Information processing method and related device

Abstract

The embodiment of the invention discloses an information processing method and a related device, and the method comprises the steps: executing an information classification task through an information generation model which completes various information generation tasks based on indication information, and adjusting the indication information corresponding to the information classification task. And the information generation model can have the capability of carrying out information classification on the new to-be-classified information. Compared with the model parameters, the indication information is less in information amount and higher in adjustment efficiency, so that model training in the mode is more efficient, less training resources and time need to be consumed, and continuously changing information classification requirements can be quickly responded.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and in particular, to an information processing method and related devices. Background Art

[0002] With the continuous development of computer technology, model technologies for information processing have become increasingly mature, and various models are widely used in multiple technical fields. Among them, classification models are one of the widely used model types. Through classification models, various information classification tasks can be performed, such as identifying object categories in images, adding content tags to information, etc.

[0003] During the application of the model, it may be necessary to classify information of information categories that did not appear during the training process. For example, it may be necessary to classify animals that did not appear during the training process. In related technologies, it is necessary to adjust the model parameters corresponding to the classification model based on the training samples corresponding to the new information classification task, so that the model can learn new classification knowledge, so that the trained classification model can be used to perform the new information classification task.

[0004] Due to the large number of model parameters, the model adjustment method in related technologies requires a large amount of information processing resources and time, and it is difficult to adapt to information classification tasks with frequent dynamic changes. Summary of the Invention

[0005] To solve the above technical problems, this application provides an information processing method, which replaces updating model parameters by correcting indication information, and improves the model training efficiency while ensuring the model training effect.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, the embodiments of this application disclose an information processing method, and the method includes:

[0008] Obtain information to be classified and initial indication information, where the initial indication information is used to describe an information classification task;

[0009] Through an information generation model, perform the information classification task on the information to be classified according to the initial indication information, and generate first undetermined label information corresponding to the information to be classified, where the first undetermined label information is used to identify the information content corresponding to the information to be classified;

[0010] Determine first accuracy information corresponding to the first undetermined label information, where the first accuracy information is used to characterize the accuracy of the first undetermined label information in identifying the information content;

[0011] Adjust the initial indication information according to the first accuracy information to obtain first indication information, where the accuracy corresponding to the label information obtained by the information generation model performing the information classification task according to the first indication information is greater than a first accuracy threshold, and the information generation model is used to perform the information classification task according to the first indication information.

[0012] In a second aspect, an embodiment of the present application discloses an information processing device, which includes a first acquisition unit, a first generation unit, a first determination unit, and a first adjustment unit:

[0013] The first acquisition unit is configured to acquire information to be classified and initial indication information, where the initial indication information is used to describe an information classification task;

[0014] The first generation unit is configured to, through an information generation model, perform the information classification task on the information to be classified according to the initial indication information, and generate first pending label information corresponding to the information to be classified, where the first pending label information is used to identify the information content corresponding to the information to be classified;

[0015] The first determination unit is configured to determine first accuracy information corresponding to the first pending label information, where the first accuracy information is used to characterize the accuracy of the first pending label information in identifying the information content;

[0016] The first adjustment unit is configured to adjust the initial indication information according to the first accuracy information to obtain first indication information, where the accuracy corresponding to the label information obtained by the information generation model performing the information classification task according to the first indication information is greater than a first accuracy threshold, and the information generation model is used to perform the information classification task according to the first indication information.

[0017] In a possible implementation manner, the first adjustment unit is specifically configured to:

[0018] Through the information generation model, perform a defect analysis task on the initial indication information according to second indication information, and generate defect information corresponding to the initial indication information, where the defect information is used to describe the information defect existing in the initial indication information, and the information defect is the reason for the accuracy characterized by the first accuracy information not being greater than the first accuracy threshold, and the second indication information is used to describe the defect analysis task;

[0019] Through the information generation model, perform an information adjustment task on the initial indication information according to the defect information and third indication information, and generate the first indication information, where the information adjustment task is used to correct the information defect, and the third indication information is used to describe the information adjustment task.

[0020] In a possible implementation, the device further includes a second acquisition unit, a second generation unit, a third generation unit, and a second adjustment unit:

[0021] The second acquisition unit is configured to acquire sample initial indication information, where the sample initial indication information has corresponding sample indication information and sample defect information, the sample defect information is used to describe the sample information defect existing in the sample initial indication information, and the sample indication information is the information obtained after correcting the sample information defect in the sample initial indication information;

[0022] The second generation unit is configured to, through an initial information generation model, perform the defect analysis task on the sample initial indication information according to the second indication information, and generate pending defect information corresponding to the sample initial indication information;

[0023] The third generation unit is configured to, through the initial information generation model, perform the information adjustment task on the sample initial indication information according to the sample defect information and the third indication information, and generate pending indication information;

[0024] The second adjustment unit is configured to adjust the corresponding old model parameters of the initial information generation model according to the difference between the pending defect information and the sample defect information, and the difference between the pending indication information and the sample indication information, to obtain the information generation model.

[0025] In a possible implementation, the first adjustment unit is specifically configured to:

[0026] Through the information generation model, according to the second indication information, the information to be classified, the first pending label information, and the first sample label information corresponding to the information to be classified, perform the defect analysis task on the initial indication information, and generate defect information corresponding to the initial indication information, where the first sample label information indicates that the accuracy of the information content corresponding to the information to be classified is greater than the first accuracy threshold.

[0027] In a possible implementation, the first acquisition unit is specifically configured to:

[0028] Acquire initial information to be classified;

[0029] Generate initial label information corresponding to the initial information to be classified and second accuracy information corresponding to the initial label information through a pre-trained classification model, where the initial label information is used to identify the information content corresponding to the initial information to be classified, and the second accuracy information is used to characterize the accuracy of the initial label information in identifying the information content;

[0030] Based on the accuracy characterized by the second accuracy information being less than the second accuracy threshold, the initial information to be classified is determined as the information to be classified, and the information generation model is used to jointly perform the information classification task with the classification model.

[0031] In a possible implementation manner, the apparatus further includes a third acquisition unit, a second determination unit, and a third adjustment unit:

[0032] The third acquisition unit is configured to acquire sample information, and the sample information has corresponding second sample label information, where the second sample label information is used to identify the information content corresponding to the sample information;

[0033] The second determination unit is configured to determine, through an initial classification model, third accuracy information corresponding to the sample information under multiple label information respectively, and determine one or more label information with the highest accuracy characterized by the corresponding third accuracy information as the second to-be-determined label information corresponding to the sample information, where the third accuracy information is used to characterize the accuracy of the corresponding label information in identifying the information content of the sample information;

[0034] The third adjustment unit is configured to adjust the model parameters corresponding to the initial classification model according to the difference between the second sample label information and the second to-be-determined label information to obtain the classification model.

[0035] In a possible implementation manner, the apparatus further includes a fourth acquisition unit, a third determination unit, a fourth determination unit, and a fourth generation unit:

[0036] The fourth acquisition unit is configured to acquire first information to be processed;

[0037] The third determination unit is configured to determine, through the classification model, a first label information and fourth accuracy information corresponding to the first information to be processed, where the fourth accuracy information is used to characterize the accuracy of the first label information in identifying the information content of the first information to be processed;

[0038] The fourth determination unit is configured to determine the first label information as the label information corresponding to the first information to be processed based on the accuracy characterized by the fourth accuracy information being not less than the second accuracy threshold;

[0039] The fourth generation unit is configured to, based on the accuracy characterized by the fourth accuracy information being less than the second accuracy threshold, perform the information classification task on the first information to be processed according to the indication information through the information generation model, and generate the label information corresponding to the first information to be processed.

[0040] In a possible implementation, the third determination unit is specifically configured to:

[0041] Perform information encoding on the first information to be processed, and extract feature information corresponding to the first information to be processed, where the feature information is used to characterize the information content corresponding to the first information to be processed;

[0042] Determine first label information corresponding to the first information to be processed and the fourth accuracy information according to the feature information;

[0043] The fourth generation unit is specifically configured to:

[0044] Perform the information classification task on the feature information according to the indication information, and generate label information corresponding to the first information to be processed.

[0045] In a possible implementation, the information to be classified is information corresponding to a first information field, and the first information field is any one of a plurality of information fields. The apparatus further includes a storage unit, a fifth acquisition unit, a fifth determination unit, and an execution unit:

[0046] The storage unit is configured to store the indication information as indication information corresponding to the first information field;

[0047] The fifth acquisition unit is configured to acquire second information to be processed;

[0048] The fifth determination unit is configured to determine a second information field corresponding to the second information to be processed, where the second information field is any one of the plurality of information fields;

[0049] The execution unit is configured to perform the information classification task on the second information to be processed through the information generation model according to the indication information corresponding to the second information field.

[0050] In a possible implementation, the information to be classified is any one of a plurality of information to be classified, and the plurality of information to be classified are all information corresponding to the first information field. The first adjustment unit is specifically configured to:

[0051] Adjust the initial indication information according to the first accuracy information respectively corresponding to the plurality of information to be classified, so as to obtain first indication information, and the accuracy of the label information obtained by the information generation model performing the information classification task on the plurality of information to be classified according to the first indication information is greater than the first accuracy threshold.

[0052] In a possible implementation, the first determination unit is specifically configured to:

[0053] Determine the semantic similarity between the first to-be-determined tag information and the information to be classified, where the semantic similarity is used to characterize the similarity between the information content corresponding to the information to be classified and the information content identified by the first to-be-determined tag information;

[0054] Determine the first accuracy information corresponding to the first to-be-determined tag information according to the semantic similarity, and the accuracy characterized by the first accuracy information is positively correlated with the semantic similarity.

[0055] In a possible implementation manner, the first determining unit is specifically configured to:

[0056] Obtain the first sample tag information corresponding to the information to be classified, where the accuracy of the information content corresponding to the first sample tag information is greater than the first accuracy threshold;

[0057] Determine the difference information corresponding to the first to-be-determined tag information according to the first sample tag information, where the difference information is used to characterize the difference between the first sample tag information and the first to-be-determined tag information;

[0058] Determine the first accuracy information corresponding to the first to-be-determined tag information according to the difference information, and the accuracy characterized by the first accuracy information is inversely correlated with the difference.

