Sample data processing method and device, equipment, storage medium and program product

Abstract

The application discloses a sample data processing method and device, equipment, a storage medium and a program product, and relates to the technical field of computers. In the method, a mark file comprises result statistical areas corresponding to mark results one by one, each sample identifier is recorded in a result statistical area of a mark result to be associated, and the sample identifiers in each result statistical area are stored in a continuous storage area of a storage medium. In this way, when sample data with a target mark result is screened, multiple sample identifiers associated with one mark result can be read in one IO reading process, that is, the association between multiple sample identifiers and mark results can be read in one IO reading process. Therefore, the number of IOs for data reading can be greatly reduced, data reading efficiency is improved, and sample data processing efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to sample data processing methods, apparatus, devices, storage media, and program products. Background Technology

[0002] Some artificial intelligence platforms allow users to import and process sample data, such as modifying low-quality sample data. They also support users in creating or importing models to be trained and training these models based on the processed sample data.

[0003] Currently, in order to ensure the quality of sample data, these artificial intelligence platforms need to check each sample data item by item and manually modify the poor-quality sample data, resulting in relatively low sample data processing efficiency. Summary of the Invention

[0004] This application provides a sample data processing method, a sample data processing apparatus, an electronic device, a non-volatile computer-readable storage medium, and a computer program product to at least solve the problem of low sample data processing efficiency in related technologies.

[0005] This application provides a sample data processing method, including:

[0006] In response to receiving a first display instruction, at least a portion of the sample data in the sample dataset is displayed, the sample dataset comprising multiple sample data entries, and the sample data entries having sample identifiers;

[0007] In response to the displayed sample data being labeled, the association between the sample identifier and the labeling result of the sample data is recorded in the labeling file. The labeling result represents the data category to which the sample data belongs. The labeling file includes a result statistics area that corresponds one-to-one with the labeling result. When recording the association between the sample identifier and the labeling result, each sample identifier is recorded in the result statistics area of ​​the labeling result to be associated, and the sample identifier in each result statistics area is stored in a contiguous storage area of ​​the storage medium.

[0008] In response to receiving a filtering instruction including target annotation results, the system filters sample data with the target annotation results from the sample dataset according to the correlation between sample identifiers and annotation results recorded in the annotation file, and processes the filtered sample data.

[0009] This application also provides a sample data processing apparatus, including:

[0010] A display module is configured to, in response to receiving a first display instruction, display at least a portion of the sample data in the sample dataset, wherein the sample dataset includes multiple sample data entries, and each sample data entry has a sample identifier;

[0011] The association module is used to respond to the display of sample data being labeled, record the association between the sample identifier and the labeling result in the labeling file, the labeling result represents the data category to which the sample data belongs, and the labeling file includes a result statistics area that corresponds one-to-one with the labeling result. When recording the association between the sample identifier and the labeling result, each sample identifier is recorded in the result statistics area of ​​the labeling result to be associated, and the sample identifier in each result statistics area is stored in a contiguous storage area of ​​the storage medium.

[0012] The filtering module is used to respond to receiving a filtering instruction including target annotation results, filter sample data with target annotation results in the sample dataset according to the correlation between sample identifiers and annotation results recorded in the annotation file, and process the filtered sample data.

[0013] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described sample data processing methods.

[0014] This application also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above-described sample data processing methods.

[0015] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described sample data processing methods.

[0016] This application provides a labeling file that includes a result statistics area corresponding one-to-one with each labeling result. Each sample identifier is recorded within the result statistics area of ​​the labeling results to be associated, and the sample identifiers in each result statistics area are stored in a contiguous storage area of ​​the storage medium. Thus, when filtering sample data with target labeling results, multiple sample identifiers associated with a single labeling result can be read in a single IO read. This is equivalent to reading the association between multiple sample identifiers and labeling results in a single IO read, significantly reducing the number of data read IO operations, improving data reading efficiency, and consequently improving the processing efficiency of sample data. Attached Figure Description

[0017] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 These are schematic diagrams of modules in some artificial intelligence platforms.

[0019] Figure 2 This is one example image of paginated display of sample data;

[0020] Figure 3 These are schematic diagrams of modules of the artificial intelligence platform provided in some embodiments of this application;

[0021] Figure 4 This is a schematic diagram of the sample data annotation interface provided in some embodiments of this application;

[0022] Figure 5 This is a schematic diagram of a sample data processing interface provided in some embodiments of this application;

[0023] Figure 6 This is a schematic diagram of a sample data processing interface provided in some other embodiments of this application;

[0024] Figure 7 This is a schematic diagram of a sample data processing interface provided in some other embodiments of this application;

[0025] Figure 8 This is a schematic diagram of a sample data processing interface provided in some other embodiments of this application;

[0026] Figure 9 This is a schematic diagram of a sample data processing interface provided in some other embodiments of this application;

[0027] Figure 10 This is a schematic diagram of a sample data processing interface provided in some other embodiments of this application;

[0028] Figure 11 This is a schematic diagram of a sample data processing interface provided in some other embodiments of this application;

[0029] Figure 12 This is a schematic diagram of a sample data processing interface provided in some other embodiments of this application;

[0030] Figure 13 This is a schematic diagram of a summary browsing interface for annotation results provided in some embodiments of this application;

[0031] Figure 14This is a flowchart illustrating a sample data processing method provided in some embodiments of this application;

[0032] Figure 15 This is a schematic diagram of a sample data processing apparatus provided in some embodiments of this application;

[0033] Figure 16 These are schematic diagrams of the structure of electronic devices provided in some embodiments of this application. Detailed Implementation

[0034] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0035] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0036] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] See also Figure 1 This is a schematic diagram of modules in an artificial intelligence platform in some technologies. Figure 1 In this system, the artificial intelligence platform includes a sample data processing system and a model training system. The sample data processing system processes user-input sample data and provides the processed data to the model training system. Based on the sample data provided by the sample data processing system, the model training system can train or fine-tune models created or imported by the user.

[0038] The sample data processing system may include an initial data management module and a dataset management module. Users can store sample data in the initial data management module through local upload, remote download, or other methods. Alternatively, the output data of data processing tasks can be used as sample data and stored in the initial data management module. The initial data management module may include storage media for storing sample data.

