Method and system for determining result for task executed in crowd-sourced environment

1. A method of determining a result for a task executed in a crowd-sourced environment, the task being of a given type of tasks requiring a result in textual format, the method being executed by a server, the method comprising:

receiving, by the server, a plurality of results of the task having been submitted by a plurality of human assessors of the crowd-sourced environment;

receiving, by the server, a quality score for each human assessor of the plurality of human assessors, the quality score being indicative of an error rate of the human assessor executing the given type of tasks;

generating, by the server, a plurality of vector representations comprising a vector representation for each of the results of the plurality of results;

associating, by the server, each vector representation of the plurality of vector representations, with the quality score of the human assessor associated with each vector representation;

mapping, by the server, the plurality of vector representations into a vector space;

clustering, by the server, the plurality of vector representations within the vector space into at least a first cluster and a second cluster, the first cluster including a first subset of the plurality of vector representations and the second cluster including a second subset of the plurality of vector representations;

executing, by the server, a machine learning (MLA) algorithm configured to determine a first confidence parameter and a second confidence parameter,

the first confidence parameter corresponding to a probability of the first subset having a vector representation of a correct result of the task, the first confidence parameter being determined based on the quality scores associated with a first subset of the plurality of human assessors previously associated with the first subset of the plurality of vector representations;

the second confidence parameter corresponding to a probability of the second subset having the vector representation of the correct result of the task, the second confidence parameter determined based on the quality scores associated with a second subset of the plurality of human assessors previously associated with the second subset of the plurality of vector representations;

in response to the first confidence parameter being above the second confidence parameter identifying, by the server, a given vector representation of the first subset of the plurality of vector representations as a representative vector for the plurality of vector representations, the given vector representation to be used as an aggregated vector representation for a single textual result of the task; and

inputting, by the server, the aggregated vector representation into a vector-to-text routine to generate the single textual result;

selecting, by the server, the single textual result as the result of the task.

2. The method of claim 1 , wherein the given type of task is one of a translation task and a speech-to-text conversion task.

3. The method of claim 1 , wherein the result is being represented in a textual form, and wherein:

the generating the plurality of vector representations comprises generating a word embedding for each of the plurality of the textual form, using one of a Word2Vec model, a continuous bag-of-words model, and a continuous skip-gram model.

4. The method of claim 1 , wherein the clustering the plurality of vector representations is executed by one of a k-means clustering algorithm, a centroid-based clustering algorithm, a distribution-based clustering algorithm, and a density-based clustering algorithm.

5. The method of claim 1 , wherein the MLA is a Dawid-Skene model algorithm.

6. The method of claim 1 , wherein

identifying the given vector representation comprises selecting the given vector representation corresponding to a medoid of the first subset of the plurality of vector representations.

7. The method of claim 1 , wherein

identifying the given vector representation comprises selecting a most frequent vector representation within the first subset of the plurality of vector representations.

8. The method of claim 1 , wherein identifying the given vector representation comprises:

generating a first given vector representation corresponding to a centroid of the first subset of the plurality of vector representations;

selecting from the first subset of the plurality of vector representations a second given vector representation corresponding to a most similar vector representation to the centroid as the given vector representation.

9. A system for determining a result for a task executed in a crowd-sourced environment, the task being of a given type of tasks requiring a result in textual format, the system comprising a server, the server comprising a processor configured to:

receive, by the server, a plurality of results of the task having been submitted by a plurality of human assessors of the crowd-sourced environment;

receive, by the server, a quality score for each human assessor of the plurality of human assessors, the quality score being indicative of an error rate of the human assessor executing the given type of tasks;

generate, by the server, a plurality of vector representations comprising a vector representation for each of the results of the plurality of results;

associate, by the server, each vector representation of the plurality of vector representations, with the quality score of the human assessor associated with each vector representation;

map, by the server, the plurality of vector representations into a vector space;

cluster, by the server, the plurality of vector representations within the vector space into at least a first cluster and a second cluster, the first cluster including a first subset of the plurality of vector representations and the second cluster including a second subset of the plurality of vector representations;

execute, by the server, a machine learning (MLA) algorithm configured to determine a first confidence parameter and a second confidence parameter,

the first confidence parameter corresponding to a probability of the first subset having a vector representation of a correct result of the task, the first confidence parameter being determined based on the quality scores associated with a first subset of the plurality of human assessors previously associated with the first subset of the plurality of vector representations;

the second confidence parameter corresponding to a probability of the second subset having the vector representation of the correct result of the task, the second confidence parameter being determined based on the quality scores associated with a second subset of the plurality of human assessors previously associated with the second subset of the plurality of vector representations;

in response to the first confidence parameter being above the second confidence parameter, identify, by the server, a given vector representation of the first subset of the plurality of vector representations as a representative for the plurality of vector representations, the given vector representation to be used as an aggregated vector representation for a single textual result of the task; and

input, by the server, the aggregated vector representation into a vector-to-text routine to generate the single textual result;

select, by the server, the single textual result as the result of the task.

10. The system of claim 9 , wherein the result is being represented in a textual form, and wherein:

to generate the plurality of vector representations, the processor is configured to generate a word embedding for each of the plurality of the textual form, using one of a Word2Vec model, a continuous bag-of-words model, and a continuous skip-gram model.

11. The system of claim 9 , wherein to cluster the plurality of vector representations, the processor is configured to execute one of a k-means clustering algorithm, a centroid-based clustering algorithm, a distribution-based clustering algorithm, and a density-based clustering algorithm.

12. The system of claim 9 , wherein the MLA is a Dawid-Skene model algorithm.

13. The system of claim 9 , wherein

to identify the given vector representation, the processor is configured to select the given vector representation corresponding to a medoid of the first subset of the plurality of vector representations.

14. The system of claim 9 , wherein

to identify the given vector representation, the processor is configured to select a most frequent vector representation within the first subset of the plurality of vector representations.

15. The system of claim 9 , wherein to identify the give vector representation, the processor is configured to:

generate a first given vector representation corresponding to a centroid of the first subset of the plurality of vector representations;

select from the first subset of the plurality of vector representations a second given vector representation corresponding to a most similar vector representation to the centroid as the given vector representation.