A Platform Content Recognition Method and System Based on AIGC

Abstract

This invention discloses a platform content recognition method and system based on AIGC. The method receives a first prompt message from the AIGC management platform and distributes it to two AIGC engines for processing. The two engines generate different feedback messages based on the same prompt message. Then, they identify and distinguish the first application scenario information corresponding to the first feedback message and the second application scenario information corresponding to the second feedback message. When the AIGC management platform determines that the similarity between the two application scenario messages exceeds a preset similarity threshold, it automatically synthesizes the first and second feedback messages to generate more comprehensive and accurate first synthesized information. This not only improves the accuracy and reliability of the information but also provides users with richer and more diverse information choices.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, and in particular relates to a platform content recognition method and system based on AIGC. Background Technology

[0002] With the rapid development of artificial intelligence technology, content generation and recognition technologies have been widely applied in various fields. In AI-generated Content (AIGC) platforms, content recognition technology is a key link in achieving intelligent content processing and generation. However, traditional content recognition methods often rely on a single model or algorithm, which not only limits the accuracy and efficiency of content recognition but also makes it difficult to cope with complex and ever-changing application scenarios.

[0003] In current AIGC platform management systems, the content recognition process is typically quite simple: after receiving user input, it is processed by a single recognition model and feedback is generated. This approach has the following drawbacks:

[0004] Limited accuracy: When dealing with complex or diverse inputs, a single model may lead to recognition errors due to insufficient generalization ability.

[0005] Poor adaptability: A single model is difficult to adjust and optimize flexibly to meet the needs of different application scenarios, which limits the scalability and maintainability of the system. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention proposes a platform content recognition method and system based on AIGC.

[0007] The technical solution of the present invention is as follows:

[0008] A platform content recognition method based on AIGC is applied to an AIGC platform management system, which includes an AIGC engine cluster and an AIGC management platform. Each AIGC engine in the AIGC engine cluster is connected to the AIGC management platform. The method includes:

[0009] The AIGC management platform obtains the first prompt information and inputs the first prompt information into the first AIGC engine and the second AIGC engine in the AIGC engine cluster.

[0010] The first AIGC engine responds to the first prompt information, generates first feedback information, and submits the first feedback information to the AIGC management platform; the second AIGC engine responds to the first prompt information, generates second feedback information, and submits the second feedback information to the AIGC management platform.

[0011] The AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, respectively.

[0012] If the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is greater than a preset similarity threshold, then the AIGC management platform outputs first synthesized information, which is a synthesized information of the first feedback information and the second feedback information.

[0013] Furthermore, if the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is less than the preset similarity threshold, then the AIGC management platform outputs the first feedback information or the second feedback information.

[0014] Furthermore, the AIGC management platform outputs either the first feedback information or the second feedback information, including:

[0015] The AIGC management platform inputs the first prompt information to the third AIGC engine in the AIGC engine cluster, so that the AIGC management platform responds to the first prompt information, generates third feedback information, and submits the third feedback information to the AIGC management platform;

[0016] The AIGC management platform determines the third application scenario information corresponding to the third feedback information, and determines the second scenario similarity between the first application scenario information and the third application scenario information, as well as the third scenario similarity between the second application scenario information and the third application scenario information.

[0017] If the AIGC management platform determines that the similarity of the second scene is greater than that of the third scene, then the AIGC management platform outputs the first feedback information;

[0018] If the AIGC management platform determines that the similarity of the second scene is less than that of the third scene, then the AIGC management platform outputs the second feedback information.

[0019] Furthermore, if the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is less than the preset similarity threshold, then the AIGC management platform will input the first prompt information to each AIGC engine in the AIGC engine cluster to generate a set of feedback information.

[0020] The AIGC management platform uses a preset classifier to identify the application scenario information corresponding to each feedback information in the feedback information set, so as to generate an application scenario information set.

[0021] The AIGC management platform generates a scenario similarity matrix based on the application scenario information set.

[0022] The AIGC management platform determines the classification feature vector corresponding to the feedback information set based on the scene similarity matrix;

[0023] The AIGC management platform determines the target feedback information based on the classification feature vector, and the target feedback information is the feedback information corresponding to the target classification characteristic value;

[0024] The AIGC management platform outputs the target feedback information.

[0025] Furthermore, the AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, including:

[0026] The AIGC management platform extracts features from the first feedback information to generate a first feature word sequence; the AIGC management platform extracts features from the second feedback information to generate a second feature word sequence.

[0027] The AIGC management platform inputs the first feature word sequence and the second feature word sequence into a preset application scenario classifier to determine the first application scenario information and the second application scenario information, respectively.

[0028] Furthermore, before the AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, it further includes:

[0029] Construct a training dataset, wherein each training sample in the training dataset includes question information and labeled application scenario information, wherein the question information includes question statements about products;

[0030] Based on each training sample in the constructed training dataset, including the question information, a corresponding question feature word sequence is generated. The question feature word sequence and the corresponding labeled application scenario information are used to train a preset initial classifier to generate the preset application scenario classifier. The preset initial classifier is a vector machine-based classifier model.