[0059] In a possible implementation manner, the first determining unit is specifically configured to:

[0060] Determine the sub-accuracy information respectively corresponding to the first to-be-determined tag information under the multiple information analysis methods through multiple information analysis methods, where the sub-accuracy information is used to characterize, through the corresponding information analysis method, the accuracy of the first to-be-determined tag information in identifying the information content;

[0061] Determine the first accuracy information corresponding to the first to-be-determined tag information according to the weights respectively corresponding to the multiple information analysis methods and the sub-accuracy information respectively corresponding to the first to-be-determined tag information under the multiple information analysis methods. The first information analysis method is any one of the multiple information analysis methods, and the weight corresponding to the first information analysis method is used to indicate the influence degree of the sub-accuracy information corresponding to the first to-be-determined tag information under the first information analysis method on the first accuracy information.

[0062] In a third aspect, an embodiment of the present application discloses a computer device, where the computer device includes a processor and a memory:

[0063] The memory is used to store a computer program and transmit the computer program to the processor;

[0064] The processor is configured to execute the information processing method according to any one of the first aspects based on the instructions in the computer program;

[0065] In a fourth aspect, an embodiment of the present application discloses a computer-readable storage medium for storing a computer program, where the computer program is configured to execute the information processing method according to any one of the first aspects;

[0066] In a fifth aspect, an embodiment of the present application discloses a computer program product including a computer program, which, when running on a computer device, causes the computer device to execute the information processing method according to any one of the first aspects.

[0067] It can be seen from the above technical solutions that the information generation model can perform an information classification task on the information to be classified through the description of the information classification task by the initial indication information, and obtain the first undetermined label information as the classification result. The first undetermined label information is the label attached by the information generation model to the information content of the information to be classified. Since the information generation model itself has the information generation ability, and whether this ability can be accurately applied to the information classification task depends on whether the guiding effect of the initial indication information on the information classification task is accurate. Therefore, in order to enable the information generation model to accurately perform the information classification task for the information to be classified, when the accuracy of the first undetermined label information is relatively low, the initial indication information can be adjusted until the accuracy of the label information obtained by performing the information classification task on the information to be classified based on the adjusted indication information is greater than the first accuracy threshold. In this way, by adjusting the indication information, the information generation model can use the description function of the indication information for the information classification task to generate accurate label information for the new information to be classified. Throughout the process, there is no need to adjust the model parameters corresponding to the information generation model. On the one hand, the information generation ability brought by the original model parameters is retained, avoiding the impairment of the model ability caused by the adjustment of the model parameters; on the other hand, compared with the model parameters, the amount of information of the indication information is less, so the adjustment efficiency of the indication information is higher. While ensuring that the information generation model has new information classification capabilities, the model training efficiency is improved, which is applicable to application scenarios where information classification tasks are rapidly iterated. Description of the Drawings

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0069] Figure 1Schematic diagram of an information processing method provided by an embodiment of the present application in an actual application scenario;

[0070] Figure 2 Flowchart of an information processing method provided by an embodiment of the present application;

[0071] Figure 3 Schematic diagram of an information processing method provided by an embodiment of the present application;

[0072] Figure 4 Schematic diagram of an information processing method provided by an embodiment of the present application;

[0073] Figure 5 Schematic diagram of an information processing method provided by an embodiment of the present application;

[0074] Figure 6 Schematic diagram of an information processing method provided by an embodiment of the present application;

[0075] Figure 7 Schematic diagram of an information processing method provided by an embodiment of the present application;

[0076] Figure 8 Flowchart of an information processing method provided by an embodiment of the present application in an actual application scenario;

[0077] Figure 9 Structural block diagram of an information processing device provided by an embodiment of the present application;

[0078] Figure 10 Structural diagram of a terminal provided by an embodiment of the present application;

[0079] Figure 11 Structural diagram of a server provided by an embodiment of the present application. Detailed implementation manners

[0080] The embodiments of the present application will be described below with reference to the accompanying drawings.

[0081] The information classification task is one of the key tasks in the data processing tasks executed by the model. Through the information classification task, various tag information for identifying the information content can be generated. The information classification task can include multiple types. For example, for image information, information classification can generate tag information for identifying the object type of the object shown in the image information. For text information, information classification can generate tag information for identifying the information content of the text information, etc.

[0082] In related technologies, information classification tasks are mainly implemented through classification models. During the training process, relevant personnel input information to be classified into the classification model to obtain the pending label information output by the classification model. Then, based on the difference between the pending label information and the sample label information as the accurate classification result, the corresponding model parameters of the classification model are adjusted so that the classification model can learn the model knowledge for information classification. If an information type that has not appeared during the training process appears, new training samples need to be constructed for the new information type, and the corresponding model parameters of the classification model are re-adjusted based on the new training samples so that the classification model can learn the ability to classify new information types.

[0083] However, since the model usually contains a large number of model parameters, the efficiency of adjusting the model parameters is low. On the one hand, a large amount of training resources are consumed, and on the other hand, the training time is long. It is difficult to provide a classification model that can accurately perform information classification tasks for new types of information in a short time, so it is difficult to be applicable to information classification tasks with constantly changing information types and unable to quickly respond to changing information classification requirements.

[0084] To solve the above technical problems, the present application provides an information processing method that uses an information generation model based on indication information to complete various information generation tasks to perform information classification tasks, and by adjusting the indication information corresponding to the information classification tasks, the information generation model can have the ability to classify new information to be classified. Since the indication information has less information volume and higher adjustment efficiency compared with the model parameters, the model training is more efficient in this way, consuming less training resources and time, and can quickly respond to constantly changing information classification requirements.

[0085] It can be understood that this method can be applied to a computer device, which is a computer device capable of information processing, such as a terminal device or a server. This method can be independently executed by the terminal device or the server, or can be applied to a network scenario where the terminal device and the server communicate and executed in cooperation by the terminal device and the server. Among them, the terminal device can be a mobile phone, a tablet computer, a laptop computer, a desktop computer, etc. The terminal device can also include a variety of virtual reality devices, such as augmented reality (AR) devices, such as AR glasses, AR screens, etc., and can also include virtual reality technology (VR) devices, such as head-mounted VR glasses, etc. The server can be understood as an application server or a Web server. In actual deployment, the server can be an independent server, a cluster server, or a cloud server, etc.

[0086] To facilitate the understanding of the technical solution provided by this application, next, an information processing method provided by this application will be introduced in combination with an actual application scenario.

[0087] See Figure 1 , Figure 1 which is a schematic diagram of an information processing method in an actual application scenario provided by an embodiment of this application. In this actual application scenario, the computer device can be a server 101 with information processing capabilities.

[0088] The server 101 can obtain a to-be-classified image as to-be-classified information. The information classification task for the to-be-classified image is used to generate label information that identifies the information content expressed by the to-be-classified image, such as an event label that can be used to describe the event occurring in the to-be-classified image. The initial indication information is used to describe this information classification task. For example, it can be "Please describe the event occurring in the image". The information generation model is a model that can perform various information generation tasks based on the indication information. For example, it can be a large language model, etc.

[0089] The server 101 can input the to-be-classified image and the initial indication information into the information generation model together. By analyzing the initial indication information, the information generation model can be guided to perform the information classification task described for the to-be-classified image and generate first tentative label information. To measure the execution effect of the information generation model for performing the information classification task on the to-be-classified image, the server 101 will determine the first accuracy information corresponding to the first tentative label information. This first accuracy information is used to characterize the accuracy of the information content identified by the first tentative label information.

[0090] The server 101 can adjust the initial indication information based on the accuracy characterized by the first accuracy information, so that the adjusted indication information can more accurately and reasonably describe the information classification task for the to-be-classified image, thereby guiding the information generation model to generate more accurate label information (i.e., label information whose accuracy of the characterized information content reaches the first accuracy threshold) for the to-be-classified image. For example, since the initial indication information does not define the scope of event analysis in the image, the information generation model will summarize events for each image area in the to-be-classified image, resulting in the event content identified by the first tentative label information being too messy to accurately summarize the actual event occurring in the to-be-classified image. Therefore, the indication information obtained after adjusting the initial indication information can be "Please describe the event occurring to the object that occupies a larger image area in the image", so that when generating label information, the information generation model can analyze the main event expressed by the to-be-classified image and obtain more accurate label information. Combining this indication information can enable the information generation model to perform a more accurate information classification task for the to-be-classified image used to express events.

[0091] As can be seen from the above, when training an information generation model to enable it to accurately perform an information classification task for analyzing events expressed in images, the model training can be completed by adjusting the indication information instead of adjusting the model parameters. Compared with the model parameters, the amount of information of the indication information is less, and the information structure is relatively simple. Therefore, the adjustment difficulty is lower and the adjustment efficiency is higher, enabling the information generation model to quickly master the information classification ability, and then being able to be quickly put into the information classification task scenario of event analysis for images.

[0092] Next, the technical solutions provided by this application will be introduced in detail with reference to the accompanying drawings.

[0093] See Figure 2 , Figure 2 which is a flowchart of an information processing method provided by an embodiment of this application. In this embodiment, the computer device can be any of the above computer devices with information processing capabilities. The method includes:

[0094] S201: Obtain the information to be classified and the initial indication information.

[0095] Among them, the information to be classified can be any one or more pieces of information that can be classified, that is, information that can generate label information for identifying its information content. This application can involve various information classification scenarios, and the information types of the information to be classified in different information classification scenarios can be different. For example, for image information, it can generate category label information for identifying the object categories of the objects included in the image information, generate event label information for identifying the event content represented by the image information, and for text information, it can also generate domain label information for identifying the information fields involved in the text information, etc., which is not limited here.

[0096] The information generation model with information generation capabilities is used in this application to perform the information classification task. This information generation model can perform diverse information generation tasks according to the description of the task by the indication information. The information classification task can be regarded as an information generation task for generating corresponding label information. For example, the information generation model can be various large language models (LLMs for short), etc., and the indication information can be the Prompt information provided to the large language model.

[0097] The initial indication information can be used to describe the information classification task, so that after the initial indication information is input into the information generation model, the model can be instructed to perform the information classification task.

[0098] S202: Through the information generation model, perform an information classification task on the information to be classified according to the initial indication information, and generate the first pending label information corresponding to the information to be classified.

[0099] The information generation model can perform an information classification task on the information to be classified according to the description function of the initial indication information for the information generation task, so as to generate first undetermined label information, which is used to identify the information content corresponding to the information to be classified.

[0100] S203: Determine the first accuracy information corresponding to the first undetermined label information.

