A reranking system, method, and program for extracting passages that are highly relevant to a query.

Abstract

A reranking system, method, and program for extracting passages with a high degree of relevance to a query are disclosed. The system includes at least one processor and memory for storing one or more instructions, wherein the at least one processor performs multiple tournament sortings to rerank multiple passages by executing the one or more instructions stored in the memory, and each tournament sorting includes the steps of: pairing a specific number of first passages contained in the passage into multiple groups, evaluating the degree of relevance between the first passages contained in each group and the query, extracting passages from the first passages that have a high degree of relevance up to a predetermined rank and outputting them as second passages; and pairing a specific number of second passages into multiple groups, evaluating the degree of relevance between the second passages contained in each group and the query, extracting passages from the second passages that have a high degree of relevance up to a predetermined rank and outputting them as third passages.

Claims

1. A system comprising at least one processor and one or more memory for storing instructions, The at least one processor performs multiple tournament sorting and reranks multiple passages by executing one or more instructions stored in the memory. The aforementioned tournament lineup is, The first passages included in the aforementioned passage are paired in specific numbers to divide them into multiple groups, the degree of relevance between the first passages included in each group and the query is evaluated, and passages with a high degree of relevance up to a predetermined rank are extracted from the first passages and output as a second passage. The process includes the steps of: pairing the second passages in specific numbers to divide them into multiple groups; evaluating the degree of relevance between the second passages included in each group and the query; and extracting passages from the second passages that have a high degree of relevance up to a predetermined rank and outputting them as a third passage. system.

2. The first passage, which is identical to the third passage output in at least one of the tournament arrangements, is not used as the object for evaluating the degree of relevance in any one of the subsequent tournament arrangements. The system according to claim 1.

3. The results of at least one correlation assessment performed in at least one of the tournament arrangements are reused in the subsequent step of evaluating correlation in any one of the tournament arrangements. The system according to claim 1.

4. The step of evaluating the degree of relevance generates outputs in order from the lowest degree of relevance to the highest degree of relevance. The system according to claim 1.

5. The aforementioned degree of relevance is evaluated by the encoder-decoder structure. The encoder-decoder structure is an FiD or T5-base architecture. The system according to claim 1.

6. The aforementioned query or passage is a vector representation of data stored in the database. The system according to claim 1.

7. The aforementioned specific number is 5. The system according to claim 1.

8. The aforementioned predetermined ranking is either the top 1st place or the top 2nd place. The system according to claim 1.

9. A method for performing reranking of multiple passages by performing multiple tournament alignments, which are executed by at least one processor, The first passages included in the aforementioned passage are paired in specific numbers to divide them into multiple groups, the degree of relevance between the first passages included in each group and the query is evaluated, and passages with a high degree of relevance up to a predetermined rank are extracted from the first passages and output as a second passage. The process includes the steps of: pairing the second passages in specific numbers to divide them into multiple groups; evaluating the degree of relevance between the second passages included in each group and the query; and extracting passages from the second passages that have a high degree of relevance up to a predetermined rank and outputting them as a third passage. method.

10. The first passage, which is identical to the third passage output in at least one of the tournament arrangements, is not used as the object for evaluating the degree of relevance in any one of the subsequent tournament arrangements. The method according to claim 9.

11. The results of at least one correlation assessment performed in at least one of the tournament arrangements are reused in the subsequent step of evaluating correlation in any one of the tournament arrangements. The method according to claim 9.

12. The step of evaluating the degree of relevance generates outputs in order from the lowest degree of relevance to the highest degree of relevance. The method according to claim 9.

13. The aforementioned degree of relevance is evaluated by the encoder-decoder structure. The encoder-decoder structure is an FiD or T5-base architecture. The method according to claim 9.

14. The aforementioned query or passage is a vector representation of data stored in the database. The method according to claim 9.

15. The aforementioned specific number is 5. The method according to claim 9.

16. The aforementioned predetermined ranking is either the top 1st place or the top 2nd place. The method according to claim 9.

17. A computer-readable recording medium is used to cause a computer to perform the method described in any one of claims 9 to 16, program.

Images