Systems and methods for training machine learning models with user-specific knowledge graphs to predict outcomes
A method includes receiving interaction data including more than one interaction type and providing the interaction data to a first machine learning model to generate a vector representation defining similarity measures between subsets of the interaction data and update a knowledge graph. The method includes retrieving data associated with a user based on a target associated with a prediction associated with a capability level the user and providing inputs to multiple machine learning models to define relative complexity scores, pseudoguessing weights, and an item discrimination index, which are provided to a transformer trained on the knowledge graph to generate at least one prediction. The method includes providing the at least one prediction to a machine learning model to generate a plurality of simulations. The method includes identifying, based on the plurality of simulations, a prediction associated with a capability level of the user.
1 . A non-transitory processor-readable medium storing code representing instructions to be executed by one or more processors, the instructions comprising code to cause the one or more processors to:
receive interaction data associated with a user, the interaction data including more than one interaction type;
provide the interaction data as input to a first machine learning model to generate a vector representation of the interaction data, the vector representation defining similarity measures between subsets of the interaction data regardless of interaction type;
remove, from the vector representation, similarity measures below a predetermined threshold, the predetermined threshold associated with a bi-partite graph;
update, based on the vector representation, a knowledge graph associated with relationships between units from clusters of units;
train, based on the knowledge graph, an auto regressive transformer;
retrieve, based on a target for a prediction associated with a capability level of the user, data associated with the user, the target associated with at least one unit from the clusters of units;
provide the target as input to a second machine learning model to define a relative complexity score for the data associated with user;
provide a format associated with the data associated with the user as input to a third machine learning model to generate pseudoguessing weights for the data associated with the user;
provide point-biserial correlation coefficients associated with the data as input to a fourth machine learning model to generate an item discrimination index for the data associated with the user;
generate at least one prediction associated with the target by providing the relative complexity score, the pseudoguessing weights, and the item discrimination index as input to the auto regressive transformer;
provide the at least one prediction as input to a fifth machine learning model to generate a plurality of simulations; and
identify, based on the plurality of simulations, the prediction associated with the capability level of the user.
2 . The non-transitory processor-readable medium of claim 1 , wherein the plurality of simulations is based on item response theory (IRT).
3 . The non-transitory processor-readable medium of claim 1 , wherein the instructions further comprise code to cause the one or more processors to:
extract, from the target, an identified context, wherein defining the relative complexity score is based on the identified context.
4 . The non-transitory processor-readable medium of claim 3 , wherein the identified context is a subject associated with the user.
5 . The non-transitory processor-readable medium of claim 1 , wherein generating the item discrimination index is based on data associated with a plurality of students, the plurality of students including relatively high knowledge students and relatively low knowledge students.
6 . The non-transitory processor-readable medium of claim 1 , wherein the instructions further comprise code to cause the one or more processors to:
receive user information associated with the user; and
update the knowledge graph based on the user information.
7 . The non-transitory processor-readable medium of claim 6 , wherein the user information includes at least one of user device information, a user profile, chatbot conversations, input device information, explored content, or user device conversations.
8 . The non-transitory processor-readable medium of claim 1 , wherein the more than one interaction type includes more than one of conversation data, assessment data, assignment data, question data, or engagement data.
9 . The non-transitory processor-readable medium of claim 1 , wherein the at least one prediction includes a probability associated with the target.
10 . A non-transitory processor-readable medium storing code representing instructions to be executed by one or more processors, the instructions comprising code to cause the one or more processors to:
receive interaction data, the interaction data including more than one interaction type;
provide the interaction data to as an input to a first machine learning model to generate a vector representation of the interaction data, the vector representation defining similarity measures between subsets of the interaction data regardless of interaction type;
remove, from the vector representation, similarity measures below a predetermined threshold, the predetermined threshold associated with a bi-partite graph;
generate, based on the vector representation, a knowledge graph associated with relationships between units of clusters of units;
train a transformer based on the knowledge graph;
retrieve, based on a target for a prediction associated with a capability level of a user, data associated with the user, the target associated with at least one unit from the clusters of units;
provide a context associated with the target as an input to a second machine learning model to define a relative complexity score for the data associated with the user;
provide a format associated with the data associated with the user as an input to a third machine learning model to define pseudoguessing weights for the data associated with the user;
provide point-biserial correlation coefficients associated with the data as an input to a fourth machine learning model to define an item discrimination index for the data associated with the user;
generate at least one prediction associated with the target by providing the relative complexity score, the pseudoguessing weights, and the item discrimination index as an input to the transformer;
provide the at least one prediction to a fifth machine learning model to generate a plurality of simulations; and
identify, based on the plurality of simulations, the prediction associated with the capability level of the user.
11 . The non-transitory processor-readable medium of claim 10 , wherein generating the vector representation is based on semantic similarities between the data in the interaction data.
12 . The non-transitory processor-readable medium of claim 10 , wherein the first machine learning model is a natural language model.
13 . The non-transitory processor-readable medium of claim 10 , wherein each of the second machine learning model, the third machine learning model, and the fourth machine learning model is a different type of machine learning model.
14 . The non-transitory processor-readable medium of claim 10 , wherein the plurality of simulations is based on item response theory (IRT).
15 . The non-transitory processor-readable medium of claim 10 , wherein the more than one interaction type includes more than one of conversation data, assessment data, assignment data, question data, or engagement data.
16 . An apparatus comprising:
one or more memories; and
one or more processors operatively coupled to the one or more memories, the one or more processors configured to:
retrieve, based on a target, associated with at least one unit from clusters of units, for a prediction associated with a capability level of a user, data associated with the user, the data associated with a knowledge graph associated with relationships between units of the clusters of units;
provide the target as an input to a first machine learning model to generate a vector representation of the target;
remove, from the vector representation, one or more representations having a similarity value below a predetermined threshold, the predetermined threshold associated with a bi-partite graph;
training a transformer based on the knowledge graph;
update the knowledge graph based on the vector representation;
provide a context associated with the target as an input to a second machine learning model to define a relative complexity score for the data associated with the user;
provide a format associated with the data associated with the user as an input to a third machine learning model to define pseudoguessing weights for the data associated with the user;
provide point biserial correlation coefficients associated with the data to a fourth machine learning model to define an item discrimination index for the data associated with the user;
generate at least one prediction associated with the target by providing the relative complexity score, the pseudoguessing weights, and the item discrimination index as an input to the transformer;
provide the at least one prediction to a fifth machine learning model to generate a plurality of simulations; and
identify, based on the plurality of simulations, the prediction associated with the capability level of the user.
17 . The apparatus of claim 16 , wherein the plurality of simulations is based on item response theory (IRT).
18 . The apparatus of claim 16 , wherein the data associated with the knowledge graph includes more than one of conversation data, assessment data, assignment data, question data, or engagement data.
19 . The apparatus of claim 16 , wherein the transformer is an auto regressive transformer model.
20 . The apparatus of claim 16 , wherein the one or more processors are further configured to:
define a subset of data, wherein the subset of data includes data with at least one of an item discrimination index below zero or an item discrimination index within a predefined threshold of zero; and
store the subset of data in a database for review.
21 . The apparatus of claim 16 , wherein the target is a request for information associated with a future user skill prediction.