[0031] Furthermore, after the AIGC management platform outputs the target feedback information, it also includes:

[0032] If the AIGC management platform receives the first feedback instruction and, after re-outputting the new target feedback information, receives the second feedback instruction, then it generates first feature data. The first feedback instruction includes an instruction to regenerate the content corresponding to the first prompt information, and the second feedback instruction is an instruction to accept the new target feedback information. The first feature data includes the first prompt information, the target feedback information, and the new target feedback information.

[0033] The AIGC management platform adds the first feature data to the training dataset to iteratively train the preset application scenario classifier.

[0034] A platform content recognition system based on AIGC is applied to an AIGC platform management system including an AIGC engine cluster and an AIGC management platform. Each AIGC engine in the AIGC engine cluster is connected to the AIGC management platform. The system includes a first data processing module, a second data processing module, a third data processing module, and a fourth data processing module.

[0035] The first data processing module is used to obtain first prompt information through the AIGC management platform and input the first prompt information into the first AIGC engine and the second AIGC engine in the AIGC engine cluster;

[0036] The second data processing module is used for the first AIGC engine to generate first feedback information in response to the first prompt information and submit the first feedback information to the AIGC management platform, and the second AIGC engine to generate second feedback information in response to the first prompt information and submit the second feedback information to the AIGC management platform;

[0037] The third data processing module is used by the AIGC management platform to determine the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, respectively.

[0038] The fourth data processing module is used to output synthesized information based on the first scene similarity between the first application scene information and the second application scene information. If the AIGC management platform determines that the first scene similarity between the first application scene information and the second application scene information is greater than a preset similarity threshold, the AIGC management platform outputs first synthesized information, which is the synthesized information of the first feedback information and the second feedback information.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] This invention proposes an AIGC-based platform content recognition method that improves recognition accuracy and efficiency. By introducing diverse AIGC engines, this approach can address complex and ever-changing content recognition needs. Different engines may excel at handling different types or domains of information, thus providing more accurate and professional feedback. Simultaneously, intelligent analysis and merging functions further enhance recognition accuracy. Furthermore, the parallel processing approach significantly improves content recognition efficiency.

[0041] This invention enhances the system's flexibility and scalability: compared to traditional single models or algorithms, the AIGC engine cluster and management platform in this solution are easier to upgrade and optimize. As artificial intelligence technology continues to develop, new and more advanced engines can be continuously introduced to replace older ones, thereby improving the overall system performance. Simultaneously, the management platform also supports dynamic adjustment and optimization of the recognition process based on changes in user needs.

[0042] The method of this invention provides a richer and more diverse range of information choices: because it can generate multiple feedback messages and allow users to select the most suitable answer based on the similarity of the application scenario, it offers users a wider range of information options. This helps users gain a more comprehensive understanding of the problem and make more informed decisions.

[0043] The method of this invention achieves continuous learning and optimization: This scheme also possesses the capability for continuous learning and optimization. Each time a user provides feedback, the system records this information and uses it to optimize the classifier and improve the accuracy of future recognition. In this way, over time, the system will increasingly understand the user, providing more personalized and accurate services. Attached Figure Description

[0044] Figure 1 This is one of the flowcharts illustrating a platform content recognition method based on AIGC.

[0045] Figure 2 This is the second flowchart illustrating the AIGC-based platform content recognition method. Detailed Implementation

[0046] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0047] Example 1:

[0048] This invention discloses a platform content recognition method based on AIGC, applied to an AIGC platform management system including an AIGC engine cluster and an AIGC management platform. Each AIGC engine in the AIGC engine cluster is connected to the AIGC management platform, such as... Figure 1 As shown, the method includes:

[0049] S101. Obtain the first prompt information through the AIGC management platform and input the first prompt information into the first AIGC engine and the second AIGC engine in the AIGC engine cluster;

[0050] In this step, the AIGC management platform receives initial prompts from users or generated by the system. These prompts can be descriptions or keywords related to a specific topic, product, or issue.

[0051] Specifically, users can input prompts about products, services, or questions via text or voice input. For example, users can enter product names, descriptions, or related keywords in the search box, or ask questions through a voice assistant. For voice input, speech recognition technology can be used to convert speech signals into text information. This requires the system to have a highly efficient speech recognition module capable of accurately converting user voice commands into initial text prompts.

[0052] The received initial prompts undergo preprocessing, including removing irrelevant characters, correcting spelling errors, and performing synonym replacements, to improve the accuracy and efficiency of subsequent AIGC engine processing. This is particularly important for product names, as the system may need to identify and handle different meanings in different contexts.

[0053] The preprocessed first prompt message is sent simultaneously to multiple engines in the AIGC engine cluster (such as the first AIGC engine and the second AIGC engine in this example). These engines may generate different feedback messages based on the same prompt message.

