CN-121981300-A - Content conversion method, device, equipment, medium and product

Abstract

The application discloses a content generation method, a content generation device, a content generation equipment, a content generation medium and a content generation product, and belongs to the field of natural language processing. The method is executed by a content conversion system, the content conversion system comprises a supervision model and a conversion model, the supervision model generates ith round of scheduling information based on input source information in the ith round of scheduling process when i is equal to 1, the supervision model generates the ith round of scheduling information based on the i-1 th round of feedback information in the ith round of scheduling process when i is greater than 1, the ith round of scheduling information is used for scheduling the conversion model to execute a first subtask, the conversion model executes the first subtask based on the ith round of scheduling information and generates the ith round of feedback information, and the supervision model acquires the ith round of feedback information to execute a subsequent round of scheduling process. And an automatic closed-loop content conversion task is supported, and the content conversion efficiency is improved.

Inventors

• GUO HUI

Assignees

• 腾讯科技（深圳）有限公司

Dates

Publication Date

20260505

Application Date

20260121

Claims (20)

1. 1. A content conversion method, the method being performed by a content conversion system, the content conversion system comprising a supervision model and a conversion model, the method comprising: when i is equal to 1, the supervision model generates ith round of scheduling information based on input source information in the ith round of scheduling process; when i is greater than 1, the supervision model generates the ith round of scheduling information based on the ith-1 round of feedback information in the ith round of scheduling process; the conversion model executes a first subtask based on the ith round of scheduling information and generates ith round of feedback information, wherein the first subtask is one of at least one subtask corresponding to a content conversion task, and the first subtask indicates the scheduling of the subtask executed by the conversion model for the ith round of scheduling information; the supervision model acquires the ith round of feedback information to execute a scheduling process of a subsequent round; The content conversion task is used for processing the input source information to obtain target information, the source information and the target information are information which comprises the same core information elements and/or core logic association relations but different expression forms, the core information elements are information elements with importance degrees meeting element screening conditions, and the core logic association relations are used for indicating association relations with importance degrees meeting the association relation screening conditions.
2. 2. The method of claim 1, wherein the supervisory model obtains the i-th round of feedback information to perform a scheduling process for a subsequent round, comprising: And generating the (i+1) -th round of scheduling information in the (i+1) -th round of scheduling process under the condition that the (i) -th round of feedback information indicates that a first execution result does not meet a quality screening condition, wherein the (i+1) -th round of scheduling information is used for scheduling the conversion model to re-execute the first subtask, and the first execution result is an execution result obtained by executing the first subtask in the (i) -th round of scheduling process.
3. 3. The method of claim 2, wherein the i+1th round of scheduling information includes at least one of: a prompt word for the first subtask; The first state reasoning information is used for indicating a process of reasoning and obtaining the (i+1) th round of scheduling information based on the (i) th round of feedback information; The first task description information is used for indicating constraint conditions for the first subtasks after being corrected based on the first execution result; A first module, the first module being scheduled to perform the first subtask, the first module being one of at least one module comprised by the conversion model.
4. 4. A method according to any one of claims 1 to 3, wherein the at least one sub-task is a sub-task that is executed in sequence, wherein in the at least one sub-task, a third sub-task is executed in sequence after a second sub-task, the method further comprising: And generating j+1th round of scheduling information in a j+1th round of scheduling process under the condition that the j-th round of feedback information indicates that a second execution result meets a quality screening condition, wherein the j+1th round of scheduling information is used for scheduling the conversion model to execute the third subtask, and the second execution result is an execution result obtained by executing the second subtask in the j-th round of scheduling process, and j is a positive integer.
5. 5. The method of claim 4, wherein the j+1th round of scheduling information comprises at least one of: a prompt for the third subtask; the second state reasoning information is used for indicating a process of reasoning and obtaining the j+1th round of scheduling information based on the j th round of feedback information; the second task description information is provided with a second task description information, the second task description information is used for indicating constraint conditions for the third subtask; And a second module, the second module being scheduled to perform the third subtask, the second module being one of at least one module included in the conversion model.
6. 6. The method according to any one of claims 2 to 4, wherein the quality screening conditions comprise at least one of the following conditions: The evaluation score of the execution result is higher than the quality score threshold; the execution duration of the subtasks is smaller than the execution time threshold; The execution times of the subtasks are smaller than the execution times threshold.
7. 7. The method according to any one of claims 1 to 6, wherein the feedback information generated by the conversion model comprises at least one of the following information: executing a subtask; The execution state of the subtasks; A correction mode for the subtasks; The cause is corrected.
8. 8. The method according to any one of claims 1 to 7, wherein the subtasks corresponding to the content conversion task include at least one of: Planning a subtask; retrieving the subtasks; The architecture generates subtasks; Generating a subtask by using the target information; Evaluating the subtasks; and correcting the subtasks.
9. 9. The method of claim 8, wherein the content transformation task comprises the planning sub-task; The method further comprises the steps of: the supervision model determines planning information aiming at the source information based on the input source information; Wherein the planning information includes at least one of the following information: the execution sequence of the at least one subtask; the output format of each of the at least one subtask; quality judgment criteria for each of the at least one subtask; and priorities corresponding to the dimensions in the quality judgment standard.
10. 10. The method of claim 8, wherein the content transformation task comprises the retrieval subtask; The method further comprises the steps of: The conversion model is based on input source information, at least one reference case is retrieved from a database, and the association degree between the at least one reference case and the input source information meets a reference condition; Each reference case in the at least one reference case comprises at least one of reference source information, reference target information, reference core information elements, reference logic association relations and reference architecture information.
11. 11. The method of claim 10, wherein the transformation model retrieves at least one reference case from a database based on the input source information, comprising: The conversion model extracts key information elements corresponding to the source information based on the input source information; The conversion model determines a search keyword based on the key information element, wherein the search keyword is an information element of which the key information element is most matched with a conversion rule in at least one category, and the conversion rule is a conversion direction aiming at the source information and the target information; the conversion model is based on the search keywords, and the at least one reference case is obtained from the database in a search mode.
