Processing method and device for optimizing module generation content, medium and product
By constructing semantic graphs and mapping the deviation of generated content, the artificial intelligence model is dynamically optimized to solve the problem of low accuracy of generated content and achieve the accuracy and continuous optimization of generated content.
Patent Information
- Application Number
- CN202511045014.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-10-21
AI Technical Summary
Existing AI models lack a dynamic feedback mechanism when generating content, resulting in reduced accuracy of generated content.
By obtaining the task features of the input task, building a semantic graph, and mapping the generated content and the semantic graph to the set semantic space, the target deviation is calculated, and the artificial intelligence model is adjusted based on the deviation to achieve dynamic optimization.
It improves the accuracy of content generated by AI models, ensures that the generated content meets the intent of the input task through detailed feature capture and semantic alignment, and provides structured deviation feedback to facilitate model optimization.
Smart Images
Figure CN120822639A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present disclosure relate to the technical field of artificial intelligence, and more specifically, to a processing method, device, medium, and product for optimizing model-generated content. Background Art
[0002] With the rapid development of generative artificial intelligence (AIGC) technology, it has been widely used in various content productions such as text-to-image, text-to-video, and speech synthesis.
[0003] At present, existing artificial intelligence models can generate content based on tasks and perform a one-time score on the generated content through an evaluation model. The lack of a dynamic feedback mechanism reduces the accuracy of content generated by artificial intelligence models. Summary of the Invention
[0004] One purpose of the embodiments of the present disclosure is to provide a new technical solution for optimizing the processing of model-generated content.
[0005] According to a first aspect of the present disclosure, a processing method for optimizing model-generated content is provided, the method comprising:
[0006] Obtaining an input task of an artificial intelligence model and generated content generated by the artificial intelligence model based on the input task;
[0007] Extracting task features of the input task and constructing a semantic graph of the input task;
[0008] Mapping the semantic graph and the generated content to a set semantic space to obtain a target deviation between the generated content and the semantic graph;
[0009] Adjust the artificial intelligence model based on the target optimization matters associated with the target deviation amount.
[0010] Optionally, extracting the task features of the input task and constructing a semantic graph of the input task includes:
[0011] Performing deep semantic parsing on the input task to obtain task features reflecting task elements of the input task;
[0012] The task features are converted into graph composition units to construct a semantic graph of the input task.
[0013] Optionally, before mapping the semantic graph and the generated content to a set semantic space and obtaining a target deviation between the generated content and the semantic graph, the method further includes:
[0014] Acquiring content features and content semantics of the generated content; wherein the content features are image features and / or voice features;
[0015] Determining, based on the content features, first feature quantities corresponding to the content emotional features, temporal structure features, and logical structure features reflected by the generated content;
[0016] Comparing the first feature with the content semantics to obtain a semantic comparison feature;
[0017] The semantic comparison feature is updated to the generated content to obtain the updated generated content.
[0018] Optionally, mapping the semantic graph and the generated content to a set semantic space to obtain a target deviation between the generated content and the semantic graph includes:
[0019] Mapping the content features of the generated content, the semantic comparison features, and the semantic features of the semantic graph to a set semantic space to obtain a first embedding vector corresponding to the generated content and a second embedding vector corresponding to the semantic graph;
[0020] The first embedding vector and the second embedding vector are input into a preset vector distance algorithm to obtain an offset distance as a target deviation between the generated content and the semantic graph.
[0021] Optionally, adjusting the artificial intelligence model based on the target optimization item associated with the target deviation includes:
[0022] determining whether the deviation value reflected by the target deviation amount is less than or equal to a set threshold;
[0023] When the deviation value reflected by the target deviation is less than or equal to a set threshold, determining a target deviation item corresponding to the target deviation according to a preset first mapping relationship; wherein the first mapping relationship reflects that different deviation items correspond to different deviation items;
[0024] According to a preset second mapping relationship, the target optimization item corresponding to the target deviation item is determined; wherein the second mapping relationship reflects that different deviation items correspond to different optimization items.
[0025] Optionally, the method further includes:
[0026] Generate a deviation optimization report based on the target deviation item and the target optimization item;
[0027] Feedback the deviation optimization report to the associated object of the artificial intelligence model.
[0028] According to a second aspect of the present disclosure, there is also provided an optimization processing device for an artificial intelligence model, the device comprising:
[0029] An acquisition module, configured to acquire an input task of an artificial intelligence model and generated content generated by the artificial intelligence model based on the input task;
[0030] An extraction module, configured to extract task features of the input task and construct a semantic graph of the input task;
[0031] an obtaining module, configured to map the semantic graph and the generated content to a set semantic space, and obtain a target deviation between the generated content and the semantic graph;
[0032] An adjustment module is used to adjust the artificial intelligence model based on the target optimization items associated with the target deviation amount.
