Search

CN-121980025-A - Method and device for generating abstract by combining large model technology

CN121980025ACN 121980025 ACN121980025 ACN 121980025ACN-121980025-A

Abstract

The method comprises the steps of determining a plurality of abstract generation large models, wherein the abstract generation large models are used for generating original abstracts of technical documents in coal industry, selecting standby abstract generation large models corresponding to the original abstracts generated by the abstract generation large models when scores are larger than a set threshold value, acquiring question-answer pairs of the technical documents, adjusting the standby abstract generation large models through the question-answer pairs to obtain target abstract generation large models, acquiring standby abstracts corresponding to the target abstract generation large models when the technical documents in coal industry are processed, and adjusting the standby abstracts based on knowledge maps corresponding to the technical documents to obtain the target abstract. Therefore, the method enables the abstract to be continuously optimized by adjusting the large model generated by the query and the answer on the standby abstract and adjusting the knowledge graph on the standby abstract, is close to the vertical field of coal exploitation, and remarkably improves the quality and efficiency of the abstract of the scientific literature.

Inventors

  • LV YIMENG

Assignees

  • 中煤科工开采研究院有限公司

Dates

Publication Date
20260505
Application Date
20251230

Claims (10)

  1. 1. A summary generation method combined with a large model technology, the method comprising: Determining a plurality of abstract generation large models, wherein the abstract generation large models are used for generating original abstracts of scientific and technological documents in the coal industry; Selecting a corresponding standby abstract generation large model when scores in original abstracts generated by the plurality of abstract generation large models are larger than a set threshold value; obtaining question-answer pairs of the scientific literature, and adjusting the standby abstract generation large model through the question-answer pairs to obtain a target abstract generation large model; Obtaining a standby abstract corresponding to the target abstract generation large model when processing technical literature of the coal industry; And adjusting the standby abstract based on the knowledge graph corresponding to the scientific literature so as to obtain a target abstract.
  2. 2. The method according to claim 1, wherein selecting the backup summary-generation large model corresponding to the original summary generated by the plurality of summary-generation large models when the score is greater than the set threshold value comprises: Obtaining a first score of an original abstract generated by each abstract generating large model, wherein the first score is obtained by calculating a text quality evaluation index between the original abstract and a standby abstract; Obtaining a second score of the original abstract generated by each abstract generating large model, wherein the second score is obtained by calculating a preset user scoring mechanism, the browsing number of the question-answer pairs, the number of the adoption and the discussion number; obtaining a scoring average value and a weight of each abstract generation large model, wherein the scoring average value is the average value of scores of a plurality of original abstracts generated by each abstract generation large model; Calculating the score of the original abstract generated by each abstract generation large model based on the first score, the second score, the liveness parameter of the question-answer pair corresponding to the knowledge service platform, the score average value and the weight; And selecting a corresponding standby abstract to generate a large model when the score of the original abstract is larger than a set threshold value.
  3. 3. The method according to claim 2, wherein the liveness parameter of the question-answer pair corresponding to the knowledge service platform is calculated by month liveness of the knowledge service platform, browse count, adoption count and discussion count of the question-answer pair.
  4. 4. The method of claim 2, wherein the obtaining the question-answer pair of the scientific literature comprises: Acquiring a plurality of initial question-answer pairs corresponding to the scientific literature from the knowledge service platform; Determining the selection probability of each initial question-answer pair according to the browsing number, the adoption number and the discussion number of each initial question-answer pair; And selecting question-answer pairs with probability larger than a set probability threshold as question-answer pairs of scientific and technical literature.
  5. 5. The method of claim 1, wherein the adjusting the backup abstract based on the knowledge-graph corresponding to the scientific literature to obtain the target abstract comprises: Acquiring a knowledge graph corresponding to the scientific literature; Determining the text of the coal industry in the standby abstract; and combing and adjusting the text of the coal industry by taking the knowledge graph as a vector to obtain a target abstract.
  6. 6. A summary generation apparatus incorporating large model technology, the apparatus comprising: The determining module is used for determining a plurality of abstract generation large models, wherein the abstract generation large models are used for generating original abstracts of scientific and technological documents in the coal industry; the selecting module is used for selecting a standby abstract generation large model corresponding to the original abstract generated by the plurality of abstract generation large models when the score is larger than a set threshold value; The first adjusting module is used for acquiring question-answer pairs of the scientific literature and adjusting the standby abstract generation large model through the question-answer pairs to obtain a target abstract generation large model; The acquisition module is used for acquiring a standby abstract corresponding to the target abstract generated large model when the technical literature of the coal industry is processed; and the second adjusting module is used for adjusting the standby abstract based on the knowledge graph corresponding to the scientific literature so as to obtain a target abstract.
  7. 7. The apparatus of claim 6, wherein the selecting module is specifically configured to: Obtaining a first score of an original abstract generated by each abstract generating large model, wherein the first score is obtained by calculating a text quality evaluation index between the original abstract and a standby abstract; Obtaining a second score of the original abstract generated by each abstract generating large model, wherein the second score is obtained by calculating a preset user scoring mechanism, the browsing number of the question-answer pairs, the number of the adoption and the discussion number; obtaining a scoring average value and a weight of each abstract generation large model, wherein the scoring average value is the average value of scores of a plurality of original abstracts generated by each abstract generation large model; Calculating the score of the original abstract generated by each abstract generation large model based on the first score, the second score, the liveness parameter of the question-answer pair corresponding to the knowledge service platform, the score average value and the weight; And selecting a corresponding standby abstract to generate a large model when the score of the original abstract is larger than a set threshold value.
  8. 8. The apparatus of claim 7, wherein the liveness parameter of the question-answer pair corresponding to the knowledge service platform is calculated by a month liveness number of the knowledge service platform, a browse number, an adoption number, and a discussion number of the question-answer pair.
  9. 9. The apparatus of claim 7, wherein the first adjustment module is specifically configured to: Acquiring a plurality of initial question-answer pairs corresponding to the scientific literature from the knowledge service platform; Determining the selection probability of each initial question-answer pair according to the browsing number, the adoption number and the discussion number of each initial question-answer pair; And selecting question-answer pairs with probability larger than a set probability threshold as question-answer pairs of scientific and technical literature.
  10. 10. The apparatus of claim 6, wherein the second adjustment module is specifically configured to: Acquiring a knowledge graph corresponding to the scientific literature; Determining the text of the coal industry in the standby abstract; and combing and adjusting the text of the coal industry by taking the knowledge graph as a vector to obtain a target abstract.

