clusters are modified in accordance with a selection from the group consisting of:
(i) split, depending on a presence of different class labelled samples in the same cluster, and
(ii) merged, if adjacent clusters have the same class labelled samples.
4. The method of claim 1 , wherein the one or more quality metrics comprise label noise, class imbalance, class overlap, and outliers.
5. The method of claim 1 , further comprising collecting the additional data based on the final dataset not meeting the one or more quality metrics.
6. A computer system for collecting and annotating data based on user preference, the computer system comprising:
one or more computer-readable non-transitory storage media configured to store computer program code; and
one or more computer processors configured to access said computer program code and operate as instructed by said computer program code, said computer program code including:
receiving code configured to cause the one or more computer processors to receive unlabeled data corresponding to one or more entries within a dataset;
generating code configured to cause the one or more computer processors to generate pseudo-labeled data based on the unlabeled data and based on unsupervised clustering of the unlabeled data;
receiving user-provided annotations corresponding to entries of the pseudo-labeled data;
re-clustering the unlabeled data based on the user-provided annotations;
regenerating the pseudo-labeled data based on the re-clustering;
determining code configured to cause the one or more computer processors to determine, based on one or more quality metrics, whether each entry from among the pseudo-labeled data is to be included within a final dataset, wherein the one or more entries are selected by two or more hidden layers corresponding to, at least, an input layer and an output, wherein the input layer corresponds to a feature map of input samples and the output layer corresponds to a probability of considering each entry for inclusion in the final dataset;
upon determining that user preferences are not met, providing a recommendation to a user to collect additional data, wherein the recommendation includes providing samples to the user from labeled data that contribute most to the one or more quality metrics; and
receiving additional user-provided annotations corresponding to the samples.
7. The computer system claim 6 , wherein the unsupervised clustering is performed by an unsupervised machine learning model.
8. The computer system of claim 6 , wherein based on the user-provided annotations, at least one of the clusters are modified in accordance with a selection from the group consisting of:
(i) split, depending on a presence of different class labelled samples in the same cluster, and
(ii) merged, if adjacent clusters have the same class labelled samples.
9. The computer system of claim 6 , wherein the one or more quality metrics comprise label noise, class imbalance, class overlap, and outliers.
10. The computer system of claim 6 , further comprising collecting code configured to cause the one or more computer processors to collect the additional data based on the final dataset not meeting the one or more quality metrics.
11. A non-transitory computer readable medium having stored thereon a computer program for collecting and annotating data based on user preference, the computer program configured to cause one or more computer processors to:
receive unlabeled data corresponding to one or more entries within a dataset;
generate pseudo-labeled data based on the unlabeled data and based on unsupervised clustering of the unlabeled data;
receive user-provided annotations corresponding to entries of the pseudo-labeled data;
re-cluster the unlabeled data based on the user-provided annotations;
regenerate the pseudo-labeled data based on the re-clustering;
determine, based on one or more quality metrics, whether each entry from among the pseudo-labeled data is to be included within a final dataset, wherein the one or more entries are selected by two or more hidden layers corresponding to, at least, an input layer and an output, wherein the input layer corresponds to a feature map of input samples and the output layer corresponds to a probability of considering each entry for inclusion in the final dataset;
upon determining that user preferences are not met, providing a recommendation to a user to collect additional data, wherein the recommendation includes providing samples to the user from labeled data that contribute most to the one or more quality metrics; and
receiving additional user-provided annotations corresponding to the samples.
12. The computer readable medium of claim 11 , wherein the unsupervised clustering is performed by an unsupervised machine learning model.
13. The computer readable medium of claim 11 , wherein based on the user-provided annotations, at least one of the clusters are modified in accordance with a selection from the group consisting of:
(i) split, depending on a presence of different class labelled samples in the same cluster, and
(ii) merged, if adjacent clusters have the same class labelled samples.
14. The computer readable medium of claim 11 , wherein the one or more quality metrics comprise label noise, class imbalance, class overlap, and outliers.
15. The computer readable medium of claim 11 , wherein the computer program is further configured to cause the one or more computer processors to collect the additional data based on the final dataset not meeting the one or more quality metrics.