[0101] In order to measure whether the information generation model can accurately perform the information classification task for the information to be classified, the computer device analyzes the accuracy of the representation of the information content by the first undetermined label information to obtain the first accuracy information, which is used to represent the accuracy of the first undetermined label information in identifying the information content. Among them, the determination method of the first accuracy information can include various types, which will be introduced in detail below and will not be elaborated here.

[0102] S204: Adjust the initial indication information according to the first accuracy information to obtain the first indication information.

[0103] In the above information generation model, the factors affecting the accuracy of the information classification task execution are mainly in two aspects. On the one hand, it is the accuracy and rationality of the description of the information classification task by the initial indication information. On the other hand, it is the information generation ability of the information generation model itself, and the latter is determined by the model parameters corresponding to the information generation model. In this application, the model parameters corresponding to the information generation model have been adjusted during the training process. Therefore, the information generation model itself has a relatively effective information generation ability. Therefore, in order to enable the information generation model to accurately perform the information classification task for the information to be classified, the computer device can adjust the initial indication information.

[0104] The computer device can continuously adjust the initial indication information according to the accuracy represented by the first accuracy information, and measure the accuracy of performing the information classification task by the indication information by continuously determining the accuracy information corresponding to the indication information during the adjustment process. Finally, the first indication information is obtained. The accuracy corresponding to the label information obtained by the information generation model performing the information classification task according to the first indication information is greater than the first accuracy threshold. The first accuracy threshold is used to measure whether the accuracy of performing the information classification task reaches a relatively high level. The first accuracy threshold can be characterized in various forms. For example, under different accuracy information determination methods, the first accuracy threshold can correspond to different accuracy information. If the accuracy represented by the accuracy information is greater than the first accuracy threshold, it can be explained that the accuracy of performing the information classification task is relatively high and meets the task execution requirements. Among them, the first accuracy threshold can be set based on actual needs. For example, when the requirement for model training efficiency is relatively high and the requirement for the execution accuracy of the information classification task is relatively low, a lower first accuracy threshold can be set; on the contrary, when the requirement for the execution accuracy of the information classification task is relatively high, a higher first accuracy threshold can be set.

[0105] Through the above information adjustment, the first indication information can accurately and reasonably describe the information classification task for the information to be classified. Therefore, the information generation model can be used to perform the information classification task according to the first indication information to achieve accurate information classification for information similar to the information to be classified.

[0106] As can be seen from the above, compared with the related art, the present application has the following technical effects in multiple aspects:

[0107] 1. The present application can enable the information generation model to use the description function of the indication information to generate accurate label information for new information to be classified by adjusting the indication information. The whole process does not require adjusting the model parameters corresponding to the information generation model. Compared with the model parameters, the indication information has less information volume, so the adjustment efficiency of the indication information is relatively high. While ensuring that the information generation model has new information classification capabilities, the model training efficiency is improved, which is applicable to application scenarios where the information classification task iterates rapidly.

[0108] 2. Since the model parameters of the information generation model are not adjusted, the information generation capabilities brought by the original model parameters are retained, avoiding damage to the model capabilities caused by adjusting the model parameters. Therefore, while having new information generation capabilities, the model can still accurately execute the original information generation tasks, improving the comprehensiveness of the information generation model.

[0109] Next, the technical solution provided by the present application will be introduced in detail with reference to the accompanying drawings.

[0110] First, the adjustment method for adjusting the indication information in this application will be introduced in detail. In a possible implementation manner, in order to enable the information generation model to efficiently and automatically adjust the initial indication information, the thinking mode of the human brain for information adjustment can be simulated to construct a thinking chain for adjusting the indication information for the information generation model.

[0111] When performing step S204, the computer device may execute steps S2041 - S2042 (not shown in the figure). Steps S2041 - S2042 are a possible implementation manner of step S204 and include:

[0112] S2041: Through the information generation model, perform a defect analysis task on the initial indication information according to the second indication information, and generate defect information corresponding to the initial indication information.

[0113] When the human brain corrects information, the thinking logic usually used is "analyze the reason for the problem in the information - modify the information based on the reason analysis". Based on this, the computer device can enable the information generation model to accurately complete the adjustment of the initial indication information based on this thinking mode. In this implementation manner, as Figure 3 shown, the adjustment of the initial indication information can be divided into a defect analysis task and an information adjustment task. The defect analysis task is used to analyze the defects in the indication information that cause inaccurate execution of the information classification task, and the information adjustment task is used to correct the information defects and generate the adjusted first indication information.

[0114] Similar to the information classification task, in order to enable the information generation model to execute the above two information generation tasks, the computer device will set corresponding indication information. Among them, the second indication information is used to describe the defect analysis task. Based on the description of the task by the second indication information, the information generation model can perform a defect analysis on the initial indication information and generate corresponding defect information. This defect information is used to describe the information defects existing in the initial indication information, and the information defects are the reasons for the accuracy represented by the first accuracy information being not greater than the first accuracy threshold.

[0115] For example, in Figure 1 , the defect information corresponding to the initial indication information may be "the scope of analyzing the image content is not limited", which is the reason for the information generation model to perform event analysis on each image area in the image to be classified, resulting in the image events identified by the generated label information being too messy and having low accuracy.

[0116] S2042: Through the information generation model, perform an information adjustment task on the initial indication information according to the defect information and the third indication information, and generate the first indication information.

[0117] The third indication information is used to describe an information adjustment task, and the information adjustment task is used to correct information defects. After inputting the defective information and the third indication information into the information generation model, the information generation model can learn that it is necessary to correct the information defects identified based on the defective information in the initial indication information, and finally obtain the corrected first indication information, which is the initial indication information after correcting the above-mentioned information defects.

[0118] For example, in Figure 1 , the initial indication information is "Please describe the event that occurred in the image", and the defective information is "There is no limit to the scope of analyzing the image content". When performing information adjustment based on this defective information, the information generation model will add relevant information for limiting the scope of image content analysis to the initial indication information. The adjusted indication information is "Please describe the event that occurred to the object that occupies a relatively large area in the image", and "the object that occupies a relatively large area in the image" is the information for correcting the information defect.

[0119] In this way, the process of adjusting the indication information can be split into multiple thinking processes, enabling the information generation model to simulate the thinking logic of the human brain to adjust the indication information, thereby making the information adjustment process more accurate, reasonable, and logical, enabling the information generation model to more quickly analyze how to adjust the indication information to more accurately describe the information classification task, and then being able to generate the adjusted indication information more efficiently and accurately, further improving the model training efficiency.

[0120] In the above model training process, whether the information generation model can accurately perform the defect analysis task and the information adjustment task is the key to accurately adjusting the indication information. Based on this, in a possible implementation manner, the computer device can strengthen the ability of the information generation model to perform the above information generation task through a pre-training process.

[0121] The computer device can first obtain sample initial indication information. The sample initial indication information can be any indication information used to describe any information generation task. The sample initial indication information has corresponding sample indication information and sample defective information. The sample defective information is used to describe the sample information defects existing in the sample initial indication information, and the sample indication information is the information obtained after correcting the sample information defects in the sample initial indication information. That is, if the model can accurately perform the defect analysis task for the sample initial indication information, it should generate defective information that is relatively close to the sample defective information; if the model can accurately perform the information adjustment task for the sample initial indication information based on the sample defective information, it should generate indication information that is relatively close to the sample indication information.

[0122] The computer device can generate an initial information generation model, perform a defect analysis task on the sample initial indication information according to the second indication information, and generate the to-be-determined defect information corresponding to the sample initial indication information. Moreover, the computer device can generate an initial information generation model, perform an information adjustment task on the sample initial indication information according to the sample defect information and the third indication information, and generate the to-be-determined indication information. The reason for having the model perform the information adjustment task based on the sample defect information rather than the generated to-be-determined defect information is that since the sample defect information is accurate defect information, the accuracy of the to-be-determined indication information is only affected by the execution ability of the initial information generation model itself for the information adjustment task, reducing the influence of the information input side on the accuracy of the to-be-determined indication information. Thus, the ability of the initial information generation model to perform the information adjustment task can be analyzed more accurately in combination with the to-be-determined indication information.

[0123] Combining the above content, it can be seen that the difference between the to-be-determined defect information and the sample defect information can accurately represent the ability of the initial information generation model to perform the defect analysis task, and the difference between the to-be-determined indication information and the sample indication information can accurately represent the ability of the initial information generation model to perform the information adjustment task. Thus, as Figure 5 shown, the computer device can adjust the corresponding old model parameters of the initial information generation model according to the difference between the to-be-determined defect information and the sample defect information, and the difference between the to-be-determined indication information and the sample indication information. During this parameter adjustment process, the differences between the two-dimensional parameters can be continuously reduced, so that the initial information generation model can learn how to accurately perform the defect analysis task and the information adjustment task, and finally obtain an information generation model, enabling the information generation model to have the ability to accurately perform the defect analysis task and the information adjustment task.

[0124] In this way, the computer device can, through parameter adjustment, enable the model to have the above-mentioned defect analysis ability and information adjustment ability. During the subsequent process of training the information classification ability, the model can have the ability to accurately adjust the indication information, providing an accurate information basis for the model to adjust the indication information. At the same time, in the actual application process, there is no need to adjust the model parameters anymore, so it will not affect the training efficiency of the training process for the information classification ability.

[0125] Of course, in addition to adjusting the model parameters, during the training process, the computer device can also further adjust the second indication information and the third indication information to enable them to better describe the defect analysis task and the information adjustment task, which will not be elaborated here.

[0126] In a possible implementation manner, in order to enable the information generation model to more accurately perform the defect analysis task, the computer device can also enrich the types of information input into the information generation model.

[0127] When performing step S2041, the computer device may execute step S20411 (not shown in the figure). Step S20411 is a possible implementation of step S2041 and includes:

[0128] S20411: Through the information generation model, perform a defect analysis task on the initial indication information according to the second indication information, the information to be classified, the first undetermined label information, and the first sample label information corresponding to the information to be classified, and generate defect information corresponding to the initial indication information.

[0129] Among them, the second indication information plays a role in instructing the information generation model to perform the defect analysis task. The first sample label information indicates that the accuracy of the information content corresponding to the information to be classified is greater than the first accuracy threshold, that is, the first sample label information is the accurate label information corresponding to the information to be classified. Thus, through the information to be classified and the first sample label information, the information generation model can understand the accurate information classification method for the information to be classified. At the same time, based on the difference between the first sample label information and the first undetermined label information, the information generation model can understand the difference between the current information classification result and the required information classification result, and then the information generation model can analyze the reason for this difference caused by the initial indication information, and further can more accurately and reasonably analyze the information defects existing in the initial indication information.