[0033] According to the third aspect of the present disclosure, an optimization processing device for an artificial intelligence model is also provided, comprising a memory and a processor, wherein the memory is used to store a computer program; the processor is used to execute the computer program to implement the method described in the first aspect of the present disclosure.
[0034] According to a fourth aspect of the present disclosure, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the method according to the first aspect of the present disclosure is implemented.
[0035] According to a fifth aspect of the present disclosure, a computer program product is further provided, comprising a computer program, wherein when the computer program is executed by a processor, the method according to the first aspect of the present disclosure is implemented.
[0036] One beneficial effect of the embodiments of the present disclosure is that the optimization processing method of the artificial intelligence model provided by the present invention can
[0037] Other features and advantages of the embodiments of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the embodiments of the present disclosure.
[0039] Figure 1 is a flowchart of an optimization processing method for an artificial intelligence model according to one embodiment;
[0040] Figure 2 is a block diagram of an optimization processing device for an artificial intelligence model according to one embodiment;
[0041] Figure 3 It is a schematic diagram of the hardware structure of an optimization processing device for an artificial intelligence model according to one embodiment. DETAILED DESCRIPTION
[0042] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of parts and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention.
[0043] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.
[0044] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.
[0045] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0046] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.
[0047] <Method Example>
[0048] Figure 1 This is a flowchart of a method for optimizing model-generated content according to an embodiment. The implementation subject is a server, a personal computer, or a mobile phone, etc., which is not limited here.
[0049] like Figure 1 As shown, the processing method for optimizing model generation content in this embodiment may include the following steps S110 to S140:
[0050] Step S110: Obtain the input task of the artificial intelligence model and the generated content generated by the artificial intelligence model based on the input task.
[0051] In this embodiment, the input task can be a short text prompt, a complex script paragraph, or a scene instruction, etc., which is not limited here. The artificial intelligence model can be a generative artificial intelligence model, which can use the input task as input to generate corresponding generated content.
[0052] Step S120 , extracting the task features of the input task and constructing a semantic graph of the input task.
[0053] In some embodiments, step S120 may include the following steps S210 and S220:
[0054] Step S210 , performing deep semantic parsing on the input task to obtain task features reflecting task elements of the input task.
[0055] Step S220: convert the task features into graph component units to construct a semantic graph of the input task.
[0056] In some examples, the input task is to generate a short video that shows the tranquil atmosphere of autumn. Through preset analysis instructions, the input task is deeply semantically parsed and processed to extract task elements, namely "autumn", "tranquil atmosphere" and "short video", as task features of the task elements. Among them, "autumn" represents the season, and an autumn scene is generated. "Tranquil atmosphere" represents emotion, and the short video is required to convey a sense of calm and tranquility. "Short video" means generating a short visual content. And according to these task features, it is converted into a structured semantic graph, which can contain the following information: autumn-scene (such as fallen leaves, cool), tranquility-emotion (such as quiet, soft), short video-content (short video).
[0057] Step S130 : Mapping the semantic graph and the generated content to a set semantic space to obtain a target deviation between the generated content and the semantic graph.
[0058] In some embodiments, before step S130, the method further includes the following steps S310 to S340:
[0059] Step S310: Acquire content features and content semantics of generated content; wherein the content features are image features and / or voice features.
[0060] In this embodiment, the processing device for optimizing model-generated content performs feature extraction and semantic modeling on the generated content, and can obtain content features and content semantics of the generated content.
[0061] Step S320 : determining first feature quantities corresponding to the content emotion feature, temporal structure feature, and logical structure feature respectively reflected in the generated content based on the content feature.
[0062] In this embodiment, the processing device for optimizing model-generated content performs sentiment analysis, temporal structure analysis, and logical structure analysis on image features and / or speech features using an analysis model (such as a benchmark model, GPT-40 or Gemini) to determine first feature quantities corresponding to the content's sentiment, temporal structure, and logical structure features, respectively. The sentiment analysis of image features and / or speech features can specifically include checking whether the generated content conveys the correct sentiment. For example, if the input task requires a "serene autumn atmosphere," the generated content is evaluated for calm expressions and warm tones, reflecting the sentiment of "tranquility," to obtain a first feature quantity reflecting the content's sentiment. The temporal structure analysis can specifically include evaluating whether the generated content's chronological order is reasonable and whether its pacing meets the requirements of the input task. For example, if the input task requires a smooth pacing, the generated content is evaluated for excessively rapid scene cuts or incoherent shot changes, to obtain a first feature quantity reflecting the temporal structure. The logical structure analysis of image features and / or speech features can specifically include checking whether the generated content is logically organized. For example, if the input task requires the generated content to display "autumn leaves," we evaluate whether the generated content contains autumn leaves and their placement is prominent, meeting the structural requirements of the input task, and thus obtaining a first feature quantity that reflects the logical structure. This first feature quantity is generally a feature vector.
