PERSONALIZED RETRIEVAL-AUGMENTED GENERATION SYSTEM
The present disclosure relates to systems, non-transitory computer-readable media, and methods for generating personal responses through retrieval-augmented generation. In particular, the disclosed systems can generate a query embedding from a query generated by an entity and determine data context specific to the entity by comparing the query embedding with a plurality of vectorized segments of content items associated with the entity. The disclosed systems can provide the data context to a large language model and generate a personalized response informed by the data context. Subsequently, the disclosed systems can provide the personalized response for display on a client device associated with the entity.
1 . A computer-implemented method comprising:
determining, for a query received from a client device associated with an entity of a content management system, a data context corresponding to the query and utilized in generative functions by a personalized retrieval-augmented generation model;
generating an augmented query by augmenting the query based on the entity;
generating a personalized response specific to the entity by providing the augmented query and a modified data context corresponding to the augmented query to the personalized retrieval-augmented generation model; and
providing the personalized response for display on the client device associated with the entity.
2 . The computer-implemented method of claim 1 , wherein generating the augmented query comprises modifying a query structure or query language based on the entity.
3 . The computer-implemented method of claim 1 , further comprising:
determining a relationship between the query and a content item associated with the entity;
generating the augmented query by adding metadata reflecting the relationship of the content item to the query; and
generating the personalized response based utilizing the personalized retrieval-augmented generation model to process the augmented query reflecting the relationship.
4 . The computer-implemented method of claim 1 further comprising:
generating the augmented query by processing the query with an augmentation layer;
receiving feedback data indicating a relevance of the personalized response; and
modifying the augmentation layer based on the feedback data.
5 . The computer-implemented method of claim 1 , further comprising:
detecting a modification to a component of the personalized retrieval-augmented generation model; and
based on the modification to the component, modifying an augmentation layer.
6 . The computer-implemented method of claim 1 , further comprising:
determining a query type for the query;
selecting an embedding model that corresponds to the query type;
generating the augmented query based on the query type;
generating an augmented query embedding by processing the augmented query with the embedding model; and
generating the personalized response using the personalized retrieval-augmented generation model to retrieve digital content corresponding to the augmented query embedding.
7 . The computer-implemented method of claim 1 , further comprising:
detecting a change to a characteristic associated with the entity;
based on the change to the characteristic associated with the entity, modifying an augmentation layer to align with the entity; and
generating the augmented query using the augmentation layer modified to align with the entity.
8 . A system comprising:
at least one processor; and
a non-transitory computer readable medium comprising instructions that, when executed by the at least one processor, cause the system to:
determine, for a query received from a client device associated with an entity of a content management system, a data context corresponding to the query and utilized in generative functions by a personalized retrieval-augmented generation model;
generate, utilizing an augmentation layer, an augmented query by augmenting a structure of the query based on the entity;
generate a personalized response specific to the entity by providing the augmented query and a modified data context corresponding to the augmented query to the personalized retrieval-augmented generation model; and
provide the personalized response for display on the client device associated with the entity.
9 . The system of claim 8 , wherein generating the augmented query comprises one or more of restructuring, translating, or expanding language of the query.
10 . The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:
access metadata reflecting relationships, locations, or timing of a plurality of content items associated with the entity; and
generate the augmented query by adding the metadata to the query.
11 . The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:
receive explicit feedback data indicating approval or disapproval of the personalized response by receiving an indication of a selection of a selectable element indicating helpfulness of the personalized response; and
modify the augmentation layer based on the explicit feedback data.
12 . The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:
detect a modification to an embedding model of the personalized retrieval-augmented generation model; and
based on the modification to the embedding model, modify the augmentation layer to generate the augmented query tailored to the embedding model.
13 . The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:
determine a query type for the query;
select a large language model as part of the personalized retrieval-augmented generation model based on the query type;
generate the augmented query with the structure tailored to the large language model; and
generate the personalized response to the augmented query utilizing the large language model.
14 . The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:
access one or more features associated with the entity;
determine a learning rate based on the one or more features of the entity; and
modify the personalized retrieval-augmented generation model according to the learning rate specific to the entity.
15 . A non-transitory computer readable medium comprising instructions that, when executed by at least one processor, cause the at least one processor to:
determine, for a query received from a client device associated with an entity of a content management system, a data context corresponding to the query and utilized in generative functions by a personalized retrieval-augmented generation model;
generate an augmented query by augmenting the query based on the entity;
generate a personalized response specific to the entity by providing the augmented query and a modified data context corresponding to the augmented query to the personalized retrieval-augmented generation model;
receive feedback data associated with the personalized response; and
modify an augmentation layer specific to the entity based on the feedback data.
16 . The non-transitory computer readable medium of claim 15 , wherein generating the augmented query comprises generating a first subquery and a second subquery from the query.
17 . The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:
access, from a metadata database, metadata of a plurality of content items associated with the entity;
generate the augmented query by adding the metadata to the query; and
utilize the metadata to select one or more vectorized segments to include in the modified data context.
18 . The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:
receive implicit feedback data indicating approval or disapproval of the personalized response by monitoring one or more user interactions with the personalized response; and
modify the augmentation layer based on the implicit feedback data.
19 . The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:
modify the augmentation layer by fine-tuning the augmentation layer based on explicit feedback data.
20 . The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:
detect a modification to at least one of a large language model or an embedding model of the personalized retrieval-augmented generation model; and
based on the modification to the at least one of the large language model or the embedding model, modify the augmentation layer to generate an augmented query tailored to the at least one of the large language model or the embedding model.