[0039] In the initial data management module, users can select at least a portion of the sample data as a sample dataset and import it into the dataset management module. The dataset management module may include a data processing interface. This interface displays at least a portion of the sample data in the sample dataset. Users can process the sample data in the interface, for example, modifying poor-quality sample data. The processed sample data can be saved to the model training system's cache. This allows the model training system to train or fine-tune the model based on the cached sample data.

[0040] Sample data can be text data, and the unit of measurement can be "strips". The sample data can vary for different model training or fine-tuning tasks. For example, for the training or fine-tuning task of a question-answering model, each sample data can include both a question and an answer, and the questions and answers in different sample data can be different, as shown in Table 1.

[0041] Table 1 Sample Data

[0042]

[0043] In Table 1 above, the questions and answers in the second row can be used as one sample data point, and the questions and answers in the third row can be used as another sample data point. After training or fine-tuning the question-answering model using sample data similar to that in Table 1, the question-answering model can answer the questions posed by users.

[0044] For example, for the task of training or fine-tuning a recommendation model, each sample data can include the user's historical behavior, user characteristics, item characteristics, and interaction results. The historical behavior, user characteristics, item characteristics, and interaction results in different sample data are not completely the same, as shown in Table 2.

[0045] Table 2 Sample Data

[0046]

[0047] In Table 2 above, the data in the first, second, and third rows can each be used as a sample data point. After training or fine-tuning the recommendation model using sample data similar to that in Table 2, the model can learn the potential relationships between user behavior and interaction results, such as the potential relationship between user age and movie IDs, user IDs and article IDs, and reading duration. Thus, after the recommendation model is trained, inputting user IDs, user age, and other information into the model will allow it to recommend movie IDs and article IDs that match the user ID.

[0048] Sample data can be stored in sample files. Each sample file can contain multiple sample data entries, such as 10,000 or 20,000 entries. In the storage medium, the sample data from all sample files is stored aggregated, and each sample data entry has its own corresponding row number, which serves as the sample identifier. For example, Table 3 provides an illustrative correspondence between sample data and sample identifiers:

[0049] Table 3. Correspondence between sample identifiers and sample data

[0050]

[0051] Users can select one or more sample files in the initial data management module. The sample data in the selected sample files constitutes the sample dataset. When there is a large amount of sample data in the sample dataset, the dataset management module can display the sample data in pages on the data processing interface. Page display means dividing the sample data in the sample dataset into multiple subsets, with each page displaying the sample data from one subset. For easier understanding, please refer to the relevant documentation. Figure 2 This is one example image used to display sample data in paginated format. Figure 2 In this case, assuming the data processing interface supports displaying 3 sample data at a time, and the sample dataset contains 300 sample data, then the sample data in the sample dataset can be divided into 100 subsets, and each subset contains 3 sample data. The sample data in each subset is displayed as one page.

[0052] In the data processing interface, users can sequentially check each sample data point according to the target conditions that the model training data needs to meet, and determine whether the sample data meets the target conditions. These target conditions include, but are not limited to, high data accuracy, representativeness, good timeliness, completeness, and low noise. For any given sample data point, the data quality can be determined based on the number of target conditions met. For example, if the sample data meets all target conditions, its data quality is considered good; if it meets some, its data quality is considered medium; and if it does not meet any, its data quality is considered poor. Based on the data quality, users can process the sample data in the data processing interface, such as modifying poor-quality samples to meet the target conditions. This results in higher-quality sample data, ensuring the accuracy of subsequent model training. However, these techniques require users to check and modify each sample data point individually, which is labor-intensive and not practical.

[0053] In view of this, this application addresses Figure 1The AI ​​platform shown has been improved. (See also...) Figure 3 This is a schematic diagram of the modules of an artificial intelligence platform provided in some embodiments of this application. Figure 3 and Figure 1 The basic structure is similar, with the main difference being the addition of data annotation, data filtering, and data export functions to the dataset management module. Based on these new functions, users can annotate the data quality of sample data within the dataset management module. The annotation results characterize the data category to which the sample data belongs, and this category can be based on the quality level of the sample data. For example, high-quality sample data can be classified as the first category and labeled "Good"; medium-quality sample data can be classified as the second category and labeled "Medium"; and low-quality sample data can be classified as the third category and labeled "Poor". Based on the annotation results, sample data that meets the data quality requirements can be selected as training data for the model. For example, in a labeled dataset, sample data labeled "Good" can be selected as training data for the model. This allows for the direct removal of low- or medium-quality sample data without manually modifying low-quality data, thus reducing workload.

[0054] To implement the aforementioned new functions, the dataset management module of this application may specifically include a sample data annotation interface and a sample data processing interface. These will be described separately below.

[0055] See also Figure 4 This is a schematic diagram of the sample data annotation interface provided in some embodiments of this application. Figure 4 It includes a sample data display area 41, a sample annotation area 42, and a sample data switching area 43. The sample data display area 41 can display one sample data point from the sample dataset. The sample annotation area 42 can display multiple data category options, such as "Good," "Medium," and "Poor."

[0056] Users can annotate the sample data displayed in sample data display area 41 by selecting options in sample annotation area 42, and obtain the annotation results of the sample data. For example Figure 4 In the sample labeling area 42, the option "Good" is selected, which means that the sample data displayed in the sample data display area 41 will be labeled as "Good", that is, the labeling result of the sample data is "Good".

[0057] When the sample data in the sample data display area 41 has been labeled, the user can update the labeling result of the sample data by changing the selected option in the sample labeling area 42. For example, when the selected option in the sample labeling area 42 is changed from "good" to "medium", it means that the labeling result of the sample data is updated from "good" to "medium".

[0058] After the labeling of the sample data in the sample data display area 41 is completed, based on the first switching component 431 or the second switching component 432 in the sample data switching area 43, the sample data displayed in the sample data display area 41 can be switched. For example, assume that the sample data currently displayed in the sample data display area 41 is sample data A1. After the labeling of sample data A1 is completed, click the first switching component 431, and the previous sample data of sample data A1 can be displayed in the sample data display area 41. Click the second switching component 432, and the next sample data of sample data A1 can be displayed in the sample data display area 41.