[0130] It can be understood that in the related art, although the training method of adjusting model parameters to make the model have information classification ability cannot achieve efficient model training, the trained classification model can perform relatively effective information classification for the information types that have appeared during the training process. Based on this, in a possible implementation, the computer device can combine the pre-trained classification model and the above information generation model to jointly perform the information classification task, and make full use of the information classification knowledge possessed by the classification model.

[0131] In this implementation, when performing step S201, the computer device may execute steps S2011 - S2013 (not shown in the figure). Steps S2011 - S2013 are a possible implementation of step S201 and include:

[0132] S2011: Obtain the initial information to be classified.

[0133] Among them, the initial information to be classified can be any one or more pieces of information that can be classified.

[0134] S2012: Generate the initial label information corresponding to the initial information to be classified and the second accuracy information corresponding to the initial label information through the pre-trained classification model.

[0135] The classification model is a traditional classification model obtained through pre-training and capable of performing information classification. Its input is the information to be classified, and its output is the classified label information and the accuracy information corresponding to the label information. This information classification process does not require the participation of indication information. Among them, the accuracy information is usually the information generated by the classification model during the generation of label information. The ability of the classification model is to analyze the probabilities of the information corresponding to each label information and finally output one or more label information with the highest probabilities. The accuracy information is the information determined based on this probability.

[0136] Taking the result of the classification model for the initial information to be classified as an example, the initial label information is used to identify the information content corresponding to the initial information to be classified, and the second accuracy information can be used to characterize the accuracy of the initial label information in identifying the information content.

[0137] S2013: Based on the accuracy characterized by the second accuracy information being less than the second accuracy threshold, determine the initial information to be classified as the information to be classified.

[0138] The processing device can preset a second accuracy threshold, which is used to measure whether the label information generated by the classification model has sufficient accuracy. If the accuracy characterized by the second accuracy information is not less than the second accuracy threshold, it means that the classification model can perform accurate information classification for the initial information to be classified without the help of the information generation model. When performing information classification for information similar to the initial information to be classified, the classification model itself can complete the information classification task. Therefore, the information generation model does not need to learn the ability to perform information classification for this part of the information.

[0139] Based on the accuracy characterized by the second accuracy information being less than the second accuracy threshold, it means that the classification model cannot perform accurate information classification for the initial information to be classified. At this time, the computer device can determine the initial information to be classified as the information to be classified, so that the information generation model can learn the ability to perform accurate information classification for the information to be classified through the above process, thereby making up for the deficiency of the information classification ability of the classification model. The information generation model is used to jointly perform the information classification task with the classification model.

[0140] In this way, on the one hand, the computer device does not need the information generation model to adjust the indication information for each piece of information to be classified, but only needs to learn the information that the classification model cannot accurately classify. While ensuring the versatility of the information classification function, it reduces the training volume required by the model and further improves the model training efficiency. On the other hand, through the combination of the classification model and the information generation model, the information classification ability learned by the classification model during the pre-training process can be fully utilized. At the same time, during the model training process, it is not necessary to adjust the model parameters of the classification model, effectively retaining the model knowledge of the classification model, thereby further improving the accuracy of information classification.

[0141] Specifically, the above classification model can be trained through the following training method:

[0142] For example Figure 5 As shown, the computer device can first obtain sample information. The sample information can be any information that can be classified. The sample information has corresponding second sample label information. The second sample label information is used to identify the information content corresponding to the sample information. The second sample label information is the accurate label information corresponding to the sample information. That is, if the model can accurately classify the sample information, it should obtain label information that is relatively similar to the second sample label information. The higher the similarity, the more accurate the information classification.

[0143] Then, the computer device can determine the third accuracy information corresponding to the sample information under multiple label information through the initial classification model, and determine one or more label information with the highest accuracy represented by the corresponding third accuracy information as the second pending label information corresponding to the sample information. Among them, the initial classification model can be any model that can perform information classification tasks. The information classification principle of the initial classification model is based on the input information, analyzes the probability of the information corresponding to each label information, and outputs one or more label information with the highest probability as the information classification result. The number of output label information can be set according to actual needs and is not limited here. Among them, the third accuracy information is used to represent the accuracy of the corresponding label information in identifying the information content of the sample information, that is, the probability that the corresponding label information is the label information corresponding to the sample information.

[0144] As can be seen from the above, the difference between the second sample label information and the second to-be-determined label information can characterize the accuracy of the initial classification model in classifying sample information. This accuracy is inversely correlated with the difference. Based on this, the computer device can adjust the model parameters corresponding to the initial classification model according to the difference between the second sample label information and the second to-be-determined label information, so that the difference between the second to-be-determined label information determined by it and the second sample label information gradually decreases. In this process, the initial classification model can learn how to accurately analyze the accuracy information corresponding to sample information under multiple label information, and then learn how to accurately classify sample information to obtain the above classification model.

[0145] As can be seen from this training process, through the pre-training process of adjusting model parameters, the classification model can have the ability to accurately analyze label information for the information types that appear in the training process. Thus, on the one hand, it can accurately perform information classification for the information types that appear in the training process. On the other hand, it can determine the to-be-classified information that actually needs to be learned by the information generation model through the accurate accuracy analysis ability of the classification model, which helps to screen the to-be-classified information required for the training process of the information generation model and further improves the training efficiency.

[0146] Next, the process of performing an information classification task by combining the classification model and the information generation model will be introduced in detail.

[0147] First, the computer device can obtain the first to-be-processed information, and the first to-be-processed information can be any information that needs to be classified. As Figure 6 shown, the computer device can first determine the first label information and the fourth accuracy information corresponding to the first to-be-processed information through the classification model. The fourth accuracy information is used to characterize the accuracy of the first label information in identifying the information content of the first to-be-processed information. The method of analyzing accuracy information through the classification model has been introduced in detail above and will not be elaborated here.

[0148] Then, the computer device can compare the accuracy characterized by the fourth accuracy information with the above second accuracy threshold. Based on the fact that the accuracy characterized by the fourth accuracy information is not less than the second accuracy threshold, it indicates that the classification model itself can already accurately classify the first to-be-processed information without the help of the information generation model. At this time, the computer device can directly determine the first label information as the label information corresponding to the first to-be-processed information, that is, determine the first label information as the classification result of classifying the first to-be-processed information.

[0149] If the accuracy based on the fourth accuracy information representation is less than the second accuracy threshold, it indicates that the classification model cannot accurately classify the first information to be analyzed and requires the assistance of the information classification ability of the information generation model. At this time, the computer device can use the information generation model to perform an information classification task on the first information to be processed according to the indication information, and generate label information corresponding to the first information to be processed, that is, determine the classification result of the information classification of the first information to be processed through the information generation model.

[0150] Through this information classification method, on the one hand, the computer device can make full use of the effective information classification knowledge learned by the classification model through the pre-training process, so that the classification model can generate accurate information classification results for the information that it can accurately classify; on the other hand, for the information that the classification model cannot accurately classify, the fast learning ability of the information generation model can be used to learn the model knowledge for classifying this part of the information, so as to make up for the deficiency of the information classification ability of the classification model and improve the comprehensiveness and versatility of the information classification function. The process of information classification through this method can be shown by the following formula:

[0151] T o ′ utput = T output + ΔL

[0152] Among them, T o ′ utput is the set of label information that can be accurately analyzed through the above method, T output is the set of label information that can be accurately analyzed by the classification model, and ΔL is the set of new label information that can be accurately analyzed through the information generation model.

[0153] It can be understood that when the model processes information, it usually first encodes the information (for example, extracts information through a convolutional layer) to extract the feature information used to represent the information content, and then performs corresponding information processing based on the feature information. Based on this, in a possible implementation manner, in the scenario where this application involves two models for information classification, the computer device can further improve the information processing efficiency in information encoding.

[0154] In this implementation manner, see Figure 7, when determining the first label information and the fourth accuracy information corresponding to the first piece of information to be processed through the classification model, the classification model will first perform information encoding on the first piece of information to be processed and extract the feature information corresponding to the first piece of information to be processed. The feature information is used to represent the information content corresponding to the first piece of information to be processed. When performing information encoding, information that is less helpful for information classification in the first piece of information to be processed will be screened out, and information that is more helpful for information classification will be retained, so as to make the subsequent information classification process more efficient.

[0155] Then, the classification model will determine the first label information and the fourth accuracy information corresponding to the first piece of information to be processed based on the feature information. When triggering information classification through the information generation model due to the accuracy represented by the fourth accuracy information being less than the second accuracy threshold, since the feature information that can accurately represent the information content of the first piece of information to be processed has been generated and this feature information is helpful for information classification, the information generation model does not need to perform information encoding on the first piece of information to be processed again. When performing the information classification task on the first piece of information to be processed according to the indication information and generating the label information corresponding to the first piece of information to be processed, the information generation model can directly perform the information classification task on the feature information according to the indication information and generate the label information corresponding to the first piece of information to be processed.

[0156] Thus, it can be seen that in the entire information classification process, although two models are involved in information classification, there is at most only one information encoding process, thereby further simplifying the information classification process combining multiple models while ensuring the information classification effect and improving the information classification efficiency.

[0157] In addition, it can be understood that information in different information fields usually has large differences in information content. Therefore, the classification methods for information classification of information in different information fields usually vary greatly. For example, for the information classification task of identifying the object types corresponding to the objects included in an image, the classification methods for the two different information classification tasks of identifying the species of animals and identifying the vehicle models are quite different. If the model is mainly trained based on animal images during the training process, it is difficult to learn the model knowledge for information classification of vehicle images.

[0158] Based on this, in a possible implementation manner, in order to further improve the accuracy of information classification through the information generation model, the computer device can analyze the information fields corresponding to the information involved in the training process of the information generation model and perform targeted information classification on the information in these information fields during the application process.

[0159] In this implementation manner, the information to be classified is the information corresponding to the first information field, and the first information field is any one of multiple information fields. After adjusting the initial indication information to obtain the indication information through the above-mentioned manner, the computer device can store the indication information as the indication information corresponding to the first information field. That is, when adjusting the indication information based on the information to be classified applied during the training process, the computer device will bind the obtained indication information to the information field corresponding to the information to be classified, so as to obtain the indication information corresponding to each information field respectively. From the above training manner, it can be seen that the indication information corresponding to the information field can be used to accurately classify the information under this information field. That is, the indication information corresponding to each information field can match the information classification method of the information under this information field.