[0063] Step S330: Compare the first feature value with the content semantics to obtain a semantic comparison feature.
[0064] In this embodiment, the first feature quantity can be compared and combined with the semantic embedding vector of the generated content to form an enhanced semantic vector, namely, a semantic comparison feature.
[0065] Step S340 , updating the semantic comparison feature into the generated content to obtain updated generated content.
[0066] In this embodiment, by embedding the first feature quantities of the generated content's content sentiment, temporal structure, and logical structure into the semantic embedding vector, the accuracy of subsequent pure semantic embedding based on the CLIP model in capturing detail features is effectively improved.
[0067] In some embodiments, step S130 may include the following steps S410 and S420:
[0068] Step S410 , mapping the content features, semantic comparison features, and semantic features of the generated content to a set semantic space to obtain a first embedding vector corresponding to the generated content and a second embedding vector corresponding to the semantic graph.
[0069] In this embodiment, the content features of the generated content, the semantic comparison features, and the semantic features of the semantic graph are mapped to the same high-dimensional semantic space, and a first embedding vector corresponding to the generated content and a second embedding vector corresponding to the semantic graph can be obtained.
[0070] In this embodiment, when the generated content is text, a feature vector reflecting the content characteristics is generated through a preset text model (such as a BERT model or a GPT model), and is combined with a feature vector reflecting the semantic comparison characteristics to obtain a first embedding vector. When the generated content is an image or video, a feature vector reflecting the content characteristics is generated through a preset image model (such as a CLIP model), and is combined with a feature vector reflecting the semantic comparison characteristics to obtain a first embedding vector.
[0071] In step S420 , the first embedding vector and the second embedding vector are input into a preset vector distance algorithm to obtain an offset distance as a target deviation between the generated content and the semantic graph.
[0072] In this embodiment, the vector distance algorithm can be a Euclidean distance calculation algorithm, that is, by calculating the similarity between the first embedded vector and the second embedded vector and using the vector distance algorithm, the straight-line distance between the two vectors in the semantic space can be obtained as the offset distance.
[0073] In this embodiment, by mapping content features, semantic contrast features and semantic features to the set semantic controls, these scattered features can be converted into vectors of the same dimension, achieving cross-dimensional alignment of concrete details and abstract semantics, and effectively avoiding the overall deviation of the generated content from the intention of the input task when the details are correct, so as to ensure the efficiency of iterative optimization of the artificial intelligence model.
[0074] Step S140: Adjust the artificial intelligence model based on the target optimization items associated with the target deviation.
[0075] In this embodiment, the artificial intelligence model can be continuously optimized based on the target deviation amount and the associated target optimization items in a structured form as guidance signals.
[0076] In some embodiments, after step S140, the method further includes the following steps S510 to S530:
[0077] Step S510 , determining whether the deviation value reflected by the target deviation is less than or equal to a set threshold.
[0078] Step S520 , when the deviation value reflected by the target deviation is less than or equal to the set threshold, determining the target deviation item corresponding to the target deviation according to a preset first mapping relationship; wherein the first mapping relationship reflects that different deviation items correspond to different deviation items.
[0079] Step S530 : determining the target optimization item corresponding to the target deviation item according to a preset second mapping relationship; wherein the second mapping relationship reflects that different deviation items correspond to different optimization items.
[0080] In this embodiment, when the deviation value is lower than a set threshold, it indicates that the semantic deviation between the input task and the generated content is large. The set threshold here can be set manually and is not limited here.
[0081] In some examples, the deviation can be represented by encoding. For example, the deviation is A1+B3+1, where A1 represents the character's expression, B3 represents the "serious" emotion required by the input task, and 1 represents non-compliance. In this case, the deviation item is the "serious" emotion mismatch, and the optimization issue can be adjusting the character's expression to make it more serious. For example, if the deviation item represents an overly cheerful expression, the deviation item mapped to this deviation item can be the "calm" emotion mismatch, and the optimization issue can be adjusting the character's expression to make it calmer. If the deviation item represents the absence of "autumn leaves," the deviation item mapped to this deviation item can be the omission of the "autumn leaves" element, and the optimization issue can be adding the autumn leaf element to highlight the autumnal feel. If the deviation item represents overly rapid scene changes, the deviation item mapped to this deviation item can be the overly fast timing, and the optimization issue can be slowing down the scene changes to maintain a smooth rhythm. If the deviation item represents the use of cool colors that are inconsistent with the warmth of autumn, the deviation item mapped to this deviation item can be the mismatch of the autumnal style, and the optimization issue can be adjusting to warm colors to reflect the warmth of autumn.