[0059] After the user completes the labeling of any sample data displayed in the sample labeling area 42, the dataset management module can record the association relationship between the sample identifier and the labeling result of the sample data. For example, assume that the sample identifier of sample data A2 is "201". After the user labels sample data A2 as "medium" in the sample labeling area 42, the dataset management module can record the association relationship between the sample identifier "201" and the labeling result "medium".

[0060] Based on the association relationship between the sample identifier and the labeling result, it can be Figure 5 shown in the sample data processing interface as shown, and each sample data and the labeling result of the sample data can be displayed. For example, taking the display of sample data A1 and the labeling result of sample data A1 as an example for illustration. First, sample data A1 and the sample identifier 200 of sample data A1 can be read from the sample dataset. According to the sample identifier 200, the corresponding labeling result of the sample identifier can be found in the association relationship between the sample identifier and the labeling result, and the sample data A1 and the labeling result corresponding to the sample identifier 200 can be displayed simultaneously in the sample data processing interface. In this way, the user can view the labeling results of each sample data one by one.

[0061] Refer to in combination Figure 6In some embodiments, the sample data processing interface further includes a filtering component for the annotation results, namely the component within the first dashed box 61. The filtering component can be a drop-down list. The drop-down list of the filtering component can include options representing various data categories, such as "Good," "Medium," and "Poor," and can also include options representing all annotation results, such as "All." Users can select one option from the drop-down list of the filtering component according to their actual needs. Based on the selected option, only sample data with corresponding annotation results can be displayed in the sample data processing interface. For example... Figure 6 In the sample data processing interface, if the "Good" option is selected in the drop-down list of the filtering component, only sample data labeled "Good" will be displayed. For example, in... Figure 7 In the sample data processing interface shown, the "All" option is selected in the drop-down list of the filtering component, indicating that all labeled sample data will be displayed. Based on Figure 6 and Figure 7 As can be seen, after setting up the filtering component, users can browse sample data with target annotation results according to their actual needs, making it more applicable.

[0062] See also Figure 8 In some embodiments, the sample data processing interface also includes an export component, namely the component within the second dashed box 81. Figure 8 In the sample data processing interface shown, the data filtering function based on the filtering component can display only sample data with target annotation results. When the export component is triggered, this sample data with target annotation results can be exported, and the exported sample data can be used as a new sample dataset. The new sample dataset can include only sample data with target annotation results. For example, assuming the option "Good" is selected in the drop-down list of the filtering component, the sample data processing interface can only display sample data with the annotation result of "Good". At this time, if the export component is triggered, the sample data with the annotation result of "Good" can be exported, resulting in a new sample dataset. The new sample dataset only includes sample data with the annotation result of "Good", eliminating the poor and medium quality sample data in the original sample dataset. Therefore, the new sample dataset can be used for model training or fine-tuning. In the solution of this application, the method based on quality annotation can select high-quality sample data from the sample dataset as sample data for model training or fine-tuning, without requiring the user to modify the poor quality sample data one by one, greatly reducing the workload.

[0063] In summary, in the artificial intelligence platform of this application, users can obtain a new sample dataset for model training based on the following steps:

[0064] 1) In the initial data management module, select the required sample files and import them into the dataset management module to obtain the sample dataset;

[0065] 2) In Figure 4 The sample data annotation interface shown allows you to annotate each sample data in the sample dataset, obtaining the correlation between the sample identifier and the annotation result for each sample data.

[0066] 3) Based on the correlation between sample identifiers and annotation results, in Figure 8 In the sample data processing interface shown, selecting the target annotation result (e.g., "Good") using the filtering component allows the interface to display only sample data with target annotation results. Then, triggering the export component allows exporting the sample data with target annotation results. This yields a new sample dataset for model training.

[0067] See also Figure 9 In some embodiments, each sample data in the sample data processing interface also has a corresponding annotation component, namely the component within the third dashed box 91. For any sample data, if the annotation component corresponding to that sample data is triggered, the sample data annotation interface for that sample data can be displayed. Users can update the annotation results of the sample data in the sample data annotation interface. In this way, when users browse the annotation results of the sample data in the sample data processing interface, if they find that the annotation results of the sample data are incorrect, they can modify the annotation results of the sample data in a timely and convenient manner.

[0068] Of course, even if the sample data is unlabeled, labeling can be triggered through the sample data processing interface. For example, Figure 10 In some cases, some sample data is not labeled. In such cases, the labeling component corresponding to the sample data can be used to trigger the labeling of each sample data.

[0069] Continue reading Figure 10 In the sample data processing interface, if multiple unlabeled sample data can have the same labeling result, such as "good," performing labeling operations on each of these sample data separately would inevitably lead to cumbersome operations. Therefore, in conjunction with the relevant references... Figure 11 In some embodiments, the sample data processing interface also includes a batch annotation component, namely the component within the fourth dashed box 111. If multiple unannotated sample data can have the same annotation result, these multiple sample data can be selected, and the batch annotation component can be triggered. At this time, the sample data annotation interface for the multiple sample data can be displayed. In the sample data annotation interface, the multiple sample data can be batch-annotated. This can greatly simplify the annotation process and improve annotation efficiency.

[0070] Of course, this is understandable. Figure 11 In the process, if it is necessary to update the annotation results of multiple sample data, and the updated annotation results of these multiple sample data are the same, it can also be done according to... Figure 10 By selecting multiple sample data points and triggering the batch annotation component, the annotation results of these multiple sample data points can be updated in batches on the displayed sample data annotation interface, thereby improving the efficiency of updating the annotation results.

[0071] See also Figure 12 In some embodiments, the sample data processing interface also includes a progress viewing component, namely the component within the fifth dashed box 121. When the progress viewing component is triggered, it can display, for example... Figure 13 The interface shown is a summary view of the annotation results. This interface displays the total number of data entries in the sample dataset and the number of annotated data entries. It also shows the number of data entries corresponding to each annotation result. This allows users to easily understand the summary of the sample data annotation.

[0072] In summary, the sample data annotation scheme proposed in this application can select high-quality sample data from the sample dataset as model training data, thus avoiding the large workload caused by modifying each sample data item individually. However, this sample data annotation scheme requires storing the association between sample identifiers and annotation results. Currently, in some technologies, the association between the sample identifier and annotation result of each sample data item can be stored as a record in a database, as shown in Table 4:

[0073] Table 4. Relationship between Sample Identification and Labeling Results

[0074]

[0075] The association between sample identifiers and annotation results can be stored in the database using a row-based storage method. Row-based storage means that the data for each record in the database is stored in a contiguous area of ​​the storage medium, while data from different records can be stored in non-contiguous areas. This storage method has the following two problems:

[0076] 1) Since the number of sample data is usually large (e.g., millions of data), there are also many records of the relationship between sample identifiers and annotation results in the database, which will put a lot of performance pressure on the database.