12. 12. The method of claim 8, wherein the content transformation task comprises the architecture generation subtask; The method further comprises the steps of: The conversion model extracts architecture information meeting conversion rules from the input source information based on first input information, wherein the architecture information is information obtained by summarizing and reorganizing the input source information based on the expression form of the target information, and the first input information comprises at least one of at least one reference case, input core information elements and input core logic association relations.
13. 13. The method of claim 8, wherein the content transformation task comprises the target information generation subtask, an evaluation subtask; The method further comprises the steps of: The conversion model sequentially converts at least one information subunit based on the architecture information and/or the input source information, wherein the at least one information subunit is a subunit of the target information; the conversion model determines the evaluation score of a first information subunit under the condition that the first information subunit is obtained through conversion; The conversion model continues to execute conversion aiming at a second information subunit under the condition that the evaluation score of the first information subunit meets the quality screening condition, and the execution sequence of the information generation subtasks corresponding to the second information subunit is behind the information generation subtasks corresponding to the first information subunit; the conversion model re-executes an information generation subtask for the first information subunit if the evaluation score of the first information subunit does not satisfy the quality screening condition.
14. 14. The method of any one of claims 1 to 13, wherein the content conversion system further comprises a data module; the supervision model and the conversion model are used for completing the exchange of feedback information and/or scheduling information based on the data module; And/or the supervision model and the conversion model are used for storing and exchanging all or part of information in the execution process of the at least one subtask based on the data module.
15. 15. A content conversion method, wherein the method is performed by a content conversion system, the content conversion system comprising a scheduling module and an execution module, the method comprising: the scheduling module generates scheduling information based on feedback information or input source information, wherein the scheduling information is used for scheduling the execution module to execute sub-tasks, and the sub-tasks are at least one sub-task corresponding to the content conversion task; The execution module executes the subtasks based on the scheduling information and generates feedback information, wherein the feedback information is used for indicating the execution condition of the subtasks; The scheduling module acquires the feedback information and judges whether the execution quality of the subtasks meets quality screening conditions or not; When the execution quality does not meet the quality screening condition, the scheduling module generates corrected scheduling information based on the feedback information, wherein the corrected scheduling information is used for scheduling the execution module to execute the subtasks again or execute corrected subtasks; under the condition that the execution quality meets the quality screening condition, the scheduling module generates subsequent scheduling information, wherein the subsequent scheduling information is used for scheduling the execution module to execute the next subtask; The content conversion task is configured to process the input source information to obtain target information, where the source information and the target information are information that includes the same core information element and/or core logic association relationship, but have different expression forms, and the core information element is an information element in the source information, where the importance degree of the information element meets an element screening condition, and the core logic association relationship is used to indicate an association relationship in the source information, where the importance degree of the information element meets an association relationship screening condition.
16. 16. A content conversion apparatus, characterized in that the apparatus comprises: the supervision module is used for generating the ith round of scheduling information based on the input source information in the ith round of scheduling process under the condition that i is equal to 1; The supervision module is used for generating the ith round of scheduling information based on the ith-1 round of feedback information in the ith round of scheduling process under the condition that i is larger than 1; The conversion module is configured to execute the first subtask based on the i-th round of scheduling information, and generate i-th round of feedback information, where the first subtask is one of at least one subtask corresponding to a content conversion task, and the i-th round of scheduling information indicates to schedule the subtask executed by the conversion module; The supervision module is used for acquiring the ith round of feedback information so as to execute a scheduling process of a subsequent round; The content conversion task is used for processing the input source information to obtain target information, the source information and the target information are information which comprises the same core information elements and/or core logic association relations but different expression forms, the core information elements are information elements, the importance degree of the information elements meets element screening conditions, in the source information, and the core logic association relations are used for indicating association relations, in the source information, the importance degree of the information elements meets the association relation screening conditions.
17. 17. A content conversion apparatus, characterized in that the apparatus comprises: The scheduling module is used for generating scheduling information based on the feedback information or the input source information, wherein the scheduling information is used for scheduling the execution module to execute the subtasks, and the subtasks are one of at least one subtask corresponding to the content conversion task; the execution module is used for executing the subtasks based on the scheduling information and generating feedback information, wherein the feedback information is used for indicating the execution condition of the subtasks; the scheduling module is used for acquiring the feedback information and judging whether the execution quality of the subtasks meets the quality screening condition or not; The scheduling module is used for generating corrected scheduling information based on the feedback information when the execution quality does not meet the quality screening condition, wherein the corrected scheduling information is used for scheduling the execution module to execute the subtasks again or execute corrected subtasks; The scheduling module is used for generating subsequent scheduling information when the execution quality meets the quality screening condition, and the subsequent scheduling information is used for scheduling the execution module to execute the next subtask; The content conversion task is configured to process the input source information to obtain target information, where the source information and the target information are information that includes the same core information element and/or core logic association relationship, but have different expression forms, and the core information element is an information element in the source information, where the importance degree of the information element meets an element screening condition, and the core logic association relationship is used to indicate an association relationship in the source information, where the importance degree of the information element meets an association relationship screening condition.
18. 18. A computer device comprising a processor and a memory, wherein the memory has at least one program stored therein, and wherein the processor is configured to execute the at least one program in the memory to implement the content conversion method of any of claims 1 to 15.
19. 19. A computer readable storage medium having stored therein computer instructions that are loaded and executed by a processor to implement the content conversion method of any one of the preceding claims 1 to 15.
20. 20. A computer program product comprising computer instructions stored in a computer readable storage medium, the computer instructions being read from the computer readable storage medium and executed by a processor to implement the content conversion method of any one of claims 1 to 15.