[0054] S102, the first AIGC engine and the second AIGC engine respectively respond to the first prompt information and generate the first feedback information and the second feedback information;

[0055] In this step, the first AIGC engine responds to the first prompt information, generates first feedback information, and submits the first feedback information to the AIGC management platform; the second AIGC engine responds to the first prompt information, generates second feedback information, and submits the second feedback information to the AIGC management platform.

[0056] Specifically, the received first prompt message is simultaneously input to both the first and second AIGC engines in the AIGC engine cluster. These two engines may generate different feedback messages based on the same prompt message; for example, if the first prompt message input by the user is incomplete, the outputs of different AIGC engines may be inconsistent. The first AIGC engine responds to the first prompt message, generates a first feedback message, and submits this message to the AIGC management platform. The second AIGC engine also responds to the first prompt message, generates a second feedback message, and submits this message to the AIGC management platform.

[0057] S103, the AIGC management platform respectively determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information;

[0058] Specifically, the AIGC management platform processes the received first and second feedback information to determine their corresponding first and second application scenario information, respectively. This step can be achieved by extracting features from the feedback information and then inputting them into a preset application scenario classifier. Specifically, the AIGC management platform can extract features from the first feedback information to generate a first feature word sequence, and then input this sequence into the preset application scenario classifier to determine the first application scenario information. Similarly, features are extracted and classified from the second feedback information to determine the second application scenario information.

[0059] S104. The AIGC management platform outputs synthesized information based on the first scene similarity between the first application scenario information and the second application scenario information. In this step, the AIGC management platform calculates the first scene similarity between the first application scenario information and the second application scenario information. This step can be achieved by comparing the feature words, semantic relationships, etc., in the two application scenario information.

[0060] If the AIGC management platform determines that the similarity between the first application scenario information and the second application scenario information is greater than a preset similarity threshold, then the AIGC management platform outputs the first composite information. This first composite information is a combination of the first feedback information and the second feedback information, providing more comprehensive and accurate information. When the application scenario information similarity is low, outputting a single feedback information instead of composite information avoids confusion for users due to information redundancy or inconsistency. This approach improves the system's ability to filter and select information, making the output information more accurate and targeted. Furthermore, by selecting to output either the first or second feedback information, the AIGC management platform can ensure that users receive information most likely to meet their needs.

[0061] Specifically, the process can begin by removing irrelevant characters, redundant information, and special symbols from the first and second feedback messages to ensure information purity. This can be achieved by using techniques such as regular expressions and natural language processing to preprocess the first and second feedback messages, such as removing HTML tags and special characters, and correcting spelling errors. Next, key features are extracted from the cleaned feedback messages. By comparing keywords in the two sets of feedback messages, commonalities and differences are identified to prepare for fusion. The commonalities and differences between the two sets of feedback messages are then integrated to avoid information duplication and omissions. Finally, a syntax checker is used to perform a grammar check on the synthesized information. Finally, the synthesized information is presented to the user through an AIGC management platform.

[0062] In another possible implementation, the first feedback information and the second feedback information can be displayed side by side. Alternatively, the first and second feedback information can be re-inputted into the first AIGC engine, the second AIGC engine, or another AIGC engine, with a prompt indicating that the first and second feedback information be merged to output composite information.

[0063] In the above scheme, the first prompt information received through the AIGC management platform is distributed to two AIGC engines (the first AIGC engine and the second AIGC engine) for processing. This parallel processing method significantly improves the speed and efficiency of content generation. Simultaneously, the two engines generate different feedback information based on the same prompt information, providing a content foundation for subsequent application scenario analysis and information synthesis. Then, the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information are identified and distinguished. When the AIGC management platform determines that the similarity between the two application scenario information exceeds a preset similarity threshold, it automatically synthesizes the first and second feedback information to generate more comprehensive and accurate first synthesized information. This not only improves the accuracy and reliability of the information but also provides users with richer and more diverse information choices.

[0064] Furthermore, if the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is less than a preset similarity threshold, the AIGC management platform outputs either the first feedback information or the second feedback information. This selection can be achieved through further analysis and comparison, for example, by introducing a third AIGC engine to generate third feedback information, and making a decision based on the similarity between the application scenario information of the third feedback information and the first and second application scenario information.

[0065] In the above solution, when the application scenario information is not highly similar, outputting a single feedback message instead of synthesized information can avoid user confusion caused by information redundancy or inconsistency. This approach improves the system's ability to filter and select information, making the output information more accurate and targeted. Furthermore, by selecting to output either the first or second feedback message, the AIGC management platform can ensure that users receive information most likely to meet their needs.

[0066] Furthermore, the AIGC management platform outputs first or second feedback information, including:

[0067] The AIGC management platform inputs the first prompt information into the third AIGC engine in the AIGC engine cluster, so that the AIGC management platform responds to the first prompt information, generates the third feedback information, and submits the third feedback information to the AIGC management platform.