[0082] In this embodiment, by setting a set threshold, a first mapping relationship, and a second mapping relationship, some deviation amounts can be accurately converted into deviation items and optimization matters for natural language feedback, effectively improving the continuous optimization of the artificial intelligence model in subsequent input tasks.
[0083] In some embodiments, in order to visualize deviation items such as content omission, style mismatch, emotion mismatch, and timing error, the method further includes the following steps S610 and S620:
[0084] Step S610: Generate a deviation optimization report based on the target deviation items and target optimization items.
[0085] Step S620: Feedback the deviation optimization report to the associated object of the artificial intelligence model.
[0086] In this embodiment, the associated object may be one or more users using the artificial intelligence model.
[0087] In some examples, a deviation optimization report may include the following: 1. Mood mismatch. Issue: The character's expression is too joyful and fails to embody the emotion of "tranquility." Suggestion: Adjust the character's expression to make it more calming. 2. Content omission. Issue: The "autumn leaves" element is not displayed. Suggestion: Add autumn leaves to highlight the autumn feeling. 3. Timing error. Issue: The scene switching is too fast, affecting the tranquil atmosphere. Suggestion: Slow down the scene switching to maintain a smooth rhythm. 4. Stylistic mismatch. Issue: The use of cool colors does not match the warmth of autumn. Suggestion: Adjust to warm colors to reflect the warmth of autumn.
[0088] In this embodiment, a deviation optimization report is provided to facilitate user understanding and manual adjustment.
[0089] <Equipment Example 1>
[0090] Figure 2 FIG. 1 is a block diagram of an optimization processing device for an artificial intelligence model according to an embodiment. Figure 2 As shown, the optimization processing device 200 of the artificial intelligence model may include:
[0091] An acquisition module 210 is configured to acquire an input task of an artificial intelligence model and generated content generated by the artificial intelligence model based on the input task;
[0092] An extraction module 220 is configured to extract task features of the input task and construct a semantic graph of the input task;
[0093] An obtaining module 230 is configured to map the semantic graph and the generated content to a set semantic space to obtain a target deviation between the generated content and the semantic graph;
[0094] The adjustment module 240 is used to adjust the artificial intelligence model based on the target optimization items associated with the target deviation.
[0095] In some embodiments, the extraction module 220 is further used to perform deep semantic analysis on the input task to obtain task features reflecting the task elements of the input task; convert the task features into graph component units to construct a semantic graph of the input task.
[0096] In some embodiments, the optimization processing device 200 of the artificial intelligence model also includes a content acquisition module for acquiring content features and content semantics of the generated content; wherein the content features are image features and / or voice features; based on the content features, determining the first feature quantities corresponding to the content emotional features, temporal structure features and logical structure features reflected by the generated content; comparing the first feature quantities with the content semantics to obtain semantic comparison features; updating the semantic comparison features to the generated content to obtain the updated generated content.
[0097] In some embodiments, the obtaining module 230 is also used to map the content features of the generated content, the semantic comparison features and the semantic features of the semantic graph to a set semantic space to obtain a first embedding vector corresponding to the generated content and a second embedding vector corresponding to the semantic graph; the first embedding vector and the second embedding vector are input into a preset vector distance algorithm to obtain an offset distance as a target deviation between the generated content and the semantic graph.
[0098] In some embodiments, the adjustment module 240 is also used to determine whether the deviation value reflected by the target deviation is less than or equal to a set threshold; when the deviation value reflected by the target deviation is less than or equal to the set threshold, the target deviation item corresponding to the target deviation is determined according to a preset first mapping relationship; wherein, the first mapping relationship reflects that different deviation amounts correspond to different deviation items; according to a preset second mapping relationship, the target optimization item corresponding to the target deviation item is determined; wherein, the second mapping relationship reflects that different deviation items correspond to different optimization items.
[0099] In some embodiments, the optimization processing device 200 of the artificial intelligence model also includes a feedback module for generating a deviation optimization report based on the target deviation item and the target optimization item; and feeding back the deviation optimization report to the associated object of the artificial intelligence model.
