Content recommendation using retrieval augmented artificial intelligence
Systems, methods, apparatuses, and computer program products are disclosed for using retrieval augmented artificial intelligence to provide content recommendations. A first feature vector is generated based at least on user contextual information. Second feature vectors are determined based on a comparison of the first feature vector to a plurality of second feature vectors. Content items corresponding to the determined second feature vectors are retrieved. An augmented prompt generated based on the user contextual information and the retrieved content items is provided to a large language model to request a recommendation. A content recommendation is received from the large language model based on the augmented prompt.
1 . A method for improving the relevancy of a content recommendation, comprising:
receiving user contextual information comprising at least an identifier of a current webpage;
encoding a concatenation of at least the user contextual information and a content of the current webpage into a first feature vector, the first feature vector comprising a low-dimensional dense vector;
encoding concatenations of user historical information, product information, and content information into a plurality of second feature vectors, the second feature vectors comprising low-dimensional dense vectors;
determining a cosine similarity between a first portion of the first feature vector and a corresponding portion of each of the plurality of second feature vectors, wherein determining the cosine similarity ignores a second portion of the first feature vector corresponding to user contextual information having a null value;
retrieving text associated with a second feature vector having a cosine similarity to the first feature vector that satisfies a predetermined condition;
providing, to a large language model, an augmented prompt generated based at least on the user contextual information and the retrieved text; and
receiving, from the large language model and based on the augmented prompt, a summary of a recommended text.
2 . The method of claim 1 , wherein the summary comprises a user-selectable link to the recommended text.
3 . The method of claim 1 , wherein said retrieving text associated with a second feature vector comprises:
determining the second feature vectors having a cosine similarity to the first feature vector that satisfies a first predetermined relationship with a first predetermined threshold;
determining a first predetermined number of second feature vectors having highest cosine similarities to the first feature vector; or
determining a second predetermined number of second feature vectors having highest cosine similarities to the first feature vector that satisfy a second predetermined relationship with a second predetermined threshold.
4 . The method of claim 1 , wherein the augmented prompt comprises:
a query requesting a recommendation for a user based at least on the retrieved text; and
the retrieved text.
5 . The method of claim 1 , wherein at least one of the first feature vector or the plurality of second feature vectors is encoded:
using a Generative Pre-Trained Transformer (GPT)-based or a Bidirectional Encoder Representations from Transformers (BERT)-based encoder.
6 . The method of claim 1 , wherein the user contextual information comprises one or more of:
a user identifier;
user profile information;
user location information;
temporal information;
product or service information;
user intent information; or
user interest information.
7 . The method of claim 1 , further comprising:
providing the summary of the recommended text to a user device.
8 . A system for improving the relevancy of a content recommendation, comprising:
a processor; and
a computer-readable storage medium comprising computer-executable instructions, that when executed by the processor, cause the processor to:
receive user contextual information comprising at least an identifier of a current webpage;
encoding a concatenation of at least the user contextual information and a content of the current webpage into a first feature vector, the first feature vector comprising a low-dimensional dense vector;
encode concatenations of user historical information, product information, and content information into a plurality of second feature vectors, the second feature vectors comprising low-dimensional dense vectors;
determine a cosine similarity between a first portion of the first feature vector and a corresponding portion of each of the plurality of second feature vectors, wherein determining the cosine similarity ignores a second portion of the first feature vector corresponding to user contextual information having a null value;
retrieve text associated with a second feature vector having a cosine similarity to the first feature vector that satisfies a predetermined condition;
provide, to a large language model, an augmented prompt generated based at least on the user contextual information and the retrieved text; and
receive, from the large language model and based on the augmented prompt, a summary of a recommended text.
9 . The system of claim 8 , wherein the summary comprises a user-selectable link to the recommended text.
10 . The system of claim 8 , wherein said retrieve text associated with a second feature vector comprises:
determining the second feature vectors having a cosine similarity to the first feature vector that satisfies a first predetermined relationship with a first predetermined threshold;
determining a first predetermined number of second feature vectors having highest cosine similarities to the first feature vector; or
determining a second predetermined number of second feature vectors having highest cosine similarities to the first feature vector that satisfy a second predetermined relationship with a second predetermined threshold.
11 . The system of claim 8 , wherein the augmented prompt comprises:
a query requesting a recommendation for a user based at least on the retrieved text; and
the retrieved text.
12 . The system of claim 8 , wherein at least one of the first feature vector or the plurality of second feature vectors is encoded
using a Generative Pre-Trained Transformer (GPT)-based or a Bidirectional Encoder Representations from Transformers (BERT)-based encoder.
13 . The system of claim 8 , wherein the user contextual information comprises one or more of:
a user identifier;
user profile information;
user location information;
temporal information;
product or service information;
user intent information; or
user interest information.
14 . The system of claim 8 , wherein the computer-executable instructions, when executed by the processor, further cause the processor to:
provide the summary of the recommended text to a user device.
15 . A computer-readable storage medium comprising computer-executable instructions, that when executed by a processor, cause the processor to:
receive user contextual information comprising at least an identifier of a current webpage;
encoding a concatenation of at least the user contextual information and a content of the current webpage into a first feature vector, the first feature vector comprising a low-dimensional dense vector;
encode concatenations of user historical information, product information, and content information into a plurality of second feature vectors, the second feature vectors comprising low-dimensional dense vectors;
determine a cosine similarity between a first portion of the first feature vector and a corresponding portion of each of the plurality of second feature vectors, wherein determining the cosine similarity ignores a second portion of the first feature vector corresponding to user contextual information having a null value;
retrieve text associated with a second feature vector having a cosine similarity to the first feature vector that satisfies a predetermined condition;
provide, to a large language model, an augmented prompt generated based at least on the user contextual information and the retrieved text; and
receive, from the large language model and based on the augmented prompt, a summary of a recommended text.
16 . The computer-readable storage medium of claim 15 , wherein the summary comprises a user-selectable link to the recommended text.
17 . The computer-readable storage medium of claim 15 , wherein said retrieve text associated with a second feature vector comprises:
determining the second feature vectors having a cosine similarity to the first feature vector that satisfies a first predetermined relationship with a first predetermined threshold;
determining a first predetermined number of second feature vectors having highest cosine similarities to the first feature vector; or
determining a second predetermined number of second feature vectors having highest cosine similarities to the first feature vector that satisfy a second predetermined relationship with a second predetermined threshold.
18 . The computer-readable storage medium of claim 15 , wherein the augmented prompt comprises:
a query requesting a recommendation for a user based at least on the retrieved text; and
the retrieved text.
19 . The computer-readable storage medium of claim 15 , wherein at least one of the first feature vector or the plurality of second feature vectors is encoded
using a Generative Pre-Trained Transformer (GPT)-based or a Bidirectional Encoder Representations from Transformers (BERT)-based encoder.
20 . The computer-readable storage medium of claim 15 , wherein the user contextual information comprises one or more of:
a user identifier;
user profile information;
user location information;
temporal information;
product or service information;
user intent information; or
user interest information.