[0077] 2) When filtering sample data with target annotations, it is necessary to read each record in the database and search for records with target annotations. Since different records can be stored in non-contiguous storage areas of the storage medium, and each storage area requires at least one I / O read, it is equivalent to each data record requiring at least one I / O read. Therefore, the number of I / O (Input / Output) reads required when filtering sample data with target annotations is relatively large, which greatly reduces data reading efficiency and consequently reduces the processing efficiency of sample data.

[0078] Therefore, this application provides a sample data processing method that can improve the processing efficiency of sample data. The sample data processing method can be applied to... Figure 3 The sample data processing system, or an electronic device that can be used to run the sample data processing system. Electronic devices may include, but are not limited to, tablets, laptops, desktop computers, servers, etc. (See also...) Figure 14 This is a flowchart illustrating a sample data processing method provided in some embodiments of this application. Figure 14 In this study, the sample data processing method includes the following steps:

[0079] Step S141: In response to receiving the first display instruction, at least a portion of the sample data in the sample dataset is displayed. The sample dataset includes multiple sample data entries, and each sample data entry has a sample identifier.

[0080] Specifically, the sample dataset refers to the collection of sample data imported from the initial data management module to the dataset management module. Before executing step S141, the sample data processing system can, in response to the selection of one or more sample data in the initial data management module, import the selected sample data into the dataset management module as sample data in the sample dataset. It should be noted that when sample data is stored in sample files, the sample data processing system can, in response to the selection of one or more sample text files in the initial data management module, import the selected sample files into the dataset management module and use the sample data in the selected sample files as sample data in the sample dataset.

[0081] Furthermore, when importing selected sample data into the dataset management module, storing sample data from the sample dataset in both the initial data management module and the dataset management module simultaneously would inevitably lead to duplicate storage of sample data, resulting in wasted storage space. Therefore, in this embodiment, the storage path of the selected sample data in the initial data management module can be imported into the dataset management module. When the dataset management module displays at least a portion of the sample data in the sample dataset, it can retrieve the sample data to be displayed from the initial data management module based on the storage path. That is, the entity data of the sample data is stored in the initial data management module, while the dataset management module only contains the storage path of the sample data, not the entity data. When the dataset management module needs to display the sample data, it can retrieve the sample data from the initial data management module based on the storage path. This avoids the problem of wasted storage space caused by storing the same sample data in both the initial data management module and the dataset management module simultaneously.

[0082] Furthermore, after importing the sample dataset into the dataset management module, modifying the sample data in the initial data management module may lead to inaccurate annotation results in the dataset management module. For example, if sample data A3 is labeled "good" in the dataset management module, but the word order, content, etc., of sample data A3 are adjusted in the initial data management module, causing a decrease in the quality of sample data A3, then the annotation results for sample data A3 in the dataset management module will be inaccurate.

[0083] In this embodiment, to avoid inaccurate annotation results caused by changes in sample data in the initial data management module, after importing the storage path into the dataset management module, selected sample data can be identified in the initial data management module to distinguish between selected and unselected sample data. Selected sample data refers to sample data imported into the dataset management module, while unselected sample data refers to sample data not imported into the dataset management module. The identification method for sample data can be selected according to actual needs, and this application does not impose any restrictions. For example, selected sample data can be identified using icons, while unselected sample data can be unidentified. In this way, users can determine whether sample data has been imported into the dataset management module based on the icons. For sample data that has already been imported into the dataset management module, no modifications can be made in the initial data management module, thereby ensuring the accuracy of the annotation results in the dataset management module.

[0084] Step S142, in response to a labeling operation being performed on the displayed sample data, record the association relationship between the sample identifier of the sample data and the labeling result in the labeling file. The labeling result represents the data category to which the sample data belongs. The labeling file includes result statistical regions corresponding one-to-one to the labeling results. When recording the association relationship between the sample identifier and the labeling result, record each sample identifier in the result statistical region of the labeling result to be associated, and the sample identifiers in each result statistical region are stored in a continuous storage region of the storage medium.

[0085] Regarding the labeling result, data category, association relationship between the sample identifier and the labeling result, reference can be made to the above relevant description, which will not be elaborated here.

[0086] In this embodiment, on the one hand, save the association relationship between the sample identifier and the labeling result in the labeling file, and the data in the labeling file can be directly written to the storage medium. In this way, it is possible to avoid saving the association relationship between the sample identifier and the labeling result to the database and causing an excessive performance pressure on the database. On the other hand, the labeling file includes result statistical regions corresponding one-to-one to the labeling results. Record each sample identifier in the result statistical region of the labeling result to be associated, and the sample identifiers in each result statistical region are stored in a continuous storage region of the storage medium. In this way, in one IO read process, multiple sample identifiers associated with one labeling result can be read, which is equivalent to reading multiple association relationships between sample identifiers and labeling results in one IO read process. In this way, the IO times of data reading can be greatly reduced, and thus the data reading efficiency and the processing efficiency of sample data can be improved.

[0087] For the convenience of understanding, take the labeling results of "good", "medium", and "bad" as examples. The division of the result statistical regions in the labeling file can be similar to the following:

[0088] 〝streams〞:

[0089] {

[0090] 〝outs〞:

[0091] {

[0092] 〝result〞:〝good〞,

[0093] 〝ID〞:〝0、3、4、5、……、20〞,

[0094] },

[0095] {

[0096] 〝result〞:〝medium〞,

[0097] 〝ID〞:〝1、2〞,

[0098] },

[0099] {

[0100] "result": "bad",

[0101] "ID": "6",

[0102] }

[0104] }

[0105] In the above example, "result" represents the annotation result, and "ID" represents the sample identifier. The area where each "result" is located can be used as a result statistical area. Among them:

[0106] The annotation result "good" is associated with the sample identifiers 0, 3, 4, 5, 7,..., 20, indicating that the annotation results corresponding to the sample data with sample identifiers 0, 3, 4, 5, 7,..., 20 are "good";

[0107] The annotation result "medium" is associated with the sample identifiers 1, 2, indicating that the annotation results corresponding to the sample data with sample identifiers 1, 2 are "medium";

[0108] The annotation result "bad" is associated with the sample identifier ⑥, indicating that the annotation result corresponding to the sample data with sample identifier ⑥ is "bad".

