System and method for merging annotations of datasets
Systems and methods for merging annotations of datasets are provided. For example, assignments of labels to data-points may be obtained, confidence levels associated with the assignments of labels may be obtained. Further, the assignments of labels may be merged, for example based on the confidence levels. In some cases, inference models may be generated using the merged assignment of labels. In some examples, an update to the assignments of labels to data-points and/or the confidence levels may be obtained, and the merged assignment of labels may be updated.
1. A system for merging annotations of datasets, the system comprising:
at least one processor configured to:
obtain a first assignment of labels to a plurality of data-points;
obtain a second assignment of labels to the plurality of data-points;
obtain a third assignment of labels to the plurality of data-points;
obtain confidence levels associated with the first assignment of labels;
obtain confidence levels associated with the second assignment of labels;
obtain confidence levels associated with the third assignment of labels; and
generate a merged assignment of labels to the plurality of data-points based on the first assignment of labels, the second assignment of labels, the third assignment of labels, the confidence levels associated with the first assignment of labels, the confidence levels associated with the second assignment of labels and the confidence levels associated with the third assignment of labels.
2. The system of claim 1 , wherein the at least one processor is further configured to:
using the confidence levels associated with the first assignment of labels and the confidence levels associated with the second assignment of labels, identify one or more data-points of the plurality of data-points as problematic for merging labels; and
provide a suggestion to improve the assignment of labels corresponding to the identified one or more data-points.
3. The system of claim 1 , wherein the at least one processor is further configured to minimize or maximize an objective function to generate the merged assignment of labels to the plurality of data-points.
4. The system of claim 1 , wherein generating the merged assignment of labels to the plurality of data-points is further based on a quota requirement associated with the first assignment of labels.
5. The system of claim 1 , wherein the first assignment of labels comprises an assignment of labels to the plurality of data-points by a user, the confidence levels associated with the first assignment of labels is based on an evaluation of past performance of the user, and the second assignment of labels comprises an assignment of labels to the plurality of data-points by an automated process.
6. The system of claim 1 , wherein the at least one processor is further configured to:
obtain an update to at least one of the first assignment of labels and the confidence levels associated with the first assignment of labels;
analyze the update to determine that a magnitude of the update is above a selected threshold; and
based on the determination that the magnitude of the update is above the selected threshold, update the merged assignment of labels based on the obtained update.
7. A method for merging annotations of datasets, the method comprising:
obtaining a first assignment of labels to a plurality of data-points;
obtaining a second assignment of labels to the plurality of data-points;
obtaining a third assignment of labels to the plurality of data-points;
obtaining confidence levels associated with the first assignment of labels;
obtaining confidence levels associated with the second assignment of labels;
obtaining confidence levels associated with the third assignment of labels; and
generating a merged assignment of labels to the plurality of data-points based on the first assignment of labels, the second assignment of labels, the third assignment of labels, the confidence levels associated with the first assignment of labels, the confidence levels associated with the second assignment of labels and the confidence levels associated with the third assignment of labels.
8. The method of claim 7 , further comprising:
using the confidence levels associated with the first assignment of labels and the confidence levels associated with the second assignment of labels, identifying one or more data-points of the plurality of data-points as problematic for merging labels; and
providing a suggestion to improve the assignment of labels corresponding to the identified one or more data-points.
9. The method of claim 7 , further comprising at least one of minimizing and maximizing an objective function to generate the merged assignment of labels to the plurality of data-points.
10. The method of claim 9 , wherein a term associated with a quota is used in the at least one of minimizing and maximizing of the objective function.
11. The method of claim 7 , wherein generating the merged assignment of labels to the plurality of data-points is further based on a quota requirement associated with the first assignment of labels.
12. The method of claim 7 , wherein the first assignment of labels comprises an assignment of labels to the plurality of data-points by a user, and the confidence levels associated with the first assignment of labels is based on an evaluation of past performance of the user.
13. The method of claim 7 , wherein the first assignment of labels comprises an assignment of labels to the plurality of data-points by an automated process, and the confidence levels associated with the first assignment of labels is based on an evaluation of past performance of the automated process.
14. The method of claim 7 , wherein the first assignment of labels comprises an assignment of labels to the plurality of data-points by a user, the confidence levels associated with the first assignment of labels is based on an evaluation of past performance of the user, and the second assignment of labels comprises an assignment of labels to the plurality of data-points by an automated process.
15. The method of claim 7 , wherein the confidence levels associated with the first assignment of labels is based on an output of an inference model, and the inference model is a result of applying the first assignment of labels to a machine learning algorithm.
16. The method of claim 7 , wherein generating the merged assignment of labels to the plurality of data-points is further based on the plurality of data-points.
17. The method of claim 7 , wherein the confidence levels associated with the first assignment of labels comprises a function that maps types of data-points to confidence levels, and wherein generating the merged assignment of labels to the plurality of data-points is further based on an association of types with data-points and on the function.
18. The method of claim 7 , further comprising applying the merged assignment of labels to a machine learning algorithm to obtain an inference model.
19. The method of claim 7 , further comprising:
obtaining an update to at least one of the first assignment of labels and the confidence levels associated with the first assignment of labels;
analyzing the update to determine that a magnitude of the update is above a selected threshold; and
based on the determination that the magnitude of the update is above the selected threshold, updating the merged assignment of labels based on the obtained update.
20. A non-transitory computer readable medium storing data and computer implementable instructions for carrying out a method for merging annotations of datasets, the method comprising:
obtaining a first assignment of labels to a plurality of data-points;
obtaining a second assignment of labels to the plurality of data-points;
obtaining a third assignment of labels to the plurality of data-points;
obtaining confidence levels associated with the first assignment of labels;
obtaining confidence levels associated with the second assignment of labels;
obtaining confidence levels associated with the third assignment of labels; and
generating a merged assignment of labels to the plurality of data-points based on the first assignment of labels, the second assignment of labels, the third assignment of labels, the confidence levels associated with the first assignment of labels, the confidence levels associated with the second assignment of labels and the confidence levels associated with the third assignment of labels.