Description

Content conversion method, device, equipment, medium and product Technical Field The present application relates to the field of natural language processing, and in particular, to a method, apparatus, device, medium, and product for content conversion. Background With the development of the digital culture industry, the adaptation of novels and the like into different media forms such as drama, film and television, cartoon and the like is an important path for the visual conversion and presentation of works. A great deal of conversion needs are induced in the environment with increasingly high requirements on the frequency and quality of consumption of visual form materials. In the related art, although solutions for converting novels into film and television scripts with assistance of large models exist, these solutions often require manual assistance, such as manual assistance in determining when an abnormality occurs. The conversion efficiency is low and the conversion quality is poor. Disclosure of Invention The application provides a content conversion method, a device, equipment, a medium and a product, wherein the technical scheme is as follows: according to an aspect of the present application, there is provided a content conversion method performed by a content conversion system including a supervision model and a conversion model, the method comprising: when i is equal to 1, the supervision model generates ith round of scheduling information based on input source information in the ith round of scheduling process; when i is greater than 1, the supervision model generates the ith round of scheduling information based on the ith-1 round of feedback information in the ith round of scheduling process; The conversion model executes the first subtask based on the ith round of scheduling information, and generates ith round of feedback information, wherein the first subtask is one of at least one subtask corresponding to a content conversion task, and the first subtask is the subtask which the ith round of scheduling information indicates to schedule the conversion model to execute; the supervision model acquires the ith round of feedback information to execute a scheduling process of a subsequent round; The content conversion task is used for processing the input source information to obtain target information, the source information and the target information are information which comprises the same core information elements and/or core logic association relations but different expression forms, the core information elements are information elements with importance degrees meeting element screening conditions, and the core logic association relations are used for indicating association relations with importance degrees meeting the association relation screening conditions. According to an aspect of the present application, there is provided a content conversion method performed by a content conversion system including a scheduling module and an execution module, the method including: the scheduling module generates scheduling information based on feedback information or input source information, wherein the scheduling information is used for scheduling the execution module to execute sub-tasks, and the sub-tasks are at least one sub-task corresponding to the content conversion task; The execution module executes the subtasks based on the scheduling information and generates feedback information, wherein the feedback information is used for indicating the execution condition of the subtasks; The scheduling module acquires the feedback information and judges whether the execution quality of the subtasks meets quality screening conditions or not; When the execution quality does not meet the quality screening condition, the scheduling module generates corrected scheduling information based on the feedback information, wherein the corrected scheduling information is used for scheduling the execution module to execute the subtasks again or execute corrected subtasks; under the condition that the execution quality meets the quality screening condition, the scheduling module generates subsequent scheduling information, wherein the subsequent scheduling information is used for scheduling the execution module to execute the next subtask; The content conversion task is configured to process the input source information to obtain target information, where the source information and the target information are information that includes the same core information element and/or core logic association relationship, but have different expression forms, and the core information element is an information element in the source information, where the importance degree of the information element meets an element screening condition, and the core logic association relationship is used to indicate an association relationship in the source information, where the importance degree of the information element meets an association