[0109] According to the above example, when storing data in the storage medium, the annotation result "good" and the sample identifiers 0, 3, 4, 5, 7,..., 20 can be stored in a continuous storage area, and the annotation result "medium" and the sample identifiers 1, 2 can be stored in another continuous storage area. And so on. In this way, when screening the sample data with the target annotation result, at least only 1 IO read is required to read all the association relationships between the sample identifier and the annotation result. Compared with some technologies that store the association relationships between the sample identifier and the annotation result in the database, the number of IO reads required by some technologies may be millions of times. The technical solution of this application can greatly reduce the number of IO reads, thereby improving the processing efficiency of the sample data.

[0110] Step S143, in response to receiving a screening instruction including the target annotation result, according to the association relationship between the sample identifier and the annotation result recorded in the annotation file, screen the sample data with the target annotation result in the sample dataset, and process the screened sample data.

[0111] Specifically, for screening the sample data with the target annotation result, reference can be made to the above step S142 and Figure 5 、 Figure 6 ​The relevant descriptions are not repeated here.

[0112] The filtered sample data is processed, including but not limited to displaying sample data with target annotation results (such as...). Figure 5 As shown), the displayed sample data is annotated and updated (e.g. Figure 9 , 10 As shown), export the sample data with target annotation results as a new sample dataset (e.g. Figure 8 (as shown in the image) etc.

[0113] In summary, in the technical solutions of some embodiments of this application, the annotation file includes a result statistics area that corresponds one-to-one with the annotation results. Each sample identifier is recorded in the result statistics area of ​​the annotation results to be associated, and the sample identifiers in each result statistics area are stored in a contiguous storage area of ​​the storage medium. Thus, when filtering sample data with target annotation results, multiple sample identifiers associated with a single annotation result can be read in a single IO read. This is equivalent to reading the association between multiple sample identifiers and annotation results in a single IO read, which can greatly reduce the number of data read IOs, improve data reading efficiency, and thereby improve the processing efficiency of sample data.

[0114] Furthermore, referring to the example of result statistics area division in the above annotation file, it is understandable that when there are many records of the relationship between sample identifiers and annotation results (such as millions of records), there will be many sample identifiers in at least some result statistics areas. In this case, when reading sample identifiers in a single result statistics area, the time for a single IO read may be very long, which will reduce data reading efficiency and thus affect the processing efficiency of sample data.

[0115] Therefore, in some embodiments, the sample identifiers of the sample data in the sample dataset can be divided into multiple sample identifier intervals. For example, sample identifiers 1 to 1000 are divided into the first sample identifier interval, sample identifiers 1001 to 2000 are divided into the second sample identifier interval, and so on.

[0116] When dividing the sample identifier intervals, the number of sample identifiers in each interval can be the same, or at least the number of sample identifiers in some intervals can be different. For example, each interval can include 1000 sample identifiers, or the first interval can include 500 sample identifiers and the second interval can include 1000 sample identifiers. This application does not impose any restrictions on this.

[0117] The annotation file may also include interval statistical regions corresponding one-to-one with the sample identification intervals, and each interval statistical region includes result statistical regions corresponding one-to-one with the annotation results. For ease of understanding, assume that the annotation results are "good", "medium", and "bad", and every 1000 sample identifications form a sample identification interval in the order of the size of the sample identifications. Then, the division of the interval statistical regions and result statistical regions in the annotation file may be similar to the following:

[0118] 〝streams〞:

[0119] {

[0120] 〝index〞:0,

[0121] 〝outs〞:

[0122] {

[0123] 〝index〞:1,

[0124] 〝result〞:〝good〞,

[0125] 〝ID〞:〝0, 3, 4, 5, 7, ……, 20〞,

[0126] },

[0127] { <00者0317>〝result〞:〝medium〞,

[0129] 〝ID〞:〝1, 2〞,

[0130] },

[0131] {

[0132] 〝result〞:〝bad〞,

[0133] 〝ID〞:〝6〞,

[0134] }

[0136] },

[0137] {

[0138] 〝index〞:1,

[0139] 〝outs〞:

[0140] {

[0141] 〝result〞:〝good〞,

[0142] 〝ID〞: 〝21, ……, 24, 27, ……, 40〞,

[0143] },

[0144] {

[0145] "result": "good",

[0146] "ID": "25",

[0147] },

[0148] {

[0149] "result": "bad",

[0150] "ID": "26",

[0151] }

[0153] }

[0154] In the above example, index represents the sample identification interval, result represents the annotation result, and ID represents the sample identification. Each area where an index is located represents an interval statistical area. For example, the interval statistical area with "index": 0 can correspond to the sample identification interval 1 - 1000, and the interval statistical area with "index": 1 can correspond to the sample identification interval 1001 - 2000.

[0155] Based on the above description, the association relationship between the sample identification and the annotation result recorded in the annotation file can include:

[0156] Determine the target sample identification interval where the sample identification is located according to the divided sample identification intervals;

[0157] In the interval statistical area of the target sample identification interval, record the sample identification in the result statistical area of the annotation result to be associated.

[0158] For example, assume that the sample identification of sample data A1 is 1005 and the annotation result is "good". When recording the association relationship between the sample identification and the annotation result of sample data A1 in the annotation file, it can first be determined that the sample identification 1005 is in the sample identification interval 1001 - 2000. Therefore, in the interval statistical area corresponding to the sample identification interval 1001 - 2000, search for the result statistical area of the annotation result "good" and record the sample identification 1005 in the found result statistical area. In this way, the amount of data read per single IO can be reduced, thereby improving the data reading efficiency.

