Automated determination of degree of item similarity in the generation of digitized examinations

1. A computer implemented method for identifying and removing similar examination items in a digital examination, the method comprising:

obtaining, with an examination evaluation server, the digital examination, the digital examination comprising a plurality of examination items;

tokenizing a first examination item of the plurality of examination items into a first set of n-grams;

tokenizing a second examination item of the plurality of examination items into a second set of n-grams;

generating a first item vector from the first set of n-grams;

generating a second item vector from the second set of n-grams;

generating an item congruence value by comparing the first set of n-grams with the second set of n-grams;

determining a threshold value for identifying enemy item pairs using a machine learning model, the machine learning model comprising:

obtaining a training set of digital examinations, wherein each digital examination of the training set of digital examination comprises multiple training examination items;

determining, for pairs of training examination items, training item congruence values;

obtaining, from a graphical user interface, indications as to whether training examination items from the pairs of training examination items are enemy items; and

determining the threshold value based on the congruence values and the indications; and

identifying that the first and second examination items are a potential enemy item pair when the item congruence value exceeds the threshold value.

2. The method of claim 1 , wherein the item congruence value comprises a cosine of an angle between the first item vector and the second item vector.

3. The method of claim 2 , wherein the item congruence value further comprises a weighting coefficient applied to the cosine of the angle between the first item vector and the second item vector.

4. The method of claim 2 , wherein the threshold value is about 0.75.

5. The method of claim 1 , wherein the machine learning model is a convolutional neural network or a gradient boosted learning tree.

6. The method of claim 1 , wherein generating the item congruence value comprises evaluating similarity between the first item vector and the second item vector relative to a target similarity dimension.

7. The method of claim 6 , wherein the target similarity dimension comprises semantic similarity, syntactic similarity, lexical similarity, or content similarity.

8. The method of claim 7 , further comprising applying a weighting coefficient to the item congruence value, the weighting coefficient being based on the target similarity dimension.

9. The method of claim 6 , wherein the target similarity dimension comprises semantic similarity, syntactic similarity, lexical similarity, and content similarity.

10. The method of claim 9 , wherein the identifying that the second examination item is the potential enemy item further comprises generating an array of congruence values, the array of congruence values comprising a semantic congruence value, a syntactic congruence value, a lexical congruence value, and a content congruence value.

11. The method of claim 10 , further comprising:

displaying, on the graphical user interface, the potential enemy item and the array of congruence values;

obtaining, from the graphical user interface, an indication that the potential enemy item is a confirmed enemy item; and

removing the confirmed enemy item from the digital examination.

12. The method of claim 1 , further comprising determining that the potential enemy item is a confirmed enemy item and removing the confirmed enemy item from the digital examination.

13. The method of claim 12 , wherein the determining that the potential enemy item is the confirmed enemy item comprises displaying, on the graphical user interface, the potential enemy item and obtaining, from the graphical user interface, an indication that the potential enemy item is the confirmed enemy item.

14. A system for enemy items from a digital examination, the system comprising:

an examination evaluation server;

a data store; and

a graphical user interface;

wherein the examination evaluation server comprises a processor and a non-transitory computer readable medium with computer executable instructions embedded thereon, the computer executable instructions configured to cause the processor to:

obtain the digital examination comprising a plurality of examination items;

tokenize a first examination item of the plurality of examination items into a first set of n-grams;

tokenize a second examination item of the plurality of examination items into a second set of n-grams;

generate a first item vector from the first set of n-grams;

generate a second item vector from the second set of n-grams;

evaluate similarity between the first item vector and the second item vector relative to a first target similarity dimension to generate a first item congruence value;

evaluate similarity between the first set of n-grams and the second set of n-grams relative to a second target similarity dimension to generate a second item congruence value;

store the first and second item congruence values into an item congruence array located on the data store;

determine a first threshold value and a second threshold value using a machine learning logical circuit, wherein the machine learning logical circuit is configured to:

obtain a training set of digital examinations, wherein each digital examination of the training set of digital examinations comprises multiple training examination items;

determine, for pairs of training examination items, training item congruence values;

obtain, from the graphical user interface, indications as to whether training examination items from the pairs of training examination items are enemy items; and

determine the first threshold value and the second threshold value based on the congruence values and the indications;

identify the second examination item as a potential enemy item if the first item congruence value exceeds the first threshold value and the second item congruence value exceeds the second threshold value; and

cause the graphical user interface to display the digital examination without the potential enemy item.

15. The system of claim 14 , wherein the first item congruence value is a cosine of an angle between the first item vector and the second item vector when evaluated relative to the first target similarity dimension and the second item congruence value is a cosine of an angle between the first item vector and the second item vector when evaluated relative to the second target similarity dimension.

16. The system of claim 14 , wherein the machine learning logical circuit is further configured to apply a convolutional neural network or a gradient boosted learning tree.

17. The system of claim 14 , wherein the first and second target similarity dimensions comprise semantic similarity, syntactic similarity, lexical similarity, or content similarity.

18. A computer implemented method for identifying and removing similar examination items in a digital examination, the method comprising:

obtaining, with an examination evaluation server, the digital examination, the digital examination comprising a plurality of examination items;

tokenizing a first examination item of the plurality of examination items into a first set of n-grams;

tokenizing a second examination item of the plurality of examination items into a second set of n-grams;

generating a first item vector from the first set of n-grams;

generating a second item vector from the second set of n-grams;

generating a semantic congruence value by evaluating semantic similarity between the first set of n-grams with the second set of n-grams;

generating a syntactic congruence value by evaluating syntactic similarity between the first set of n-grams with the second set of n-grams;

generating a lexical congruence value by evaluating lexical similarity between the first set of n-grams with the second set of n-grams;

generating a content congruence value by evaluating content similarity between the first set of n-grams with the second set of n-grams, wherein the semantic congruence value, the syntactic congruence value, the lexical congruence value, or the content congruence value is generated using a machine learning model, the machine learning model comprising:

obtaining a training set of digital examinations, wherein each digital examination of the training set of digital examination comprises multiple training examination items;

determining, for pairs of training examination items, the semantic congruence value, the syntactic congruence value, the lexical congruence value, or the content congruence value using the machine learning model;

obtaining, from a graphical user interface, indications as to whether training examination items from the pairs of training examination items are enemy items; and

updating the semantic congruence value, the syntactic congruence value, the lexical congruence value, or the content congruence value based on the indications;

storing the semantic, syntactic, lexical, and content congruence values in a congruence array; and

identifying that the first examination item and the second examination item are a potential enemy item pair based on a comparison of the congruence array with an array of semantic, syntactic, lexical, and content threshold congruence values.

19. The method of claim 18 , wherein the machine learning model is a convolutional neural network or a gradient boosted learning tree.

20. The method of claim 18 , further comprising applying a weighting coefficient to the semantic congruence value, the syntactic congruence value, the lexical congruence value, or the content congruence value, the weighting coefficient being based on a target similarity dimension.