Using intent-based rankings to generate large language model responses
The systems and methods disclosed herein generates responses generated by artificial intelligence (AI) models such as large language models (LLM) using intent-based rankings of retrieved information. The systems and methods disclosed herein receives an output generation request for the generation of an output using a set of AI models. Using a first AI model, a set of documents are retrieved using the received output generation request. The set of documents are partitioned into chunks. The chunks are ranked using a distance between the vector representation of the received output generation request and the vector representation of each chunk. A second AI model classifies the output generation request and chunks using an intent of the respective output generation request or chunk, and generates a second set of rankings using the intents. The set of AI models generate a response using the second set of rankings.
1. A computer-implemented method for causing artificial intelligence (AI) models to generate outputs using intent-based rankings of retrieved information, the computer-implemented method comprising:
obtaining (i) an output generation request for generation of an output using one or more AI models and (ii) one or more sets of information from one or more stored computer files,
wherein the one or more sets of information is responsive to the obtained output generation request;
generating, using a first AI model, a first set of rankings of each set of information of the one or more sets of information by associating each particular set of information with one or more indicators of the obtained output generation request;
classifying, with a set of categorical labels, both (i) the obtained output generation request and (ii) each particular set of information of the one or more sets of information, the set of categorical labels indicating attributes of one or more of: (i) an expected output of the obtained output generation request or (ii) an expected output generation request that results in the particular set of information;
generating, using a second AI model, a second set of rankings of each set of information of the one or more sets of information using (i) the classified set of categorical labels of the obtained output generation request and (ii) the classified set of categorical labels of each set of information of the one or more sets of information; and
using the second set of rankings of the one or more sets of information, causing a generation of one or more outputs responsive to the obtained output generation request.
2. The method of claim 1 , wherein the first AI model and the second AI model are the same.
3. The method of claim 1 , wherein the first AI model and the second AI model are different.
4. The method of claim 1 , further comprising:
generating a confidence score for each set of information of the one or more sets of information using the second set of rankings; and
filtering out sets of information having a corresponding confidence score satisfying a predetermined threshold.
5. The method of claim 1 , further comprising:
receiving a set of feedback data indicating an accuracy of a generated one or more outputs caused by the generation of the one or more outputs; and
adjusting the second set of rankings using the received set of feedback data.
6. The method of claim 1 ,
wherein the one or more indicators include vector representations of text content within each set of information, and
wherein generating the first set of rankings comprises:
comparing vector representations of the output generation request with the vector representations of text content, and
ranking the sets of information using the comparison.
7. A system comprising:
at least one hardware processor; and
at least one non-transitory memory storing instructions, which, when executed by the at least one hardware processor, cause the system to:
obtain (i) an output generation request for generation of an output using one or more AI models and (ii) one or more sets of information from one or more computer files,
wherein the one or more sets of information is responsive to the obtained output generation request;
determine, using a first AI model, a first set of rankings of each set of information of the one or more sets of information by associating each particular set of information with one or more indicators of the obtained output generation request;
classify, with a set of categorical labels, one or more of: (i) the obtained output generation request or (ii) each particular set of information of the one or more sets of information, the set of categorical labels indicating attributes of one or more of: (i) an expected output of the obtained output generation request or (ii) an expected output generation request associated with the particular set of information;
determine, using a second AI model, a second set of rankings of each set of information of the one or more sets of information using the one or more of: (i) the classified set of categorical labels of the obtained output generation request or (ii) the classified set of categorical labels of each set of information of the one or more sets of information; and
using the second set of rankings of the one or more sets of information, cause a generation of one or more outputs responsive to the obtained output generation request.
8. The system of claim 7 , wherein the system is further caused to:
generate a confidence score for each set of information of the one or more sets of information using the second set of rankings; and
filter out sets of information having a corresponding confidence score satisfying a predetermined threshold.
9. The system of claim 7 , wherein the classification of the one or more of: (i) the obtained output generation request or (ii) each particular set of information of the one or more sets of information with the set of categorical labels is performed by a third AI model.
10. The system of claim 9 , wherein the first AI model, the second AI model, and the third AI model are the same.
11. The system of claim 9 , wherein the first AI model, the second AI model, and the third AI model are different.
12. The system of claim 7 , wherein, when the second set of rankings are determined at least partly using the classified set of categorical labels of the obtained output generation request, sets of information with categorical labels matching the categorical label of the obtained output generation request are ranked higher than sets of information with categorical labels failing to match the categorical label of the obtained output generation request.
13. The system of claim 7 , wherein the system is further caused to:
partition the obtained sets of information into a set of chunks,
wherein a size of each chunk in the set of chunks is smaller than a size of the obtained sets of information.
14. One or more non-transitory, computer-readable media comprising instructions that, when executed by one or more processors, cause operations comprising:
obtaining (i) an output generation request for generation of an output using a set of AI models and (ii) one or more sets of information responsive to the obtained output generation request;
classifying, with a set of categorical labels, one or more of: (i) the obtained output generation request or (ii) each particular set of information of the one or more sets of information, the set of categorical labels indicating attributes of one or more of: (i) an expected output of the obtained output generation request or (ii) an expected output generation request associated with the particular set of information;
determining, using one or more AI models of the set of AI models, a set of rankings of each set of information of the one or more sets of information using the one or more of: (i) the classified set of categorical labels of the obtained output generation request or (ii) the classified set of categorical labels of each set of information of the one or more sets of information; and
using the set of rankings of the one or more sets of information, causing a generation of a set of outputs responsive to the obtained output generation request.
15. The one or more non-transitory, computer-readable media of claim 14 , wherein the operations further comprise:
comparing a set of vector representations of the output generation request with a set of vector representations of at least one set of information of the sets of information.
16. The one or more non-transitory, computer-readable media of claim 14 , wherein the operations further comprise:
extracting keywords from the sets of information and the output generation request;
generating keyword match scores based on matching keywords of the extracted keywords; and
ranking the sets of information using the keyword match scores.
17. The one or more non-transitory, computer-readable media of claim 14 , wherein the one or more AI models include multiple AI models.
18. The one or more non-transitory, computer-readable media of claim 17 , wherein the multiple AI models classify the obtained output generation request using a majority vote between or among the multiple AI models.
19. The one or more non-transitory, computer-readable media of claim 14 , wherein the one or more AI models classify the obtained output generation request using one or more of: (1) a portion of text within the obtained output generation request or (2) a pre-loaded query context within the obtained output generation request.
20. The one or more non-transitory, computer-readable media of claim 14 , wherein the operations further comprise:
partitioning the obtained sets of information into a set of semantic chunks.