Text reduction and analysis interface to a text generation modeling system
A query request identifying a query and a plurality of text portions for determining an answer to the query may be received. Relevance scores corresponding with respective ones of the text portions may be determined based on application of one or more machine learning models to the respective text portion and the query. A subset of the text portions may be selected based on the relevance scores. A response message to the query request including an answer to the query in natural language text generated by a large language model based on the first subset of text portions may be determined.
1. A method comprising:
receiving a query request via a communication interface, the query request identifying a query and a plurality of text portions for determining an answer to the query;
determining a first plurality of relevance scores via a processor, each of the first plurality of relevance scores corresponding with a respective text portion of the plurality of text portions, each of the first plurality of relevance scores being determined based on application of one or more machine learning models to the respective text portion and the query;
selecting a first subset of the plurality of text portions based on the first plurality of relevance scores;
determining a response message to the query request, the response message including an answer to the query in natural language text generated by a large language model based on the first subset of text portions; and
transmitting the response message via the communication interface.
2. The method recited in claim 1 , wherein the one or more machine learning models include a cross-encoder based on a transformer network.
3. The method recited in claim 1 , wherein the one or more machine learning models include a bi-encoder based on an embedding of text into a vector space.
4. The method recited in claim 1 , the method further comprising:
determining a text class for a designated text portion of the plurality of text portions based on application of a text classification model to the designated text portion, wherein determining the plurality of relevance scores comprises determining a designated relevance score for the designated text portion based on a designated machine learning model selected based on the text class.
5. The method recited in claim 1 , wherein selecting the first subset of the plurality of text portions comprises comparing the plurality of relevance scores against a designated relevance threshold value.
6. The method recited in claim 1 , wherein determining the response message to the query request comprises:
transmitting a plurality of input relevance prompts to a remote text generation system, each of the plurality of input relevance prompts including a respective one or more of the first subset of the plurality of text portions, each of the plurality of input relevance prompts including a query portion determined based on the query; and
receiving from the remote text generation system a plurality of completed relevance prompts, each of the plurality of completed relevance prompts including a respective set of a second respective one or more relevance scores corresponding with the respective one or more of the first subset of the plurality of text portions.
7. The method recited in claim 6 , wherein determining the response message to the query request comprises:
selecting a second subset of the plurality of text portions based on the sets of second respective one or more relevance scores, wherein the second subset of the plurality of text portions is a subset of the first subset of the plurality of text portions.
8. The method recited in claim 1 , wherein determining the response message to the query request comprises:
transmitting an input synthesis prompt to a remote text generation system, the input synthesis prompt including the query and the first subset of the plurality of text portions, the input synthesis prompt further including an instruction to generate an answer to the query based on the first subset of the plurality of text portions; and
receiving from the remote text generation system a completed synthesis prompt, the completed synthesis prompt including an answer to the query in natural language generated based on the first subset of the plurality of text portions.
9. The method recited in claim 1 , the method further comprising:
transmitting an input query decomposition prompt to a remote text generation system, the input query decomposition prompt including the query and an instruction to identify one or more subqueries based on the query; and
receiving from the remote text generation system a completed query decomposition prompt, the completed query decomposition prompt including one or more subqueries identified by the remote text generation system.
10. The method recited in claim 1 , the method further comprising:
transmitting an input training data generation prompt to a remote text generation system, the input training data generation prompt including the query and an instruction to generate training data based on the query; and
receiving from the remote text generation system a completed training data generation prompt, the completed training data generation prompt including a plurality of training data observations generated by the remote text generation system based on the query.
11. The method recited in claim 10 , the method further comprising:
training a designated machine learning model of the one or more machine learning models based on the training data.
12. The method recited in claim 1 , wherein a designated one of the plurality of text portions is a document.
13. One or more non-transitory computer readable media having instructions stored thereon for performing a method, the method comprising:
receiving a query request via a communication interface, the query request identifying a query and a plurality of text portions for determining an answer to the query;
determining a first plurality of relevance scores via a processor, each of the first plurality of relevance scores corresponding with a respective text portion of the plurality of text portions, each of the first plurality of relevance scores being determined based on application of one or more machine learning models to the respective text portion and the query;
selecting a first subset of the plurality of text portions based on the first plurality of relevance scores;
determining a response message to the query request, the response message including an answer to the query in natural language text generated by a large language model based on the first subset of text portions; and
transmitting the response message via the communication interface.
14. The one or more non-transitory computer readable media recited in claim 13 , the method further comprising:
determining a text class for a designated text portion of the plurality of text portions based on application of a text classification model to the designated text portion, wherein determining the plurality of relevance scores comprises determining a designated relevance score for the designated text portion based on a designated machine learning model selected based on the text class.
15. The one or more non-transitory computer readable media recited in claim 13 , wherein selecting the first subset of the plurality of text portions comprises comparing the plurality of relevance scores against a designated relevance threshold value.
16. The one or more non-transitory computer readable media recited in claim 13 , wherein determining the response message to the query request comprises:
transmitting a plurality of input relevance prompts to a remote text generation system, each of the plurality of input relevance prompts including a respective one or more of the first subset of the plurality of text portions, each of the plurality of input relevance prompts including a query portion determined based on the query; and
receiving from the remote text generation system a plurality of completed relevance prompts, each of the plurality of completed relevance prompts including a respective set of a second respective one or more relevance scores corresponding with the respective one or more of the first subset of the plurality of text portions.
17. The one or more non-transitory computer readable media recited in claim 16 , wherein determining the response message to the query request comprises:
selecting a second subset of the plurality of text portions based on the sets of second respective one or more relevance scores, wherein the second subset of the plurality of text portions is a subset of the first subset of the plurality of text portions.
18. The one or more non-transitory computer readable media recited in claim 13 , wherein determining the response message to the query request comprises:
transmitting an input synthesis prompt to a remote text generation system, the input synthesis prompt including the query and the first subset of the plurality of text portions, the input synthesis prompt further including an instruction to generate an answer to the query based on the first subset of the plurality of text portions; and
receiving from the remote text generation system a completed synthesis prompt, the completed synthesis prompt including an answer to the query in natural language generated based on the first subset of the plurality of text portions.
19. The one or more non-transitory computer readable media recited in claim 13 , the method further comprising:
transmitting an input query decomposition prompt to a remote text generation system, the input query decomposition prompt including the query and an instruction to identify one or more subqueries based on the query; and
receiving from the remote text generation system a completed query decomposition prompt, the completed query decomposition prompt including one or more subqueries identified by the remote text generation system.
20. A system including a communication interface and one or more processors, the system configured to perform a method comprising:
receiving a query request via the communication interface, the query request identifying a query and a plurality of text portions for determining an answer to the query;
determining a first plurality of relevance scores via a processor, each of the first plurality of relevance scores corresponding with a respective text portion of the plurality of text portions, each of the first plurality of relevance scores being determined based on application of one or more machine learning models to the respective text portion and the query;
selecting a first subset of the plurality of text portions based on the first plurality of relevance scores;
determining a response message to the query request, the response message including an answer to the query in natural language text generated by a large language model based on the first subset of text portions; and
transmitting the response message via the communication interface.