[0160] During the model application process, the computer device can obtain the second information to be processed, and the second information to be processed can be any information that needs to be classified. Then, the computer device can first determine the second information field corresponding to the second information to be processed, and the second information field can be any one of the above-mentioned multiple information fields. Among them, there are various ways to analyze the information field corresponding to the information. For example, a dedicated domain analysis model can be trained to perform information field analysis, or the similarity between the second information to be processed and the information features of the information under multiple information fields can be compared for analysis, which is not limited here.

[0161] Finally, when performing information classification, the computer device can use the information generation model to perform the information classification task on the second information to be processed according to the indication information corresponding to the second information field. Thus, the execution manner of this information classification task can conform to the information characteristics corresponding to the second information field, and further, the accuracy of the execution of the information classification task can be improved, and more accurate label information can be obtained. Through this method, the computer device can utilize different indication information adjusted for different information fields, enabling the information generation model to perform the information classification task on the information of different information fields through different information classification methods, making the execution manner of the information classification task more conform to the information characteristics of different information fields, and further improving the overall accuracy of information classification.

[0162] Among them, in order to make the indication information corresponding to each information field more suitable for classifying the information corresponding to the information field, in a possible implementation manner, the computer device can adjust the indication information for each information field through multiple information to be classified.

[0163] In this implementation manner, taking the above-mentioned information to be classified as an example, the information to be classified can be any one of multiple pieces of information to be classified, and the multiple pieces of information to be classified are all information corresponding to the first information field. When performing step S204, the computer device may execute step S2043 (not shown in the figure), and step S2043 is a possible implementation manner of step S204, including:

[0164] S2043: Adjust the initial indication information according to the first accuracy information respectively corresponding to multiple pieces of information to be classified, to obtain the first indication information.

[0165] When adjusting the initial indication information to obtain the first indication information corresponding to the first information field, the computer device may perform the adjustment in combination with multiple pieces of information to be classified in the first information field, so that the adjusted first indication information can perform relatively accurate information classification for multiple pieces of information to be classified in the first information field. That is, the adjusted first indication information satisfies that the accuracy corresponding to the label information obtained by the information generation model performing the information classification task on multiple pieces of information to be classified according to the first indication information is greater than the first accuracy threshold. Thus, during the adjustment process, the description manner of the first indication information for the information classification task can be made more in line with the common information characteristics in the first information field, and further, the first indication information can be used more accurately for information classification of the information in the first information field, reducing the probability of incorrect adjustment of the indication information due to the contingency and particularity of a single piece of information to be classified, and further improving the accuracy of information classification.

[0166] As mentioned above, the present application can analyze the accuracy of the label information indicating the information content in various ways. Next, multiple accuracy analysis methods involved in the present application will be combined.

[0167] The first one: Semantic relevance analysis

[0168] Since the role of the first pending label information is to identify the information content of the information to be classified, the more accurate the first pending label information is, the closer the information content identified by the first pending label information should be to the information content corresponding to the information to be classified. And the semantic features corresponding to the information are one of the accurate features characterizing the information content. Based on this, in a possible implementation manner, the computer device can analyze whether the first pending label information is accurate by analyzing the semantic relevance between the first pending label information and the information to be classified.

[0169] When performing step S203, the computer device may execute steps S2031 - S2032 (not shown in the figure), and steps S2031 - S2032 are a possible implementation manner of step S203, including:

[0170] S2031: Determine the semantic similarity between the first undetermined label information and the information to be classified.

[0171] Among them, the semantic similarity is used to characterize the similarity between the information content corresponding to the information to be classified and the information content identified by the first undetermined label information. There are various analysis methods for information semantics. For example, the semantic features corresponding to the information can be extracted through multiple semantic feature extraction models. However, by analyzing the feature similarity between two semantic features to analyze the semantic similarity, this analysis process can be shown by the following formula:

[0172] Rel(l i ) = sim(l i , X)

[0173] Among them, l i is the first undetermined label information, Rel(l i ) is the semantic similarity corresponding to the first undetermined label information, sim is the calculation formula of feature similarity, and X is the information to be classified.

[0174] S2032: Determine the first accuracy information corresponding to the first undetermined label information according to the semantic similarity.

[0175] As can be seen from the above, the greater the semantic similarity between the first undetermined label information and the information to be classified, the more similar the information content identified by the first undetermined label information is to the information content corresponding to the information to be classified. Furthermore, it can be shown that the role of the first undetermined label information in representing the content of the information to be classified is more accurate. Therefore, the computer device can determine the first accuracy information according to the semantic similarity, and the accuracy characterized by the first accuracy information is positively correlated with the semantic similarity.

[0176] The second type: Difference analysis with the sample label

[0177] In another possible implementation, when performing step S203, the computer device can execute steps S2033 - S2035 (not shown in the figure). Steps S2033 - S2035 are a possible implementation of step S203, including:

[0178] S2033: Obtain the first sample label information corresponding to the information to be classified.

[0179] Among them, the accuracy of the first sample label information in identifying the information content corresponding to the information to be classified is greater than the first accuracy threshold, that is, the first sample label information is the accurate label information corresponding to the information to be classified. If the information generation model can accurately perform the information classification task for the information to be classified, it should generate label information that is relatively close to the first sample label information.

[0180] S2034: Determine the difference information corresponding to the first to-be-determined label information according to the first sample label information.

[0181] Among them, the difference information is used to characterize the difference between the first sample label information and the first to-be-determined label information. This difference information can be generated through various difference analysis methods, which are not limited here. For example, the computer device can analyze the information difference by means of coverage rate. That is, the first sample label information can include multiple label information, the first to-be-determined label information can include one or more label information, and the higher the coverage rate of the first to-be-determined label information for the first sample label information, the smaller the difference between the two label information. Its analysis method can be shown in the following formula:

[0182]

[0183] Among them, Cov(l i ) is the coverage rate, l i is the first to-be-determined label information, D new (l i ) is the set of label information included in the first to-be-determined label information, D total is the set of label information included in the first sample label information. The larger Cov(l i ) is, the closer the first to-be-determined label information is to the first sample label information.

[0184] S2035: Determine the first accuracy information corresponding to the first to-be-determined label information according to the difference information.

[0185] Based on the above, it can be seen that the closer the first to-be-determined label information is to the first sample label information, the higher the accuracy of the information content identification of the first to-be-determined label information for the information to be classified. Based on this, the computer device can determine the first accuracy information according to the difference information. The accuracy characterized by this first accuracy information is inversely correlated with the difference.

[0186] It should be emphasized that the above two accuracy analysis methods can be applied separately or in combination to jointly analyze the accuracy corresponding to the label information. In one possible implementation, in order to further improve the accuracy analysis accuracy, the computer device can combine multiple accuracy analysis methods to determine the first accuracy information corresponding to the first to-be-determined label information.

[0187] When executing step S203, the computer device can execute steps S2036 - S2037 (not shown in the figure). Steps S2036 - S2037 are a possible implementation of step S203, including:

[0188] S2036: Determine the sub-accuracy information corresponding to the first undetermined label information under various information analysis methods through various information analysis methods.

[0189] The various information analysis methods can be any number of analysis methods that can be used to analyze the accuracy of the information content identified by the label information. For example, the above two analysis methods can be included. The sub-accuracy information is used to characterize the accuracy of the first undetermined label information in identifying the information content through the corresponding information analysis method. Taking the first information analysis method as an example, the first information analysis method can be any one of the various information analysis methods, and the sub-accuracy information corresponding to the first information analysis method is used to characterize the accuracy of the first undetermined label information in identifying the information content through the first information analysis method.

[0190] S2037: Determine the first accuracy information corresponding to the first undetermined label information according to the weights corresponding to the various information analysis methods and the sub-accuracy information corresponding to the first undetermined label information under the various information analysis methods.

[0191] The computer device can combine the sub-accuracy information corresponding to the various information analysis methods to determine the above first accuracy information. Among them, in order to accurately combine the various sub-accuracy information to determine a higher-precision first accuracy information, the computer device can set a corresponding weight for each information analysis method. Taking the first information analysis method as an example, the weight corresponding to the first information analysis method is used to indicate the influence degree of the sub-accuracy information corresponding to the first undetermined label information under the first information analysis method on the first accuracy information. Generally, the larger the weight, the greater the influence degree on the first accuracy information. Through this method, on the one hand, it is possible to combine multiple analysis methods to improve the accuracy analysis precision and the comprehensiveness of the accuracy analysis; on the other hand, it is possible to reasonably combine the analysis results of multiple analysis methods through the weights corresponding to the various analysis methods, so that the more accurate and effective analysis methods can have a greater influence degree, and the analysis methods with lower accuracy can have a smaller influence degree. Among them, the weights corresponding to the various analysis methods can be adjusted based on actual needs and model training effects, which are not limited here.

[0192] The determination method of this accuracy information can be shown by the following formula:

[0193]

[0194] Among them, L is the first undetermined label information, L(L) is the first accuracy information, and can also be regarded as the loss corresponding to model training. Rel is the sub-accuracy information obtained through semantic relevance analysis, α 1 is the weight corresponding to the semantic relevance analysis method, Cov is the sub-accuracy information obtained through coverage analysis, α2 is the weight corresponding to the coverage analysis method, Comp is the analysis result of the overall accuracy information corresponding to multiple pieces of information to be classified involved in the training process, and α 3 is the weight corresponding to the overall accuracy analysis method. The smaller the loss, the higher the accuracy corresponding to the first piece of information to be determined.

[0195] To facilitate the understanding of the technical solution provided by this application, next, a practical application scenario will be combined to introduce the information processing method provided by this application.

[0196] See Figure 8 , Figure 8 is a flowchart of an information processing method in a practical application scenario provided by an embodiment of this application. In this practical application scenario, the computer device can be any of the above computer devices with information processing capabilities. The method includes:

[0197] S801: During the application process of the classification model, collect the information to be classified with low classification accuracy.

[0198] The classification model is any model that can be used for information classification obtained through pre-training. During the application process of the classification model, the computer device can analyze the classification effect of the classification model in real time and collect the information to be classified that the classification model cannot accurately classify. These pieces of information to be classified are the information that requires new information classification capabilities for processing.

[0199] S802: Determine the information field corresponding to the information to be classified.

[0200] S803: Through the information generation model, use the initial indication information to perform an information classification task on multiple pieces of information to be classified corresponding to the first information field, and obtain the information to be determined with a label.

