CN-122021852-A - Knowledge mining method, device, equipment, medium and product based on large model

CN122021852ACN 122021852 ACN122021852 ACN 122021852ACN-122021852-A

Abstract

The embodiment of the disclosure discloses a knowledge mining method, device, equipment, medium and product based on a large model. The method comprises the steps of determining target pre-set requirements corresponding to a target field, inputting the target pre-set requirements into a target large language model to obtain a first triplet, conducting knowledge expansion based on a first head entity of the first triplet to obtain a second triplet, conducting entity filtering processing based on a first relation entity of the first triplet and a second relation of the second triplet to obtain a third triplet, wherein the third relation entity of the third triplet accords with a preset occurrence frequency condition, conducting de-duplication processing on the third triplet to obtain a preset number of target triples corresponding to the target field, conducting parameter updating on the target large language model based on general knowledge of the target triples and the target field, and obtaining an updated large language model.

Inventors

Shi Chufan
YANG YUJIU
Lu nanchang
LIN FANG
CHEN GUANGMING
Dai Tiangong
LIN LI
JIA XIAOGUANG
HUANG ZEBO
MIAO ZIHUI

Assignees

中国移动通信集团广东有限公司
中移湾区(广东)创新研究院有限公司
清华大学深圳国际研究生院
中国移动通信集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. A knowledge mining method based on a large model, comprising: Determining a target pre-requirement corresponding to a target field, and inputting the target pre-requirement into a target large language model to obtain a first triplet; Performing knowledge expansion based on the first head entity of the first triplet to obtain a second triplet; Filtering the first relation entity of the first triplet and the second relation entity of the second triplet to obtain a third triplet, wherein the third relation entity of the third triplet accords with a preset occurrence frequency condition; Performing de-duplication treatment on the third triples to obtain target triples corresponding to a preset number of target fields; and carrying out parameter updating on the target large language model based on the target triples and the general knowledge of the target field to obtain an updated large language model.
2. The method of claim 1, wherein the performing knowledge expansion based on the first head entity of the first triplet to obtain a second triplet comprises: Performing multi-round iterative retrieval on the first head entity through the instruction retrieval model to obtain a second head entity after knowledge expansion; And carrying out structural integrity processing based on the second head entity to obtain the second triplet.
3. The method of claim 1, wherein the filtering of the entity based on the first relationship entity of the first triplet and the second relationship of the second triplet to obtain a third triplet comprises: Determining a relation entity meeting a preset occurrence frequency condition from the first triplet and the second triplet, and determining the relation entity as a third relation entity; Determining a head entity corresponding to the third relation entity, and determining the head entity as a third head entity; And generating a third tail entity which accords with a preset related condition with the third relation entity based on the third head entity and the third relation entity, and further obtaining a third triplet.
4. The method of claim 1, wherein the performing deduplication on the third triplet to obtain a preset number of target triples corresponding to the target field includes: determining a repeated triplet set in the third triplet, and reserving any triplet in the repeated triplet set to obtain a triplet after de-duplication; Under the condition that the number of triples of the triples after the duplication removal is larger than the preset number, optimizing the triples after the duplication removal to obtain a fourth triples with the preset number; And carrying out knowledge retrieval and correction on the fourth triplet through a target language big model based on the general knowledge of the target field to obtain the target triplet.
5. The method of claim 4, wherein the performing knowledge retrieval and correction on the fourth triplet through the target language big model based on the general knowledge of the target domain to obtain the target triplet comprises: Performing knowledge retrieval on the fourth triples through the target language big model based on the general knowledge in the target field to obtain fifth triples corresponding to a first retrieval result and sixth triples corresponding to a second retrieval result, wherein the first retrieval result is used for indicating fourth triples with correct knowledge, and the second retrieval result is used for indicating fourth triples with incorrect knowledge; Correcting the sixth triplet to obtain a seventh triplet; And determining the fifth triplet and the seventh triplet as the target triplet.
6. The method of claim 5, wherein said correcting said sixth triplet to obtain a seventh triplet comprises: Acquiring related knowledge and error knowledge information of the sixth triplet; correcting the error knowledge information based on the related knowledge to obtain a corrected sixth triplet; and determining the corrected sixth triplet as the seventh triplet.
7. A knowledge mining apparatus based on a large model, comprising: the determining module is used for determining target pre-requirements corresponding to the target field and inputting the target pre-requirements into the target large language model to obtain a first triplet; the expansion module is used for carrying out knowledge expansion based on the first head entity of the first triplet to obtain a second triplet; The filtering module is used for filtering the entity based on the first relation entity of the first triplet and the second relation of the second triplet to obtain a third triplet, wherein the third relation entity of the third triplet accords with a preset occurrence frequency condition; the de-duplication module is used for performing de-duplication treatment on the third triples to obtain a preset number of target triples corresponding to the target field; And the updating module is used for carrying out parameter updating on the target large language model based on the target triples and the general knowledge of the target field to obtain an updated large language model.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to carry out the steps of the method according to any one of claims 1 to 6.
9. A computer readable storage medium having stored thereon a computer program/instruction which when executed by a processor performs the steps of the method according to any of claims 1 to 6.
10. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method of any of claims 1 to 6.