Description

Method and device for generating abstract by combining large model technology Technical Field The disclosure relates to the technical field of abstract generation, and in particular relates to an abstract generation method and device combined with a large model technology. Background In the digital era, the automatic summary generation technology of the knowledge base is a mature natural language processing technology, the traditional automatic summary generation technology is divided into two types of extraction type and generation type text summary, core sentences and phrases in texts are identified, and summary texts are automatically generated, but in the coal exploitation field, the traditional automatic summary technology processing technology reports that generated summary logicality is extremely poor, professional vocabularies and logics of scientific and technical documents cannot be deeply mined, so that the summary redundancy is high and errors are high. Disclosure of Invention The present disclosure provides a summary generation method and apparatus that incorporates large model techniques to address, at least to some extent, one of the technical problems in the related art. The technical scheme of the present disclosure is as follows: According to a first aspect of the embodiment of the disclosure, a summary generation method combining a large model technology is provided, and the method comprises the steps of determining a plurality of summary generation large models, wherein the summary generation large models are used for generating original summaries of technical documents in coal industry, selecting standby summary generation large models corresponding to the raw summaries generated by the plurality of summary generation large models when scores in the original summaries are larger than a set threshold value, acquiring question-answer pairs of the technical documents, adjusting the standby summary generation large models through the question-answer pairs to obtain a target summary generation large model, acquiring standby summaries corresponding to the target summary generation large models when the technical documents in coal industry are processed, and adjusting the standby summaries based on knowledge maps corresponding to the technical documents to obtain the target summaries. According to a second aspect of the embodiment of the disclosure, a summary generation device combining a large model technology is provided, and the device comprises a determination module, a selection module, a first adjustment module, an acquisition module and a second adjustment module, wherein the determination module is used for determining a plurality of summary generation large models, the summary generation large models are used for generating original summaries of technical documents in coal industry, the selection module is used for selecting a standby summary generation large model corresponding to the original summaries generated by the plurality of summary generation large models when scores of the original summaries are larger than a set threshold value, the first adjustment module is used for acquiring question-answer pairs of the technical documents and adjusting the standby summary generation large models through the question-answer pairs to obtain a target summary generation large model, the acquisition module is used for acquiring the standby summaries corresponding to the target summary generation large models when the technical documents in coal industry are processed, and the second adjustment module is used for adjusting the standby summaries based on knowledge maps corresponding to the technical documents to obtain the target summaries. According to a third aspect of embodiments of the present disclosure, there is provided an electronic device comprising a processor, a memory for storing instructions executable by the processor, wherein the processor is configured to execute the instructions to implement a summary generation method in combination with a large model technique according to an embodiment of the first aspect of the present disclosure. According to a fourth aspect of embodiments of the present disclosure, there is provided a computer readable storage medium, which when executed by a processor of an electronic device, enables the electronic device to perform the summary generation method in combination with the large model technique as described in the embodiments of the first aspect of the present disclosure. The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects: According to the technical scheme, a plurality of abstract generation big models are determined and used for generating original abstracts of technical documents in coal industry, standby abstract generation big models corresponding to the original abstracts generated by the plurality of abstract generation big models when scores are larger than a set thresh