[0201] When obtaining the indication information that can be used to accurately describe the information classification task through model training, the computer device can divide multiple pieces of information to be classified based on the information field, and adjust and obtain the indication information used for information classification processing of the information in this information field based on multiple pieces of information to be classified corresponding to the same information field. Taking the first information field as an example, the computer device can first perform an information classification task on multiple pieces of information to be classified corresponding to the first information field through the information generation model, using the initial indication information, and obtain the information to be determined with a label.

[0202] This information generation process can be shown by the following formula:

[0203] L = M(p gen ||X||T)

[0204] Among them, L is the to-be-determined tag information generated, which can include one or more tag information, M is the information generation model, such as a large language model that can be pre-trained, etc., and p gen is the initial indication information, X is the information to be classified, T is the preset tag information set, which is used to indicate the information generation model to generate the to-be-determined tag information from this set. By setting this tag information set, the information generation model can be made to fit faster, reducing the difficulty of adjusting the indication information, so as to improve the model training efficiency. During the model application process, the computer device can also provide this tag information set, enabling the information generation model to generate tag information from the range of tag information delimited by this set.

[0205] S804: Determine the accuracy information corresponding to the to-be-determined tag information.

[0206] The determination method of the accuracy information can include various types, which have been introduced in the above content and will not be elaborated here.

[0207] S805: Adjust the initial indication information according to the accuracy information corresponding to the to-be-determined tag information to obtain the indication information corresponding to the first information field.

[0208] The computer device can first perform a defect analysis task on the initial indication information according to the second indication information through the information generation model to obtain the defect information corresponding to the initial indication information. This process can be shown by the following formula:

[0209] F = M(p fb ||X||T||L)

[0210] Among them, F is the defect information, which is used as a feedback signal to adjust the initial indication information. p fb is the second indication information, which is used to describe the defect analysis task, such as can be used to define the analysis dimension for analyzing defects.

[0211] Then, the computer device can perform an information adjustment task on the initial indication information according to the defect information through the information generation model to obtain the indication information corresponding to the first information field. This information adjustment process can be shown as follows:

[0212] L ′ = M(p refine ||X||T||L||F)

[0213] Among them, p refine is the third indication information, and L ′ is the tag information generated based on the adjusted indication information. During the adjustment process, the computer device can continuously judge the accuracy of the newly generated tag information to iteratively adjust the indication information until the accuracy of the tag information generated based on the indication information reaches the required accuracy threshold.

[0214] S806: Obtain the information to be processed.

[0215] The information to be processed can be any information that needs to be classified.

[0216] S807: Determine the first tag information corresponding to the information to be processed through the classification model.

[0217] S808: Determine whether the accuracy corresponding to the first tag information is less than the second accuracy threshold.

[0218] If it is not less than, it means that the first tag information is already relatively accurate tag information, and the computer device can execute step S809; if it is less than, it means that the accuracy of the first tag information is low, and it is difficult to accurately classify the information to be processed only through the classification model. At this time, the computer device can execute step S810 to perform information classification through the information generation model.

[0219] S809: Determine the tag information corresponding to the information to be processed as the first tag information.

[0220] S810: Determine the information field corresponding to the information to be processed.

[0221] S811: Through the information generation model, according to the indication information corresponding to the information field of the information to be processed, perform an information classification task on the information to be processed, and generate the tag information corresponding to the information to be processed.

[0222] From the above, it can be seen that this application has the following technical effects in multiple aspects:

[0223] 1. This application can adjust the indication information, enabling the information generation model to use the description function of the indication information for the information classification task to generate accurate tag information for new information to be classified. The entire process does not require adjusting the model parameters corresponding to the information generation model. Compared with model parameters, the indication information has less information, so the adjustment efficiency of the indication information is higher. While ensuring that the information generation model has new information classification capabilities, it improves the model training efficiency and is suitable for application scenarios where information classification tasks are rapidly iterated.

[0224] 2. Since the model parameters of the information generation model are not adjusted, the information generation capabilities brought by the original model parameters are retained, avoiding the impairment of model capabilities caused by adjusting model parameters. Thus, while having new information generation capabilities, the model can still accurately execute the original information generation tasks, improving the comprehensiveness of the information generation model.

[0225] 3. The present application can refine the adjustment process of automatically indicating information adjustment through the information generation model by means of the information adjustment method of the chain of thought, making the adjustment of the indication information more reasonable and accurate.

[0226] 4. The present application can jointly execute the information classification task by combining the classification model obtained through pre-training and the information generation model. On the one hand, it can make full use of the relatively accurate model knowledge in the classification model to perform information classification with higher accuracy for a part of the information. On the other hand, it can use the information generation model to perform more comprehensive and accurate information classification for another part of the information, improving the versatility of the information classification ability.

[0227] 5. The present application can train different indication information for the information in different information fields, so that the information classification task performed by the information generation model can better fit the information characteristics of different information fields, further improving the accuracy of information classification.

[0228] By comparing the information classification model training method in the existing solution with the training method in the present application, the results shown in the following table can be obtained:

[0229]

[0230] As can be seen from the table, the training method of the present application for the model requires less human resources and lower training costs, and can significantly reduce the model cost on the premise of ensuring that the model has accurate information classification ability.

[0231] Based on the information processing method provided in the above embodiments, the present application further provides an information processing device. Refer to Figure 9 , Figure 9 which is a structural block diagram of an information processing device provided in an embodiment of the present application. The device 900 includes a first acquisition unit 901, a first generation unit 902, a first determination unit 903, and a first adjustment unit 904:

[0232] The first acquisition unit 901 is configured to acquire the information to be classified and the initial indication information, where the initial indication information is used to describe the information classification task;

[0233] The first generation unit 902 is configured to perform the information classification task on the information to be classified according to the initial indication information through the information generation model, and generate first undetermined label information corresponding to the information to be classified, where the first undetermined label information is used to identify the information content corresponding to the information to be classified;

[0234] The first determination unit 903 is configured to determine first accuracy information corresponding to the first undetermined label information, where the first accuracy information is used to characterize the accuracy of the first undetermined label information in identifying the information content;

[0235] The first adjustment unit 904 is configured to adjust the initial indication information according to the first accuracy information to obtain first indication information, and the accuracy corresponding to the label information obtained by the information generation model performing the information classification task according to the first indication information is greater than a first accuracy threshold, and the information generation model is configured to perform the information classification task according to the first indication information.

[0236] In a possible implementation manner, the first adjustment unit 904 is specifically configured to:

[0237] Through the information generation model, perform a defect analysis task on the initial indication information according to second indication information to generate defect information corresponding to the initial indication information, where the defect information is used to describe the information defect existing in the initial indication information, and the information defect is the reason for the accuracy represented by the first accuracy information not being greater than the first accuracy threshold, and the second indication information is used to describe the defect analysis task;

[0238] Through the information generation model, perform an information adjustment task on the initial indication information according to the defect information and third indication information to generate the first indication information, where the information adjustment task is used to correct the information defect, and the third indication information is used to describe the information adjustment task.

[0239] In a possible implementation manner, the device further includes a second acquisition unit, a second generation unit, a third generation unit, and a second adjustment unit:

[0240] The second acquisition unit is configured to acquire sample initial indication information, where the sample initial indication information has corresponding sample indication information and sample defect information, and the sample defect information is used to describe the sample information defect existing in the sample initial indication information, and the sample indication information is the information obtained by correcting the sample information defect in the sample initial indication information;

[0241] The second generation unit is configured to, through an initial information generation model, perform the defect analysis task on the sample initial indication information according to the second indication information to generate pending defect information corresponding to the sample initial indication information;

[0242] The third generation unit is configured to, through the initial information generation model, perform the information adjustment task on the sample initial indication information according to the sample defect information and the third indication information to generate pending indication information;

[0243] The second adjustment unit is configured to adjust the corresponding old model parameters of the initial information generation model according to the difference between the to-be-determined defect information and the sample defect information, and the difference between the to-be-determined indication information and the sample indication information, so as to obtain the information generation model.

[0244] In a possible implementation manner, the first adjustment unit 904 is specifically configured to:

[0245] Through the information generation model, perform a defect analysis task on the initial indication information according to the second indication information, the to-be-classified information, the first to-be-determined label information, and the first sample label information corresponding to the to-be-classified information, and generate defect information corresponding to the initial indication information, where the first sample label information indicates that the accuracy of the information content corresponding to the to-be-classified information is greater than the first accuracy threshold.

[0246] In a possible implementation manner, the first acquisition unit 901 is specifically configured to:

[0247] Acquire initial to-be-classified information;

[0248] Generate initial label information corresponding to the initial to-be-classified information and second accuracy information corresponding to the initial label information through a pre-trained classification model, where the initial label information is used to identify the information content corresponding to the initial to-be-classified information, and the second accuracy information is used to characterize the accuracy of the initial label information in identifying the information content;

[0249] Based on the accuracy characterized by the second accuracy information being less than the second accuracy threshold, determine the initial to-be-classified information as the to-be-classified information, and the information generation model is used to jointly perform the information classification task with the classification model.

[0250] In a possible implementation manner, the device further includes a third acquisition unit, a second determination unit, and a third adjustment unit:

[0251] The third acquisition unit is configured to acquire sample information, where the sample information has corresponding second sample label information, and the second sample label information is used to identify the information content corresponding to the sample information;

[0252] The second determination unit is configured to determine, through an initial classification model, third accuracy information corresponding to the sample information under multiple label information respectively, and determine one or more label information with the highest accuracy characterized by the corresponding third accuracy information as the second to-be-determined label information corresponding to the sample information, where the third accuracy information is used to characterize the accuracy of the corresponding label information in identifying the information content of the sample information;

[0253] The third adjustment unit is configured to adjust model parameters corresponding to the initial classification model according to a difference between the second sample label information and the second to-be-determined label information, so as to obtain the classification model.

[0254] In a possible implementation manner, the apparatus further includes a fourth acquisition unit, a third determination unit, a fourth determination unit, and a fourth generation unit:

[0255] The fourth acquisition unit is configured to acquire first to-be-processed information;

[0256] The third determination unit is configured to determine, by using the classification model, first label information and fourth accuracy information corresponding to the first to-be-processed information, where the fourth accuracy information is used to characterize an accuracy of the first label information in identifying information content of the first to-be-processed information;

[0257] The fourth determination unit is configured to determine the first label information as label information corresponding to the first to-be-processed information based on that the accuracy characterized by the fourth accuracy information is not less than the second accuracy threshold;

[0258] The fourth generation unit is configured to, based on that the accuracy characterized by the fourth accuracy information is less than the second accuracy threshold, perform the information classification task on the first to-be-processed information according to the indication information by using the information generation model, so as to generate label information corresponding to the first to-be-processed information.