[0068] The AIGC management platform determines the third application scenario information corresponding to the third feedback information, and determines the second scenario similarity between the first application scenario information and the third application scenario information, as well as the third scenario similarity between the second application scenario information and the third application scenario information.

[0069] If the AIGC management platform determines that the similarity of the second scene is greater than that of the third scene, then the AIGC management platform outputs the first feedback information.

[0070] If the AIGC management platform determines that the similarity of the second scene is less than that of the third scene, then the AIGC management platform outputs the second feedback information.

[0071] In the above scheme, by introducing a third AIGC engine and third feedback information, the system can acquire more relevant content about the first prompt, thereby increasing the dimensionality and accuracy of information recognition. This triple verification method helps reduce the error rate of information recognition and improve the system's recognition accuracy. When determining which feedback information to output, the system considers not only the application scenario information of the first and second feedback information but also the application scenario information of the third feedback information for comparison. This comprehensive consideration of multiple factors makes the system's decision-making more comprehensive and reasonable, helping to improve the overall performance of the system. By comparing the similarity between the second and third scenarios, the system can intelligently select the feedback information most relevant to the first prompt information. This personalized output strategy helps improve user experience and satisfaction, while also improving the system's information push efficiency.

[0072] In this embodiment, the first prompt information received by the AIGC management platform is distributed to two AIGC engines for processing. The two engines generate different feedback information based on the same prompt information. Then, the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information are identified and distinguished. When the AIGC management platform determines that the similarity between the two application scenario information exceeds a preset similarity threshold, it automatically synthesizes the first feedback information and the second feedback information to generate more comprehensive and accurate first synthesized information. This not only improves the accuracy and reliability of the information, but also provides users with richer and more diverse information choices.

[0073] It is worth noting that when the above solution is applied to an e-commerce dialogue management system, it can quickly output relevant information about the user's question and cross-validate the content generated by multiple different AIGC engines. This ensures that the content generated by the AIGC engine is directly output only when the information in two application scenarios is highly similar or even identical. This reduces the mismatch between the output content and the user's input question due to the different meanings of the product name in different application scenarios.

[0074] Example 2:

[0075] The present invention provides a platform content recognition method based on AIGC, comprising:

[0076] S201. Obtain the first prompt information through the AIGC management platform and input the first prompt information into the first AIGC engine and the second AIGC engine in the AIGC engine cluster.

[0077] In this step, the AIGC management platform receives initial prompts from user input or generated by the system. These prompts can be descriptions or keywords related to a topic, product, question, etc. Optionally, the AIGC platform management system can be integrated with an e-commerce dialogue management system. The initial prompts include product feature words entered by the customer; these product feature words are used to represent different target objects in different application scenarios.

[0078] Specifically, users can input prompts about products, services, or questions via text or voice input. For example, users can enter product names, descriptions, or related keywords in the search box, or ask questions through a voice assistant. For voice input, speech recognition technology can be used to convert speech signals into text information. This requires the system to have a highly efficient speech recognition module capable of accurately converting user voice commands into initial text prompts.

[0079] The received initial prompts undergo preprocessing, including removing irrelevant characters, correcting spelling errors, and performing synonym replacements, to improve the accuracy and efficiency of subsequent AIGC engine processing. This is particularly important for product names, as the system may need to identify and handle different meanings in different contexts.

[0080] The preprocessed first prompt message is sent simultaneously to multiple engines in the AIGC engine cluster (such as the first AIGC engine and the second AIGC engine in this example). These engines may generate different feedback messages based on the same prompt message.

[0081] Furthermore, in one possible application scenario, the first AIGC engine and the second AIGC engine mentioned above can both be general-purpose AIGC engines, so as to output more accurate information to users by comparing the application scenario information of the content of different AIGC engines.

[0082] It's worth noting that in another possible application scenario, the first AIGC engine mentioned above can be an internal AIGC engine within the e-commerce dialogue management system. Its generated content is based on a database built by the e-commerce platform itself, with training data more closely aligned with the e-commerce domain. The second AIGC engine, on the other hand, can be an externally used, general-purpose AIGC engine. Since the internal AIGC engine's database has high accuracy but weak generalization ability, while the external AIGC engine's database has a wide data volume but relatively low accuracy, the application scenario information of the content generated by the external and internal AIGC engines can be compared to determine whether the scenario directionality of the content generated by the internal and external AIGC engines is the same. If the scenario directionality is the same, a composite of the two is output, thus ensuring both high generalization and good specificity in the generated content.

[0083] Furthermore, it's worth noting that the aforementioned first AIGC engine can be an AIGC engine currently being trained or awaiting testing within the e-commerce dialogue management system, while the second AIGC engine can be an externally applicable AIGC engine. In this case, the aforementioned first prompt information can be the relevant test prompt statements output by the tester. By comparing the application scenario information of the content generated by the external AIGC engine and the internal AIGC engine, it's possible to determine whether the scenario directionality of the content generated by the internal AIGC engine is the same as that generated by the external AIGC engine, thereby iteratively optimizing and training the internal AIGC engine. Here, feedback from the tester can also be incorporated to generate and iterate training data in the internal AIGC engine training database.