[0100] <Equipment Example 2>
[0101] Figure 3 It is a schematic diagram of the hardware structure of an optimization processing device for an artificial intelligence model according to another embodiment.
[0102] like Figure 3 As shown, the optimization processing device 300 of the artificial intelligence model includes a processor 310 and a memory 320, wherein the memory 320 is used to store an executable computer program, and the processor 310 is used to execute a method such as any of the above method embodiments according to the control of the computer program.
[0103] Each module of the above-mentioned artificial intelligence model optimization processing device 200 can be implemented by the processor 310 in this embodiment executing the computer program stored in the memory 320, or can be implemented by other structures, which is not limited here.
[0104] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.
[0105] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.
[0106] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.
[0107] The computer program instructions for performing the operation of the present invention can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, and conventional procedural programming languages such as "C" language or similar programming languages. The computer readable program instructions can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), is personalized by utilizing the state information of the computer readable program instructions, and the electronic circuit can execute the computer readable program instructions, thereby realizing various aspects of the present invention.
[0108] Various aspects of the present invention are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.
[0109] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.
[0110] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.
[0111] The flowcharts and block diagrams in the accompanying drawings show the possible implementation architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of an instruction, and the module, program segment or part of the instruction contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.
[0112] While various embodiments of the present invention have been described above, the foregoing description is intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.
Claims
1. A processing method for optimizing model generated content, characterized in that, The method comprises: Obtaining an input task of an artificial intelligence model and generated content generated by the artificial intelligence model based on the input task; Extracting task features of the input task and constructing a semantic graph of the input task; Mapping the semantic graph and the generated content to a set semantic space to obtain a target deviation between the generated content and the semantic graph; Adjust the artificial intelligence model based on the target optimization matters associated with the target deviation amount.
2. The method according to claim 1, characterized in that The step of extracting the task features of the input task and constructing the semantic graph of the input task includes: Performing deep semantic parsing on the input task to obtain task features reflecting task elements of the input task; The task features are converted into graph composition units to construct a semantic graph of the input task.
3. The method according to claim 1, characterized in that Before mapping the semantic graph and the generated content to a set semantic space and obtaining a target deviation between the generated content and the semantic graph, the method further includes: Acquiring content features and content semantics of the generated content; wherein the content features are image features and / or voice features; Determining, based on the content features, first feature quantities corresponding to the content emotional features, temporal structure features, and logical structure features reflected by the generated content; Comparing the first feature with the content semantics to obtain a semantic comparison feature; The semantic comparison feature is updated to the generated content to obtain the updated generated content.
4. The method according to claim 3, characterized in that Mapping the semantic graph and the generated content to a set semantic space to obtain a target deviation between the generated content and the semantic graph includes: Mapping the content features of the generated content, the semantic comparison features, and the semantic features of the semantic graph to a set semantic space to obtain a first embedding vector corresponding to the generated content and a second embedding vector corresponding to the semantic graph; The first embedding vector and the second embedding vector are input into a preset vector distance algorithm to obtain an offset distance as a target deviation between the generated content and the semantic graph.
5. The method according to claim 1, wherein The adjusting the artificial intelligence model based on the target optimization items associated with the target deviation includes: Determining whether the deviation value reflected by the target deviation is less than or equal to a set threshold; When the deviation value reflected by the target deviation is less than or equal to a set threshold, determining a target deviation item corresponding to the target deviation according to a preset first mapping relationship; wherein the first mapping relationship reflects that different deviation items correspond to different deviation items; According to a preset second mapping relationship, the target optimization item corresponding to the target deviation item is determined; wherein the second mapping relationship reflects that different deviation items correspond to different optimization items.
6. The method according to claim 5, characterized in that The method further comprises: Generate a deviation optimization report based on the target deviation item and the target optimization item; Feedback the deviation optimization report to the associated object of the artificial intelligence model.
7. An optimization processing device for an artificial intelligence model, characterized in that: The device comprises: An acquisition module, configured to acquire an input task of an artificial intelligence model and generated content generated by the artificial intelligence model based on the input task; An extraction module, configured to extract task features of the input task and construct a semantic graph of the input task; an obtaining module, configured to map the semantic graph and the generated content to a set semantic space, and obtain a target deviation between the generated content and the semantic graph; An adjustment module is used to adjust the artificial intelligence model based on the target optimization items associated with the target deviation amount.
8. An optimization processing device for an artificial intelligence model, characterized in that: The system comprises a memory and a processor, wherein the memory is used to store a computer program; and the processor is used to execute the computer program to implement the method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which implements the method according to any one of claims 1 to 6 when executed by a processor.
10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 6 when being executed by a processor.