[0259] In a possible implementation manner, the third determination unit is specifically configured to:

[0260] Perform information encoding on the first to-be-processed information, and extract feature information corresponding to the first to-be-processed information, where the feature information is used to characterize information content corresponding to the first to-be-processed information;

[0261] Determine the first label information and the fourth accuracy information corresponding to the first to-be-processed information according to the feature information;

[0262] The fourth generation unit is specifically configured to:

[0263] Perform the information classification task on the feature information according to the indication information, so as to generate label information corresponding to the first to-be-processed information.

[0264] In a possible implementation manner, the information to be classified is information corresponding to a first information field, where the first information field is any one of multiple information fields, and the apparatus further includes a storage unit, a fifth acquisition unit, a fifth determination unit, and an execution unit:

[0265] The storage unit is configured to store the indication information as the indication information corresponding to the first information domain;

[0266] The fifth acquisition unit is configured to acquire second information to be processed;

[0267] The fifth determination unit is configured to determine a second information domain corresponding to the second information to be processed, where the second information domain is any one of the multiple information domains;

[0268] The execution unit is configured to perform the information classification task on the second information to be processed according to the indication information corresponding to the second information domain through the information generation model.

[0269] In a possible implementation manner, the information to be classified is any one of multiple pieces of information to be classified, and the multiple pieces of information to be classified are all information corresponding to the first information domain. Specifically, the first adjustment unit 904 is configured to:

[0270] Adjust the initial indication information according to the first accuracy information respectively corresponding to the multiple pieces of information to be classified, so as to obtain first indication information, and the accuracy corresponding to the label information obtained by the information generation model performing the information classification task on the multiple pieces of information to be classified according to the first indication information is greater than the first accuracy threshold.

[0271] In a possible implementation manner, specifically, the first determination unit 903 is configured to:

[0272] Determine the semantic similarity between the first to-be-determined label information and the information content corresponding to the information to be classified, where the semantic similarity is used to characterize the similarity between the information content corresponding to the information to be classified and the information content identified by the first to-be-determined label information;

[0273] Determine the first accuracy information corresponding to the first to-be-determined label information according to the semantic similarity, and the accuracy characterized by the first accuracy information is positively correlated with the semantic similarity.

[0274] In a possible implementation manner, specifically, the first determination unit 903 is configured to:

[0275] Acquire first sample label information corresponding to the information to be classified, where the accuracy of the first sample label information identifying the information content corresponding to the information to be classified is greater than the first accuracy threshold;

[0276] Determine the difference information corresponding to the first to-be-determined label information according to the first sample label information, where the difference information is used to characterize the difference between the first sample label information and the first to-be-determined label information;

[0277] Determine the first accuracy information corresponding to the first to-be-determined tag information according to the difference information, where the accuracy represented by the first accuracy information is inversely correlated with the difference.

[0278] In a possible implementation manner, the first determining unit 903 is specifically configured to:

[0279] Determine the sub-accuracy information corresponding to the first to-be-determined tag information under the multiple information analysis methods through multiple information analysis methods, where the sub-accuracy information is used to represent, through the corresponding information analysis method, the accuracy of the first to-be-determined tag information in identifying the information content;

[0280] Determine the first accuracy information corresponding to the first to-be-determined tag information according to the weights corresponding to the multiple information analysis methods and the sub-accuracy information corresponding to the first to-be-determined tag information under the multiple information analysis methods. The first information analysis method is any one of the multiple information analysis methods, and the weight corresponding to the first information analysis method is used to indicate the influence degree of the sub-accuracy information corresponding to the first to-be-determined tag information under the first information analysis method on the first accuracy information.

[0281] The embodiment of the present application further provides a computer device. Please refer to Figure 10 as shown. This computer device may be a terminal device. Taking the terminal device as a mobile phone as an example:

[0282] Figure 10 Shown is a block diagram of a part of the structure of a mobile phone related to the terminal device provided in the embodiment of the present application. Refer to Figure 10 , the mobile phone includes: a radio frequency (RF) circuit 710, a memory 720, an input unit 730, a display unit 740, a sensor 750, an audio circuit 760, a wireless fidelity (WiFi) module 770, a processor 780, and a power supply 790 and other components. Those skilled in the art can understand that Figure 10 the structure of the mobile phone shown in

[0283] does not constitute a limitation on the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements. Figure 10 The following specifically introduces each component of the mobile phone in combination with

[0284] The RF circuit 710 can be used for receiving and transmitting information or signals during calls. Specifically, after receiving the downlink information from the base station, it is sent to the processor 780 for processing. Additionally, the uplink data designed is sent to the base station. Generally, the RF circuit 710 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (LNA), a duplexer, etc. In addition, the RF circuit 710 can also communicate with the network and other devices through wireless communication. The above wireless communication can use any communication standard or protocol, including but not limited to the Global System of Mobile communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0285] The memory 720 can be used to store software programs and modules. The processor 780 executes various functional applications and data processing of the mobile phone by running the software programs and modules stored in the memory 720. The memory 720 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory 720 can include a high-speed random access memory and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices.

[0286] The input unit 730 can be used to receive input numeric or character information, and generate key signal inputs related to the user settings and function controls of the mobile phone. Specifically, the input unit 730 can include a touch panel 731 and other input devices 732. The touch panel 731, also known as a touch screen, can collect touch operations of the user thereon or nearby (such as operations of the user using any suitable object or accessory such as a finger, a stylus, etc. on or near the touch panel 731), and drive corresponding connection devices according to a preset program. Optionally, the touch panel 731 can include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the touch position of the user, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into contact coordinates, and then sends it to the processor 780, and can receive and execute the commands sent by the processor 780. In addition, multiple types such as resistive, capacitive, infrared, and surface acoustic wave can be used to implement the touch panel 731. In addition to the touch panel 731, the input unit 730 can also include other input devices 732. Specifically, the other input devices 732 can include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control keys, power on / off keys, etc.), a trackball, a mouse, a joystick, etc.

[0287] The display unit 740 can be used to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 740 can include a display panel 741. Optionally, the display panel 741 can be configured in forms such as a liquid crystal display (LCD) or an organic light-emitting diode (OLED). Further, the touch panel 731 can cover the display panel 741. When the touch panel 731 detects a touch operation thereon or nearby, it transmits it to the processor 780 to determine the type of touch event. Subsequently, the processor 780 provides corresponding visual output on the display panel 741 according to the type of touch event. Although in Figure 10 the touch panel 731 and the display panel 741 are implemented as two independent components to realize the input and output functions of the mobile phone, in some embodiments, the touch panel 731 and the display panel 741 can be integrated to realize the input and output functions of the mobile phone.

[0288] The mobile phone may further include at least one sensor 750, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. Among them, the ambient light sensor can adjust the brightness of the display panel 741 according to the brightness of the ambient light, and the proximity sensor can turn off the display panel 741 and / or the backlight when the mobile phone is moved to the ear. As a kind of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary, and can be used in applications for identifying the posture of the mobile phone (such as horizontal and vertical screen switching, related games, magnetometer attitude calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for other sensors such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors that the mobile phone can also be configured with, they will not be elaborated here.

[0289] The audio circuit 760, the speaker 761, and the microphone 762 can provide an audio interface between the user and the mobile phone. The audio circuit 760 can transmit the electrical signal converted from the received audio data to the speaker 761, and the speaker 761 converts it into a sound signal for output; on the other hand, the microphone 762 converts the collected sound signal into an electrical signal, which is received by the audio circuit 760 and then converted into audio data. After the audio data is output to the processor 780 for processing, it is sent through the RF circuit 710 to, for example, another mobile phone, or the audio data is output to the memory 720 for further processing.

[0290] WiFi belongs to short - range wireless transmission technology. The mobile phone can help users send and receive emails, browse the web, and access streaming media through the WiFi module 770, which provides users with wireless broadband Internet access. Although Figure 10 the WiFi module 770 is shown, it can be understood that it does not belong to an essential component of the mobile phone and can be omitted entirely within the scope of not changing the essence of the invention according to needs.

[0291] The processor 780 is the control center of the mobile phone, connecting various parts of the entire mobile phone using various interfaces and lines. By running or executing software programs and / or modules stored in the memory 720, and by calling the data stored in the memory 720, it executes various functions of the mobile phone and processes data, thereby performing an overall detection of the mobile phone. Optionally, the processor 780 may include one or more processing units; preferably, the processor 780 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, and application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above - mentioned modem processor may not be integrated into the processor 780 either.

[0292] The mobile phone further includes a power supply 790 (such as a battery) for powering each component. Preferably, the power supply can be logically connected to the processor 780 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system.

[0293] Although not shown, the mobile phone may further include a camera, a Bluetooth module, etc., which will not be elaborated here.

[0294] In this embodiment, the processor 780 included in the terminal device further has the following functions:

[0295] Obtain the information to be classified and the initial indication information, where the initial indication information is used to describe the information classification task;

[0296] Through an information generation model, perform the information classification task on the information to be classified according to the initial indication information, and generate first undetermined label information corresponding to the information to be classified, where the first undetermined label information is used to identify the information content corresponding to the information to be classified;

[0297] Determine first accuracy information corresponding to the first undetermined label information, where the first accuracy information is used to characterize the accuracy of the first undetermined label information in identifying the information content;

[0298] Adjust the initial indication information according to the first accuracy information to obtain first indication information, and the accuracy of the label information obtained by the information generation model performing the information classification task according to the first indication information is greater than a first accuracy threshold, where the information generation model is used to perform the information classification task according to the first indication information.

[0299] The embodiment of the present application further provides a server. Please refer to Figure 11 as shown Figure 11 is a structural diagram of the server 800 provided by the embodiment of the present application. The server 800 may vary greatly due to different configurations or performances, and may include one or more central processing units (Central Processing Units, abbreviated as CPUs) 822 (for example, one or more processors) and a memory 832, and one or more storage media 830 (for example, one or more mass storage devices) for storing application programs 842 or data 844. Among them, the memory 832 and the storage media 830 may be transient storage or persistent storage. The program stored in the storage media 830 may include one or more modules (not marked in the figure), and each module may include a series of instruction operations on the server. Further, the central processing unit 822 may be configured to communicate with the storage media 830 and execute a series of instruction operations in the storage media 830 on the server 800.