[0084] S202, the first AIGC engine and the second AIGC engine respond to the first prompt information and generate the first feedback information and the second feedback information, respectively.

[0085] In this step, the first AIGC engine responds to the first prompt information, generates first feedback information, and submits the first feedback information to the AIGC management platform; the second AIGC engine responds to the first prompt information, generates second feedback information, and submits the second feedback information to the AIGC management platform.

[0086] S203 and the AIGC management platform respectively determine the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information.

[0087] Specifically, the AIGC management platform processes the received first and second feedback information to determine their corresponding first and second application scenario information, respectively. This step can be achieved by extracting features from the feedback information and then inputting them into a preset application scenario classifier. Specifically, the AIGC management platform can extract features from the first feedback information to generate a first feature word sequence, and then input this sequence into the preset application scenario classifier to determine the first application scenario information. Similarly, features are extracted and classified from the second feedback information to determine the second application scenario information.

[0088] In one possible implementation, the AIGC management platform extracts features from the first feedback information to generate a first feature word sequence; the AIGC management platform also extracts features from the second feedback information to generate a second feature word sequence. The AIGC management platform then inputs the first and second feature word sequences into a preset application scenario classifier to determine the first and second application scenario information, respectively.

[0089] Specifically, the AIGC management platform first receives first and second feedback information from different sources. This information may include data in various forms such as text, images, audio, or video. To effectively identify the application scenarios to which this information belongs, the platform first needs to extract features from this feedback information. Specifically, the platform can use natural language processing technology or corresponding multimedia processing technology to analyze the first feedback information. For text information, steps such as word segmentation, part-of-speech tagging, and named entity recognition are performed to generate a first feature word sequence representing the core content of the information. For non-text information, such as images or videos, computer vision technology may be used to extract key visual features and convert them into a processable feature vector form. Similar to the first feedback information, the platform also performs feature extraction on the second feedback information to generate a second feature word sequence or feature vector. This step ensures that regardless of the information format, it can be converted into a format that the platform can understand and compare.

[0090] Next, the AIGC management platform uses a pre-defined application scenario classifier to classify the extracted feature word sequences or feature vectors to determine the application scenario information corresponding to each feedback message. In the preliminary preparation phase, the platform needs to train an efficient application scenario classifier based on a labeled dataset. This classifier can be built based on machine learning algorithms (such as support vector machines and random forests) or deep learning models (such as convolutional neural networks and recurrent neural networks). The first and second feature word sequences extracted in the first step are input into the trained application scenario classifier. Through analysis and learning of these features, the classifier can identify the application scenario category most likely to which each feedback message belongs. The classifier outputs two results, corresponding to the application scenario information of the first and second feedback messages, respectively. These application scenarios may include, but are not limited to, multiple application areas such as food, furniture, appliances, and clothing, depending on the platform's design and service scope.

[0091] Specifically, before the AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, a training dataset needs to be constructed. Each training sample in the training dataset includes question information and labeled application scenario information. The question information includes questions about the product. Based on the question information in each training sample in the constructed training dataset, corresponding question feature word sequences are generated. These question feature word sequences and the corresponding labeled application scenario information are then used to train a preset initial classifier to generate a preset application scenario classifier. The preset initial classifier is a vector machine-based classifier model. This can be achieved by collecting a large amount of question information and organizing the collected question information and corresponding labeled application scenario information into a structured data format to form the training dataset. Each training sample in the training dataset should contain question information and labeled application scenario information. The question information in the training dataset is preprocessed, including stop word removal, word segmentation, and stemming, to clean the data and reduce noise. Natural language processing techniques, such as TF-IDF (Term Frequency-Inverse Document Frequency) or word embeddings (e.g., Word2Vec, BERT), are then used to convert the preprocessed question information into feature word sequences or vector representations. Next, an initial classifier model based on a vector machine (e.g., Support Vector Machine, SVM) is used as the training basis. The initial classifier is trained using the question feature word sequences and corresponding labeled application scenario information from the training dataset. During training, the classifier learns how to map the question feature word sequences to the correct application scenario information. Then, the classifier's performance is optimized through methods such as cross-validation and hyperparameter tuning. The accuracy, recall, and other performance metrics of the classifier are evaluated using a validation dataset to ensure that the classifier has good generalization ability. After training and optimization, a final preset application scenario classifier is generated, which can accurately classify question information into the corresponding application scenario. In the AIGC management platform, when the first and second feedback information are received, the trained application scenario classifier is first used to classify the question information in the feedback information, determining the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information. It is worth noting that the aforementioned preset application scenario classifier can be trained based on various general classifier base models. In this embodiment, no specific limitation is made on its specific form, and the support vector machine on which it is based is an exemplary implementation method.

[0092] S204. The AIGC management platform inputs the first prompt information to each AIGC engine in the AIGC engine cluster to generate a set of feedback information.