[0159] ​In some embodiments, the interval statistical region can be dynamically divided; that is, the interval statistical region corresponding to a sample identifier interval is only triggered in the annotation file when labeled sample data exists within a sample identifier interval. This avoids dividing redundant and unnecessary interval statistical regions in the annotation file. Therefore, before recording the correlation between the sample identifier and the annotation result within the interval statistical region of the target sample identifier interval, it is also possible to check whether the interval statistical region of the target sample identifier interval has been divided in the annotation file. If not, the interval statistical region of the target sample identifier interval is divided in the annotation file, and a result statistical region corresponding one-to-one with the annotation result is then divided within the interval statistical region of the target sample identifier interval.

[0160] Furthermore, based on Figure 9 As described in the relevant description, after the sample data is labeled, the labeling results can be modified. In this case, the same sample identifier and labeling result may be repeatedly written into the labeling file, leading to data conflicts. For example, suppose sample data A5 is labeled "Good" in the sample data labeling interface. In this case, the sample identifier of sample data A5 will be associated with the labeling result "Good" in the labeling file. If, after labeling sample data A5, the labeling result of sample data A5 is changed to "Medium" in the sample data labeling interface, the sample identifier of sample data A5 will be associated with the labeling result "Medium" in the labeling file. Thus, in the labeling file, the sample identifier of sample data A5 will be associated with two labeling results, causing a data conflict.

[0161] Therefore, in some embodiments, when recording the association between sample identifiers and annotation results of sample data in the annotation file, the method of this application may further include:

[0162] If a target sample identifier exists in the annotation file that is the same as the sample identifier in the sample data, and the annotation result associated with the target sample identifier is different from the annotation result in the sample data, then the target sample identifier will be deleted from the annotation file, and the association between the sample identifier and the annotation result in the sample data will be recorded in the annotation file.

[0163] For example, taking sample data A5 as an example. When recording the association between the sample identifier of sample data A5 and the annotation result "Neutral" in the annotation file, if the sample identifier of sample data A5 (i.e., the target sample identifier) ​​already exists in the annotation file, and the sample identifier of sample data A5 is associated with the annotation result "Good" in the annotation file, then the sample identifier of sample data A5 can be deleted in the annotation file, and the association between the sample identifier of sample data A5 and the annotation result "Neutral" can be recorded in the annotation file. In this way, on the one hand, when the user updates the annotation result of the sample data in the sample data annotation interface, the annotation result of the sample data can be updated synchronously in the annotation file, ensuring the accuracy of the annotation result in the annotation file; on the other hand, it can avoid the problem of associating the sample identifier of the same sample data with different annotation results in the annotation file, thus avoiding data conflicts.

[0164] Furthermore, while the annotation result associated with the target sample identifier being the same as the annotation result of the sample data may not cause data conflict, it may lead to data duplication. For example, suppose sample data A6 is labeled "Good" in the sample data annotation interface, and the sample identifier of sample data A6 is associated with the annotation result "Good" in the annotation file. If the annotation result of sample data A6 is updated in the sample data annotation interface, and the updated annotation result is still "Good", then the sample identifier of sample data A6 will be associated with the annotation result "Good" again in the annotation file. In this way, there will be two sample identifiers for sample data A6 in the annotation file, and both of these sample identifiers correspond to the annotation result "Good", thus causing the problem of duplicate sample identifiers.

[0165] Therefore, in some embodiments, if a target sample identifier exists in the annotation file, and the annotation result associated with the target sample identifier is the same as the annotation result of the sample data, then the target sample identifier is retained in the annotation file, and the association between the sample identifier and the annotation result of the sample data is stopped from being recorded in the annotation file. This avoids the problem of duplicate sample identifiers in the annotation file and reduces I / O consumption when writing data to the annotation file.

[0166] In some embodiments, the annotation file may be created when sample data in the sample dataset is first annotated. That is, when a user annotates sample data in the sample dataset for the first time in the sample data annotation interface, the creation of the annotation file corresponding to the sample dataset is triggered. Therefore, in step S142 above, before writing the sample identifier and annotation result of the sample data into the annotation file, it can be determined whether the sample data is the first sample data to be annotated in the sample dataset. If so, the creation of the annotation file is triggered, and the sample identifier and annotation result of the sample data are written into the newly created annotation file. If not, the association between the sample identifier and annotation result of the sample data can be recorded in the already created annotation file.

[0167] In some embodiments, after the annotation file is created, it is also set to a hidden state. This prevents other users from accidentally modifying or deleting the annotation file.

[0168] In some embodiments, when determining whether sample data is the first labeled sample data in the sample dataset, it can be determined whether a labeling file exists. If not, the sample data is determined to be the first labeled sample data in the sample dataset.

[0169] In other embodiments, a labeling status parameter can be set for the sample dataset. A first value for the labeling status parameter indicates that the sample data in the sample dataset is being labeled for the first time; a second value indicates that the sample data in the sample dataset is not being labeled for the first time. Before recording the association between the sample identifier and the labeling result for each sample data point, it can be determined whether the value of the labeling status parameter corresponding to the sample dataset is the first value. If so, it is determined that the sample data is the first labeled sample data in the sample dataset, and the value of the labeling status parameter is changed to the second value.

[0170] In some embodiments, when displaying sample data in the sample dataset, the dataset management module can also stop displaying the sample data in response to receiving a first exit instruction. For example, when a user exits the sample data processing interface or the sample data annotation interface, a first exit instruction can be triggered, causing the dataset management module to stop displaying the sample data. After stopping the display of sample data, the dataset management module can also respond to receiving a second display instruction to display at least a portion of the sample data in the sample dataset again. For example, when a user logs back into the sample data processing interface or the sample data annotation interface, a second display instruction can be triggered, causing the dataset management module to display at least a portion of the sample data in the sample dataset again.

[0171] In some embodiments, after the first user who performed sample data annotation exits the sample data processing interface or the sample data annotation interface, other second users may modify the sample data in the sample dataset, such as changing the sentence structure of the sample data. In this case, if the first user has already annotated the sample data in the sample dataset, the second user's modification of the sample data may affect the accuracy of the annotation results. For example, after the first user exits the sample data processing interface or the sample data annotation interface, the second user modifies the sample data A8 labeled "good," causing a decrease in the data quality of sample data A8. In this case, the annotation result of sample data A8 may no longer be accurate, but the first user is unaware of the risk of inaccurate annotation results for sample data A8.