[0300] The server 800 may also include one or more power supplies 826, one or more wired or wireless network interfaces 850, one or more input / output interfaces 858, and / or one or more operating systems 841, such as Windows Server TM , Mac OS X TM , Unix TM , Linux TM , FreeBSD TM and so on.

[0301] The steps performed by the server in the above embodiments may be based on Figure 11 the server structure shown.

[0302] The embodiments of the present application also provide a computer-readable storage medium for storing a computer program, which is used to execute any one of the information processing methods described in the foregoing various embodiments.

[0303] The embodiments of the present application also provide a computer program product including a computer program, which, when running on a computer device, causes the computer device to execute the information processing method described in any one of the above embodiments.

[0304] It can be understood that in the specific implementation of the present application, data related to user information (such as various user-related information to be classified, etc.) is involved. When the above embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0305] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium can be at least one of the following media: read-only memory (abbreviation: ROM), RAM, magnetic disk, or optical disk, etc., which can store program codes.

[0306] It should be noted that the various embodiments in this specification are described in a progressive manner. For the same or similar parts among the various embodiments, reference can be made to each other, and the key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the relevant parts of the method embodiments for the related content. The device and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0307] As described above, it is only a specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in this application should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. An information processing method, characterized in that: The method comprises: Acquire information to be classified and initial indication information, wherein the initial indication information is used to describe the information classification task; By means of an information generation model, performing the information classification task on the information to be classified according to the initial indication information, generating first pending label information corresponding to the information to be classified, wherein the first pending label information is used to identify information content corresponding to the information to be classified; Determine first accuracy information corresponding to the first pending tag information, where the first accuracy information is used to represent the accuracy of the first pending tag information in identifying the information content; The initial indication information is adjusted according to the first accuracy information to obtain first indication information, the accuracy corresponding to the label information obtained by the information generation model when performing the information classification task according to the first indication information is greater than a first accuracy threshold, and the information generation model is used to perform the information classification task according to the first indication information.

2. The method according to claim 1, characterized in that The step of adjusting the initial indication information according to the first accuracy information to obtain the first indication information includes: By means of the information generation model, a defect analysis task is performed on the initial indication information according to the second indication information to generate defect information corresponding to the initial indication information, wherein the defect information is used to describe information defects existing in the initial indication information, the information defects are the reasons causing the accuracy represented by the first accuracy information to be not greater than the first accuracy threshold, and the second indication information is used to describe the defect analysis task; Through the information generation model, an information adjustment task is performed on the initial indication information according to the defect information and the third indication information to generate the first indication information, the information adjustment task is used to correct the information defect, and the third indication information is used to describe the information adjustment task.

3. The method according to claim 2, characterized in that The method further comprises: Acquire sample initial indication information, the sample initial indication information having corresponding sample indication information and sample defect information, the sample defect information being used to describe sample information defects existing in the sample initial indication information, and the sample indication information being information obtained after the sample information defects in the sample initial indication information are corrected; The defect analysis task is performed on the sample initial indication information according to the second indication information through the initial information generation model to generate pending defect information corresponding to the sample initial indication information; By means of the initial information generation model, the information adjustment task is performed on the sample initial indication information according to the sample defect information and the third indication information to generate pending indication information; According to the difference between the pending defect information and the sample defect information, and the difference between the pending indication information and the sample indication information, the old model parameters corresponding to the initial information generation model are adjusted to obtain the information generation model.

4. The method according to claim 2, characterized in that: The step of performing a defect analysis task on the initial indication information according to the second indication information through the information generation model to generate defect information corresponding to the initial indication information includes: Through the information generation model, a defect analysis task is performed on the initial indication information according to the second indication information, the information to be classified, the first pending label information and the first sample label information corresponding to the information to be classified, and defect information corresponding to the initial indication information is generated, wherein the first sample label information identifies that the accuracy of the information content corresponding to the information to be classified is greater than the first accuracy threshold.

5. The method according to claim 1, characterized in that The obtaining of the information to be classified and the initial indication information includes: Obtaining initial information to be classified; Generate initial label information corresponding to the initial information to be classified and second accuracy information corresponding to the initial label information through the classification model obtained by pre-training, wherein the initial label information is used to identify the information content corresponding to the initial information to be classified, and the second accuracy information is used to characterize the accuracy of the initial label information in identifying the information content; Based on the accuracy of the second accuracy information representation being less than a second accuracy threshold, the initial information to be classified is determined as the information to be classified, and the information generation model is used to perform the information classification task together with the classification model.

6. The method according to claim 5, characterized in that The method further comprises: Acquire sample information, where the sample information has corresponding second sample label information, where the second sample label information is used to identify information content corresponding to the sample information; Determine, by means of an initial classification model, third accuracy information corresponding to the sample information under a plurality of label information, and determine one or more label information with the greatest accuracy represented by the corresponding third accuracy information as second pending label information corresponding to the sample information, wherein the third accuracy information is used to represent the accuracy with which the corresponding label information identifies the information content of the sample information; According to the difference between the second sample label information and the second undetermined label information, the model parameters corresponding to the initial classification model are adjusted to obtain the classification model.

7. The method according to claim 5, characterized in that The method further comprises: Obtaining first information to be processed; Determining, by means of the classification model, first label information and fourth accuracy information corresponding to the first information to be processed, wherein the fourth accuracy information is used to characterize the accuracy with which the first label information identifies the information content of the first information to be processed; Based on the accuracy represented by the fourth accuracy information being not less than the second accuracy threshold, determining the first label information as label information corresponding to the first information to be processed; Based on the accuracy represented by the fourth accuracy information being less than the second accuracy threshold, the information classification task is performed on the first information to be processed according to the indication information through the information generation model to generate label information corresponding to the first information to be processed.

8. The method according to claim 5, characterized in that The determining the first label information and the fourth accuracy information corresponding to the first information to be processed includes: Encoding the first information to be processed, and extracting characteristic information corresponding to the first information to be processed, where the characteristic information is used to characterize information content corresponding to the first information to be processed; Determine, according to the feature information, first label information and the fourth accuracy information corresponding to the first information to be processed; The performing the information classification task on the first information to be processed according to the indication information to generate label information corresponding to the first information to be processed includes: The information classification task is performed on the characteristic information according to the indication information to generate label information corresponding to the first information to be processed.

9. The method according to claim 1, characterized in that: The information to be classified is information corresponding to a first information field, the first information field is any one of a plurality of information fields, and the method further includes: storing the indication information as indication information corresponding to the first information field; Obtaining second information to be processed; Determine a second information field corresponding to the second information to be processed, where the second information field is any one of the multiple information fields; The information classification task is performed on the second information to be processed through the information generation model according to the indication information corresponding to the second information field.

10. The method according to claim 9, characterized in that The information to be classified is any one of a plurality of information to be classified, and the plurality of information to be classified are all information corresponding to the first information field, and the adjusting the initial indication information according to the first accuracy information to obtain the first indication information includes: According to the first accuracy information respectively corresponding to the multiple information to be classified, the initial indication information is adjusted to obtain the first indication information. The information generation model performs the information classification task on the multiple information to be classified according to the first indication information, and the accuracy corresponding to the label information obtained is greater than the first accuracy threshold.

11. The method according to any one of claims 1 to 10, characterized in that: The determining the first accuracy information corresponding to the first pending tag information includes: Determine the semantic similarity between the first pending label information and the information to be classified, wherein the semantic similarity is used to represent the similarity between the information content corresponding to the information to be classified and the information content identified by the first pending label information; First accuracy information corresponding to the first undetermined tag information is determined according to the semantic similarity, and the accuracy represented by the first accuracy information is positively correlated with the semantic similarity.

12. The method according to any one of claims 1 to 10, characterized in that: The determining the first accuracy information corresponding to the first pending tag information includes: Acquire first sample label information corresponding to the information to be classified, where the first sample label information indicates that the accuracy of the information content corresponding to the information to be classified is greater than the first accuracy threshold; Determine, according to the first sample label information, difference information corresponding to the first pending label information, where the difference information is used to characterize the difference between the first sample label information and the first pending label information; First accuracy information corresponding to the first pending label information is determined according to the difference information, and the accuracy represented by the first accuracy information is inversely correlated with the difference.

13. The method according to any one of claims 1 to 10, characterized in that: The determining the first accuracy information corresponding to the first pending tag information includes: Determine, by using multiple information analysis methods, sub-accuracy information corresponding to the first pending label information under the multiple information analysis methods, wherein the sub-accuracy information is used to characterize the accuracy of the first pending label information in identifying the information content by using the corresponding information analysis method; According to the weights corresponding to the multiple information analysis methods and the sub-accuracy information corresponding to the first pending label information under the multiple information analysis methods, the first accuracy information corresponding to the first pending label information is determined, the first information analysis method is any one of the multiple information analysis methods, and the weight corresponding to the first information analysis method is used to indicate the degree of influence of the sub-accuracy information corresponding to the first pending label information under the first information analysis method on the first accuracy information.

14. An information processing device, characterized in that: The device comprises a first acquisition unit, a first generation unit, a first determination unit and a first adjustment unit: The first acquisition unit is used to acquire the information to be classified and initial indication information, wherein the initial indication information is used to describe the information classification task; The first generating unit is used to perform the information classification task on the information to be classified according to the initial indication information through the information generation model, and generate first pending label information corresponding to the information to be classified, wherein the first pending label information is used to identify the information content corresponding to the information to be classified; The first determining unit is used to determine first accuracy information corresponding to the first pending tag information, where the first accuracy information is used to represent the accuracy of the first pending tag information in identifying the information content; The first adjustment unit is used to adjust the initial indication information according to the first accuracy information to obtain first indication information. The accuracy corresponding to the label information obtained by the information generation model when performing the information classification task according to the first indication information is greater than the first accuracy threshold. The information generation model is used to perform the information classification task according to the first indication information.

15. A computer device, characterized in that: The computer device comprises a processor and a memory: The memory is used to store a computer program and transmit the computer program to the processor; The processor is used to execute the information processing method according to any one of claims 1 to 13 according to the instructions in the computer program.

16. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the information processing method according to any one of claims 1 to 13.

17. A computer program product comprising a computer program, which, when executed on a computer device, enables the computer device to execute the information processing method according to any one of claims 1 to 13.

Images