Description

Knowledge mining method, device, equipment, medium and product based on large model Technical Field The disclosure relates to the technical field of data processing, in particular to a knowledge mining method, device, equipment, medium and product based on a large model. Background With the rapid development of artificial intelligence, a large language model (Large Language Model, abbreviated as LLM) is widely used in various industries. At present, in the related art, in the aspect of knowledge mining of a large language model, the problem of insufficient knowledge mining in a specific field exists, and the performance of the large model in practical application is affected. Resulting in reduced effectiveness and reliability of large language models in particular areas. Disclosure of Invention The embodiment of the disclosure provides a knowledge mining method, device, equipment, medium and product based on a large model aiming at some defects related in the background technology. In a first aspect, an embodiment of the present disclosure provides a knowledge mining method based on a large model, including: Determining a target pre-requirement corresponding to a target field, and inputting the target pre-requirement into a target large language model to obtain a first triplet; Performing knowledge expansion based on the first head entity of the first triplet to obtain a second triplet; Filtering the first relation entity of the first triplet and the second relation entity of the second triplet to obtain a third triplet, wherein the third relation entity of the third triplet accords with a preset occurrence frequency condition; Performing de-duplication treatment on the third triples to obtain target triples corresponding to a preset number of target fields; and carrying out parameter updating on the target large language model based on the target triples and the general knowledge of the target field to obtain an updated large language model. In an embodiment of the first aspect, the performing knowledge expansion based on the first head entity of the first triplet to obtain a second triplet includes: Performing multi-round iterative retrieval on the first head entity through the instruction retrieval model to obtain a second head entity after knowledge expansion; And carrying out structural integrity processing based on the second head entity to obtain the second triplet. In an embodiment of the first aspect, the filtering processing is performed on the entity based on the first relationship entity of the first triplet and the second relationship of the second triplet, to obtain a third triplet, including: Determining a relation entity meeting a preset occurrence frequency condition from the first triplet and the second triplet, and determining the relation entity as a third relation entity; Determining a head entity corresponding to the third relation entity, and determining the head entity as a third head entity; And generating a third tail entity which accords with a preset related condition with the third relation entity based on the third head entity and the third relation entity, and further obtaining a third triplet. In an embodiment of the first aspect, the performing deduplication processing on the third triples to obtain a preset number of target triples corresponding to the target fields includes: determining a repeated triplet set in the third triplet, and reserving any triplet in the repeated triplet set to obtain a triplet after de-duplication; And under the condition that the number of triples of the triples after the duplication removal is larger than the preset number, optimizing the triples after the duplication removal to obtain a preset number of fourth triples. And carrying out knowledge retrieval and correction on the fourth triplet through the target language big model based on the general knowledge of the target field to obtain the target triplet. In an embodiment of the first aspect, the performing, by the target language big model, knowledge retrieval and correction on the fourth triplet based on the general knowledge of the target domain, to obtain the target triplet includes: Performing knowledge retrieval on the fourth triples through the target language big model based on the general knowledge in the target field to obtain fifth triples corresponding to a first retrieval result and sixth triples corresponding to a second retrieval result, wherein the first retrieval result is used for indicating fourth triples with correct knowledge, and the second retrieval result is used for indicating fourth triples with incorrect knowledge; Correcting the sixth triplet to obtain a seventh triplet; And determining the fifth triplet and the seventh triplet as the target triplet. In an embodiment of the first aspect, the correcting the sixth triplet to obtain a seventh triplet includes: Acquiring related knowledge and error knowledge information of the sixth triplet; correcting the error know