[0172] Therefore, upon receiving the second display instruction, it can be determined whether the sample data in the sample dataset has been updated after the display of sample data was stopped. If the sample data in the sample dataset has been updated after the display of sample data was stopped, then when at least a portion of the sample data in the sample dataset is displayed again, an update prompt for the sample data will be displayed simultaneously, for example, a pop-up prompt will be displayed in the sample data processing interface. In this way, based on the update prompt, the first user can know that other users have modified the sample data after they exited the sample data processing interface or the sample data annotation interface, and can then check the accuracy of the existing annotation results. In this way, the accuracy of the annotation results can be guaranteed.

[0173] In some embodiments, a first hash value of the sample dataset can be calculated when the first user exits the sample data processing interface or the sample data annotation interface. A second hash value of the sample dataset can be calculated when the first user logs back into the sample data processing interface or the sample data annotation interface. If the first hash value and the second hash value are different, it indicates that a second user modified the sample data after the first user exited the sample data processing interface or the sample data annotation interface; if the first hash value and the second hash value are the same, it indicates that no second user modified the sample data after the first user exited the sample data processing interface or the sample data annotation interface. However, this hash value-based comparison method requires traversing all sample data in the sample dataset, resulting in a relatively large amount of data computation.

[0174] Therefore, in some other embodiments, it can be determined whether the sample data in the sample dataset has been updated based on the following method:

[0175] When the display of sample data is stopped, the first data modification time of the sample dataset is recorded in the annotation file. The first data modification time is the last modification time of the sample dataset when the display of sample data is stopped.

[0176] When displaying at least a portion of the sample data in the sample dataset again, the second data modification time of the sample dataset is obtained, which is the last modification time of the sample dataset when the second display instruction is received;

[0177] If the modification time of the first data and the modification time of the second data are different, it means that the sample data in the sample dataset was updated after the display of sample data was stopped.

[0178] Specifically, the last modification time can be attribute information of the sample dataset. When the sample data in the sample dataset changes, the last modification time of the sample dataset will change synchronously. Based on this, the first data modification time mentioned above represents the last modification time of the sample dataset when the first user exits the sample data processing interface or the sample data annotation interface. The second data modification time represents the last modification time of the sample dataset when the first user logs back into the sample data processing interface or the sample data annotation interface. If the first data modification time and the second data modification time are different, it means that the sample data in the sample dataset has changed after the first user exits the sample data processing interface or the sample data annotation interface, that is, it has been modified by the second user. If the first data modification time and the second data modification time are the same, it means that the sample data in the sample dataset has not changed after the first user exits the sample data processing interface or the sample data annotation interface, that is, it has not been modified by the second user.

[0179] In the above embodiments, the last modification time of the sample dataset is used to determine whether the sample data in the sample dataset has been updated, without having to traverse the sample data of the sample dataset, thereby reducing the amount of computation.

[0180] Specifically, the modification time of the first data can be recorded in the filename of the annotation file. For example, the filename can be in the form of "Annotation File A + Last Modified Time". In this way, when reading the association between sample identifiers and annotation results from the annotation file, the latest last modified time of the sample dataset can be obtained based on the filename of the annotation file, which is quite convenient.

[0181] In some embodiments, the method disclosed herein further includes:

[0182] Upon receiving the second exit command, the display of sample data is stopped.

[0183] If the sample data in the sample dataset is updated after the display of sample data is stopped, the changed target sample data is recorded.

[0184] In response to receiving a third display instruction, at least a portion of the sample data in the sample dataset is displayed again, and an update prompt for the target sample data is displayed at the same time.

[0185] Specifically, when displaying update notifications for target sample data, a pop-up window can be used to display the sample identifier of the target sample data, or the font, color, etc., of the target sample data can be changed. This allows the first user to distinguish the modified target sample data from other unmodified sample data based on the update notification, and thus check the annotation results of the target sample data to ensure the accuracy of the annotation results.

[0186] In the above embodiments, by recording the changed target sample data, update prompts for the target sample data can be directly displayed. In this way, the first user does not need to perform a full check of the annotation results of the sample data, which can greatly reduce the workload.

[0187] In some embodiments, the above-mentioned simultaneous display of update notifications for target sample data includes:

[0188] Based on the correlation between the sample identifier of the target sample data, the sample identifier recorded in the annotation file, and the annotation results, determine whether the target sample data has annotation results;

[0189] If the target sample data has annotation results, an update prompt for the target sample data will also be displayed.

[0190] If the target sample data does not have annotation results, no update prompt for the target sample data will be displayed.

[0191] Specifically, in these embodiments, after obtaining the changed target sample data, it is determined whether the target sample data has been labeled. If the target sample data is not labeled, the second user's modification of the target sample data will not affect the accuracy of the labeling results, so no update prompt is needed. If the target sample data is labeled, the second user's modification of the target sample data will affect the accuracy of the labeling results, therefore, an update prompt is required. In this way, on the one hand, the number of update prompts can be reduced, and on the other hand, invalid accuracy checks on the labeling results of the sample data by the first user can be avoided.

[0192] In some embodiments, after labeling the sample data in the sample dataset, the method of this application further includes:

[0193] In response to receiving a statistical instruction, based on the association between the sample identifier subgroups and the annotation results recorded in the annotation file, the number of sample identifiers corresponding to each annotation result is counted.

[0194] Display the number of sample labels corresponding to each annotation result.

[0195] For details, please refer to Figure 13 The relevant descriptions are not repeated here.

[0196] In some embodiments, sample data with target annotation results are also used for paginated display in the sample data processing interface; the above-mentioned filtering of sample data with target annotation results in the sample dataset includes:

[0197] Based on the number of sample data to be displayed on the current page, select the corresponding number of sample data with target annotation results from the sample dataset.

[0198] In other words, when displaying sample data, the system can acquire sample data with target annotations in batches, based on the number of sample data currently available for display on the sample data processing interface, without needing to acquire the full amount of sample data with target annotations. This reduces the amount of data acquired at once and improves data processing efficiency.

[0199] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0200] See also Figure 15 This is a sample data processing apparatus provided in some embodiments of this application. Figure 15 The sample data processing device includes:

[0201] The display module 151 is used to display at least a portion of the sample data in the sample dataset in response to receiving a first display instruction. The sample dataset includes multiple sample data items, and each sample data item has a sample identifier.