[0093] If the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is less than a preset similarity threshold, then the AIGC management platform will input the first prompt information into the AIGC engine cluster E = {e1, e2, ..., e i ,…,e n Each AIGC engine in} generates a set of feedback information M = {m1, m2, ..., m}. i ,…,m n}, where n is the number of AIGC engines in AIGC engine cluster E.

[0094] The S205 AIGC management platform uses a preset classifier to identify the application scenario information corresponding to each feedback information in the feedback information set, so as to generate an application scenario information set.

[0095] The AIGC management platform uses a pre-defined classifier to identify the application scenario information corresponding to each feedback information in the feedback information set M, in order to generate an application scenario information set D = {d1, d2, ..., d...} i ,…,d n}

[0096] S206, the AIGC management platform generates a scenario similarity matrix based on the application scenario information set.

[0097] The AIGC management platform generates a scenario similarity matrix S based on the application scenario information set D:

[0098]

[0099] Wherein, s in the scene similarity matrix S ij Let be the similarity between the i-th application scenario information and the j-th application scenario information in the application scenario information set D.

[0100] S207. The AIGC management platform determines the classification feature vector corresponding to the feedback information set based on the scene similarity matrix.

[0101] The AIGC management platform uses Formula 1 and determines the classification feature vector K = {k1,k2,…,k} based on the scene similarity S. i ,…,k n}, where k i Let be the classification characteristic value corresponding to the i-th feedback information in the feedback information set M, and Formula 1 is:

[0102]

[0103] S208, the AIGC management platform determines target feedback information based on classification feature vectors.

[0104] The AIGC management platform uses Formula 2 and determines the target feedback information based on the classification feature vector K. The target feedback information is the target classification characteristic value. The corresponding feedback information, as shown in Formula 2, is:

[0105]

[0106] S209, AIGC management platform outputs target feedback information.

[0107] In the above scheme, by comparing the similarity between the first application scenario information and the second application scenario information, and when the similarity is less than a preset threshold, the AIGC engine cluster is activated to generate a set of feedback information. This step ensures that when the scenario information does not match or the differences are large, the possibility of outputting matching content can be improved by recognizing it through the AIGC engine cluster.

[0108] By using a pre-defined classifier to identify application scenario information for each piece of feedback information in the feedback information set, an application scenario information set is generated, further refining the granularity of content recognition and improving its accuracy. A scenario similarity matrix is ​​constructed by calculating the similarity between each piece of application scenario information in the application scenario information set; this step comprehensively reflects the correlation and differences between various application scenarios. Then, classification feature vectors are extracted, and the scenario similarity matrix is ​​weighted and summed using Formula 1 above, reflecting the overall accuracy of each piece of feedback information in the feedback information set generated by the AIGC engine cluster. Finally, Formula 2 is used to determine the target classification characteristic value, and target feedback information is selected accordingly; this step automatically filters out the most representative and relevant feedback information.

[0109] Furthermore, after the AIGC management platform outputs target feedback information, if the AIGC management platform receives the first feedback instruction and, after re-outputting new target feedback information, receives the second feedback instruction, then it generates first feature data. The first feedback instruction includes an instruction to regenerate the content corresponding to the first prompt information, and the second feedback instruction is an instruction to accept the new target feedback information. The first feature data includes the first prompt information, the target feedback information, and the new target feedback information. The AIGC management platform adds the first feature data to the training dataset for iterative training of the preset application scenario classifier. In the above scheme, by introducing a user feedback mechanism, when a user requests the regeneration of the initially generated content (i.e., the first prompt information) through the first feedback instruction, the platform generates new target feedback information based on this feedback. If the user accepts this new content (i.e., issues the second feedback instruction), it indicates that the new content better meets the user's needs. This mechanism ensures that the generated content is closer to the user's expectations, thereby improving content quality and satisfaction. The platform integrates user feedback data (including initial prompts, original target feedback, and regenerated target feedback) into primary feature data and adds it to the training dataset. This allows the platform to continuously enrich and optimize the model's training data, enabling the model to learn more about user-defined preferences.

[0110] Furthermore, it's worth noting that after the AIGC management platform outputs target feedback information, if it receives a second prompt and outputs new target feedback information based on that, and then receives a third feedback instruction, it generates second feature data. The third feedback instruction is an instruction to accept the new target feedback information. The second feature data includes the first prompt, the second prompt, and the new target feedback information. The AIGC management platform adds the second feature data to the training dataset for iterative training of the preset application scenario classifier. In the above scheme, after the AIGC management platform completes content recognition and determines the target feedback information, it outputs this information to the user. This feedback information is generated based on the first prompt (such as a question) provided by the user and the processing results of the preset application scenario classifier. If the user provides additional information or corrects the previous question after receiving the initial target feedback information, the AIGC management platform will receive this new prompt, i.e., the second prompt. The platform inputs the second prompt into the preset application scenario classifier, combines it with the previous processing results, and generates new target feedback information. This new target feedback information is then output to the user to meet the user's adjusted needs based on the second prompt. If the user accepts or acknowledges the new target feedback information and sends a corresponding feedback instruction (the third feedback instruction) through the AIGC management platform, the AIGC management platform integrates the first prompt information, the second prompt information, and the finally accepted new target feedback information to generate second feature data. This data contains complete process information from the user's initial question to the final acceptance of feedback information. The generated second feature data is added to the original training dataset to enrich the dataset and include more data from real-world usage scenarios. The updated training dataset is then used to iteratively train the preset application scenario classifier. This process aims to enable the classifier to learn more about changes and user feedback in real-world scenarios, thereby further improving its classification accuracy and adaptability. During the iterative training process, the classifier's performance is regularly evaluated, and corresponding optimizations are made based on the evaluation results to ensure that the classifier can continuously provide high-quality application scenario classification services.

