System for optimizing training dataset

1 . A system for reducing computation resource used by a data processing system for training a machine leaning model for classifying documents, the system comprising one or more processors configured to:

receive a training dataset comprising digital copies of sample documents;

create graph embedding vector for each of the sample documents;

cluster the graph embedding vectors of the sample documents of the training dataset into clusters based on the similarity between the graph embedding vectors;

select a first set of training data comprising a subset of the graph embedding vectors, wherein selecting the first set of training data comprises:

selecting a subset of the clusters; and

selecting graph embedding vectors from the selected clusters, wherein graph embedding vectors are selected from each of the selected clusters, and wherein in one or more of the selected clusters, having the graph embedding vectors more than a predefined upper threshold value, only a subset of the graph embedding vectors is included in the first set of training data; and

input the first set of training data as input training data for the machine learning model.

2 . The system as claimed in claim 1 , wherein to select the subset of the clusters, the one or more processor is configured to:

determine the size of the cluster, wherein the size of the cluster represents the number of graph embedding vectors of the sample documents in the cluster;

if the size of the cluster is less than a predefined lower threshold value, the cluster is ignored from being used and samples in the cluster are retained as an input training data for the machine learning model; and

else, a finite number of graph embedding vectors of the sample documents is selected from the cluster as the first set of training data, wherein the finite number of the graph embedding vectors in the first set of training data is based on the predefined upper threshold value.

3 . The system as claimed in claim 2 , wherein the selected first set of training data from the clustered training data further comprises:

positive samples, wherein the positive samples include the graph embedding vectors of the cluster that are within a boundary of the cluster; and

negative samples, wherein the negative samples include the graph embedding vectors of the cluster that are outside the boundary of the cluster.

4 . The system as claimed in claim 3 , wherein the number of positive samples is between the lower threshold value and the upper threshold value, if the size of the cluster is less than or equal to the upper threshold value.

5 . The system as claimed in claim 3 , wherein the number of positive samples selected is equal to the upper threshold value, if the size of the cluster is greater than the upper threshold value.

6 . The system as claimed in claim 3 , wherein the number of negative samples selected is equal to 10% of the number of positive samples selected.

7 . The system as claimed in claim 3 , wherein:

each of the graph embedding vectors of the sample documents in the cluster includes a threshold score;

the positive samples comprise of graph embedding vectors with threshold score in the range of 0.4-0.5; and

the negative samples comprise of graph embedding vectors with threshold score above 0.5.

8 . The system as claimed in claim 3 , wherein the one or more processors is configured to combine the positive samples and the negative samples selected from each of the clusters to obtain the first set of training data to be input to the machine learning model.

9 . The system as claimed in claim 2 , wherein:

the lower threshold value is 100; and

the upper threshold value is 500.