[0111] Furthermore, after the AIGC management platform outputs target feedback information, if the AIGC management platform receives a second prompt and outputs new target feedback information based on the second prompt, and then receives a fourth feedback instruction, and determines that the text similarity between the first and second prompts is greater than a preset similarity threshold, and that the second prompt includes feature keywords not present in the first prompt, then third feature data is generated. The third feature data includes the first prompt, the second prompt, and the new target feedback information. The fourth feedback instruction is an instruction to accept the new target feedback information. The AIGC management platform adds the third feature data to the training dataset and configures the weight values ​​of the feature keywords in subsequent training to be greater than a preset weight threshold for iterative training of the preset application scenario classifier. In this scheme, when the AIGC management platform receives the user's second prompt and outputs new target feedback information accordingly, if the user accepts this new feedback (i.e., sends a fourth feedback instruction), and the system determines that there is a high text similarity between the first and second prompts but that they contain newly added feature keywords, the system will generate third feature data. This process ensures that the classifier learns that the user's initial response to a question is insufficient to meet their needs, while a second question, enhanced with qualifiers, provides a response that satisfies their requirements. After generating the third feature data, the AIGC management platform not only adds it to the training dataset but also specifically configures the weights of these newly added feature keywords in subsequent training to be higher than a preset weight threshold. This emphasizes the importance of these keywords in classifier training, helping the classifier to more accurately identify and process this key information, thus improving content recognition accuracy. By incorporating third feature data containing new feature keywords and diverse expressions into the training dataset, the AIGC management platform can perform more comprehensive iterative training on the classifier for the preset application scenarios. This process not only enriches the diversity of training samples but also allows the classifier to focus more on feature keywords that frequently appear in practical applications and have a significant impact on classification results. Through this technical process, the AIGC management platform can more accurately understand the user's true intentions and provide feedback that better meets user expectations. This helps improve user satisfaction and loyalty, while enhancing the platform's competitiveness and market position. Furthermore, as the classifier's performance continues to improve, the platform's content recognition and processing capabilities will be significantly enhanced, providing users with a more efficient and accurate service experience. By continuously receiving user feedback, generating feature data, and conducting iterative training, the AIGC management platform achieves self-learning and evolution.

[0112] Example 3:

[0113] The present invention discloses an AIGC-based platform content recognition system, which is applied to an AIGC platform management system including an AIGC engine cluster and an AIGC management platform. Each AIGC engine in the AIGC engine cluster is connected to the AIGC management platform. The system includes a first data processing module, a second data processing module, a third data processing module and a fourth data processing module.

[0114] The first data processing module is used to obtain the first prompt information through the AIGC management platform and input the first prompt information into the first AIGC engine and the second AIGC engine in the AIGC engine cluster.

[0115] The second data processing module is used for the first AIGC engine to generate first feedback information in response to the first prompt information and submit the first feedback information to the AIGC management platform, and for the second AIGC engine to generate second feedback information in response to the first prompt information and submit the second feedback information to the AIGC management platform.

[0116] The third data processing module is used by the AIGC management platform to determine the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, respectively.

[0117] The fourth data processing module is used to output synthesized information based on the first scene similarity between the first application scene information and the second application scene information. If the AIGC management platform determines that the first scene similarity between the first application scene information and the second application scene information is greater than a preset similarity threshold, the AIGC management platform outputs the first synthesized information, which is the synthesized information of the first feedback information and the second feedback information.

[0118] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A platform content recognition method based on AIGC, applied to an AIGC platform management system including an AIGC engine cluster and an AIGC management platform, wherein each AIGC engine in the AIGC engine cluster is connected to the AIGC management platform, characterized in that, The method includes: The AIGC management platform obtains the first prompt information and inputs the first prompt information into the first AIGC engine and the second AIGC engine in the AIGC engine cluster. The first AIGC engine responds to the first prompt information, generates first feedback information, and submits the first feedback information to the AIGC management platform; the second AIGC engine responds to the first prompt information, generates second feedback information, and submits the second feedback information to the AIGC management platform. The AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, respectively. If the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is greater than a preset similarity threshold, then the AIGC management platform outputs first synthesized information, which is a synthesized information of the first feedback information and the second feedback information. If the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is less than the preset similarity threshold, then the AIGC management platform outputs the first feedback information or the second feedback information. The AIGC management platform outputs either the first feedback information or the second feedback information, including: The AIGC management platform inputs the first prompt information to the third AIGC engine in the AIGC engine cluster, so that the AIGC management platform responds to the first prompt information, generates third feedback information, and submits the third feedback information to the AIGC management platform; The AIGC management platform determines the third application scenario information corresponding to the third feedback information, and determines the second scenario similarity between the first application scenario information and the third application scenario information, as well as the third scenario similarity between the second application scenario information and the third application scenario information. If the AIGC management platform determines that the similarity of the second scene is greater than that of the third scene, then the AIGC management platform outputs the first feedback information; If the AIGC management platform determines that the similarity of the second scene is less than the similarity of the third scene, then the AIGC management platform outputs the second feedback information.

2. The platform content recognition method based on AIGC according to claim 1, characterized in that, The AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, including: The AIGC management platform extracts features from the first feedback information to generate a first feature word sequence; the AIGC management platform extracts features from the second feedback information to generate a second feature word sequence. The AIGC management platform inputs the first feature word sequence and the second feature word sequence into a preset application scenario classifier to determine the first application scenario information and the second application scenario information, respectively.

3. The platform content recognition method based on AIGC according to claim 1, characterized in that, Before the AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, it further includes: Construct a training dataset, wherein each training sample in the training dataset includes question information and labeled application scenario information, wherein the question information includes question statements about products; Based on each training sample in the constructed training dataset, including the question information, a corresponding question feature word sequence is generated. The question feature word sequence and the corresponding labeled application scenario information are used to train a preset initial classifier to generate the preset application scenario classifier. The preset initial classifier is a vector machine-based classifier model.

4. A platform content recognition method based on AIGC, applied to an AIGC platform management system including an AIGC engine cluster and an AIGC management platform, wherein each AIGC engine in the AIGC engine cluster is connected to the AIGC management platform, characterized in that, The method includes: The AIGC management platform obtains the first prompt information and inputs the first prompt information into the first AIGC engine and the second AIGC engine in the AIGC engine cluster. The first AIGC engine responds to the first prompt information, generates first feedback information, and submits the first feedback information to the AIGC management platform; the second AIGC engine responds to the first prompt information, generates second feedback information, and submits the second feedback information to the AIGC management platform. The AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, respectively. If the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is greater than a preset similarity threshold, then the AIGC management platform outputs first synthesized information, which is a synthesized information of the first feedback information and the second feedback information. If the AIGC management platform determines that the first scenario similarity between the first application scenario information and the second application scenario information is less than the preset similarity threshold, then the AIGC management platform will input the first prompt information to each AIGC engine in the AIGC engine cluster to generate a set of feedback information. The AIGC management platform uses a preset classifier to identify the application scenario information corresponding to each feedback information in the feedback information set, so as to generate an application scenario information set. The AIGC management platform generates a scenario similarity matrix based on the application scenario information set. The AIGC management platform determines the classification feature vector corresponding to the feedback information set based on the scene similarity matrix; The AIGC management platform determines the target feedback information based on the classification feature vector, and the target feedback information is the feedback information corresponding to the target classification characteristic value; The AIGC management platform outputs the target feedback information.

5. The platform content recognition method based on AIGC according to claim 4, characterized in that, The AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, including: The AIGC management platform extracts features from the first feedback information to generate a first feature word sequence; the AIGC management platform extracts features from the second feedback information to generate a second feature word sequence. The AIGC management platform inputs the first feature word sequence and the second feature word sequence into a preset application scenario classifier to determine the first application scenario information and the second application scenario information, respectively.

6. The platform content recognition method based on AIGC according to claim 4, characterized in that, Before the AIGC management platform determines the first application scenario information corresponding to the first feedback information and the second application scenario information corresponding to the second feedback information, it further includes: Construct a training dataset, wherein each training sample in the training dataset includes question information and labeled application scenario information, wherein the question information includes question statements about products; Based on each training sample in the constructed training dataset, including the question information, a corresponding question feature word sequence is generated. The question feature word sequence and the corresponding labeled application scenario information are used to train a preset initial classifier to generate the preset application scenario classifier. The preset initial classifier is a vector machine-based classifier model.

7. The platform content recognition method based on AIGC according to claim 4, characterized in that, After the AIGC management platform outputs the target feedback information, it also includes: If the AIGC management platform receives the first feedback instruction and, after re-outputting the new target feedback information, receives the second feedback instruction, then it generates first feature data. The first feedback instruction includes an instruction to regenerate the content corresponding to the first prompt information, and the second feedback instruction is an instruction to accept the new target feedback information. The first feature data includes the first prompt information, the target feedback information, and the new target feedback information. The AIGC management platform adds the first feature data to the training dataset to iteratively train the preset application scenario classifier.

Images