[0202] The association module 152 is used to respond to the display of sample data being labeled, record the association between sample identifiers and labeling results in the labeling file, and the labeling results represent the data category to which the sample data belongs. The labeling file includes result statistics areas that correspond one-to-one with the labeling results. When recording the association between sample identifiers and labeling results, each sample identifier is recorded in the result statistics area of ​​the labeling results to be associated, and the sample identifiers in each result statistics area are stored in a contiguous storage area of ​​the storage medium.

[0203] The filtering module 153 is used to respond to receiving a filtering instruction including target annotation results, filter sample data with target annotation results in the sample dataset according to the correlation between sample identifiers and annotation results recorded in the annotation file, and process the filtered sample data.

[0204] In this embodiment, the sample data processing device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0205] For a description of the features in the embodiment corresponding to the sample data processing device, please refer to the relevant description in the embodiment corresponding to the sample data processing method, which will not be repeated here.

[0206] Embodiments of this application also provide an electronic device, such as... Figure 16 As shown, it includes a memory 10 and a processor 20. The memory 10 stores a computer program, and the processor 20 is configured to run the computer program to perform the steps in any of the above-described sample data processing method embodiments.

[0207] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described sample data processing method embodiments when it is run.

[0208] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0209] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described sample data processing method embodiments.

[0210] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described sample data processing method embodiments.

[0211] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0212] The foregoing has provided a detailed description of a sample data processing method, apparatus, device, storage medium, and program product provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A sample data processing method, characterized in that, The method includes: In response to receiving a first display instruction, at least a portion of the sample data in the sample dataset is displayed, the sample dataset including multiple sample data, each sample data having a sample identifier, and in the sample dataset, the sample identifier of the sample data is divided into multiple sample identifier intervals; In response to the displayed sample data being labeled, the association between the sample identifier and the labeling result is recorded in the labeling file. The labeling result represents the data category to which the sample data belongs, and the data category is divided according to the quality of the sample data. The labeling file includes interval statistical areas corresponding one-to-one with the sample identifier intervals. Each interval statistical area includes a result statistical area corresponding one-to-one with the labeling result. When recording the association between the sample identifier and the labeling result, the target sample identifier interval where the sample identifier is located is determined according to the divided sample identifier intervals. The sample identifier is recorded in the result statistical area of ​​the labeling result to be associated within the interval statistical area of ​​the target sample identifier interval. The sample identifier in each result statistical area is stored in a contiguous storage area of ​​the storage medium. When reading the association between the sample identifier and the labeling result in the contiguous storage area, multiple associations between sample identifiers and labeling results can be read in a single reading process. In response to receiving a filtering instruction including target annotation results, the system filters sample data with target annotation results from the sample dataset according to the correlation between sample identifiers and annotation results recorded in the annotation file, and processes the filtered sample data, wherein processing the filtered sample data includes using the filtered sample data as model training data.

2. The method according to claim 1, characterized in that, Before recording the correlation between the sample identifiers and annotation results of the sample data in the annotation file, the method further includes: Determine whether the sample data is the first labeled sample data in the sample dataset. If so, trigger the creation of the label file and set the label file to a hidden state.

3. The method according to claim 2, characterized in that, The step of determining whether the sample data is the first labeled sample data in the sample dataset includes: Determine if the annotation file exists; if not, determine that the sample data is the first annotated sample data in the sample dataset; or Determine whether the value of the annotation status parameter corresponding to the sample dataset is the first value. If so, determine that the sample data is the first labeled sample data in the sample dataset, and change the value of the annotation status parameter to the second value.

4. The method according to claim 1, characterized in that, Before recording the correlation between the sample identifiers and annotation results of the sample data in the annotation file, the method further includes: Check whether the annotation file divides the interval statistical region of the target sample identification interval. If not, divide the interval statistical region of the target sample identification interval in the annotation file, and divide the result statistical region corresponding one-to-one with the annotation result in the interval statistical region of the target sample identification interval.

5. The method according to claim 1, characterized in that, When recording the correlation between the sample identifier and the annotation result of the sample data in the annotation file, the method further includes: If the annotation file contains a target sample identifier that is the same as the sample identifier of the sample data, and the annotation result associated with the target sample identifier is different from the annotation result of the sample data, then the target sample identifier is deleted from the annotation file, and the association between the sample identifier and the annotation result of the sample data is recorded in the annotation file.

6. The method according to claim 5, characterized in that, The method further includes: If the target sample identifier exists in the annotation file, and the annotation result associated with the target sample identifier is the same as the annotation result of the sample data, then the target sample identifier will continue to be retained in the annotation file, and the association between the sample identifier and the annotation result of the sample data will be stopped in the annotation file.

7. The method according to claim 1, characterized in that, The method further includes: Upon receiving the first exit command, the display of the sample data is stopped; In response to receiving a second display instruction, at least a portion of the sample data in the sample dataset is displayed again, wherein if the sample data in the sample dataset is updated after the display of the sample data is stopped, an update prompt for the sample data is displayed simultaneously when at least a portion of the sample data in the sample dataset is displayed again.

8. The method according to claim 7, characterized in that, The following method is used to determine whether the sample data in the sample dataset has been updated: When the display of the sample data is stopped, the first data modification time of the sample dataset is recorded in the annotation file. The first data modification time is the last modification time of the sample dataset when the display of the sample data is stopped. When displaying at least a portion of the sample data in the sample dataset again, a second data modification time of the sample dataset is obtained, wherein the second data modification time is the last modification time of the sample dataset when the second display instruction is received; If the modification time of the first data is different from the modification time of the second data, it means that the sample data in the sample dataset was updated after the display of the sample data was stopped.

9. The method according to claim 1, characterized in that, The method further includes: Upon receiving a second exit command, the display of the sample data is stopped; If the sample data in the sample dataset is updated after the display of the sample data is stopped, the changed target sample data is recorded. In response to receiving a third display instruction, at least a portion of the sample data in the sample dataset is displayed again, and an update prompt for the target sample data is displayed simultaneously.

10. The method according to claim 9, characterized in that, The simultaneous display of update notifications for the target sample data includes: Based on the correlation between the sample identifier of the target sample data, the sample identifier recorded in the annotation file, and the annotation results, it is determined whether the target sample data has annotation results; If the target sample data has annotation results, an update prompt for the target sample data will be displayed simultaneously. If the target sample data does not have annotation results, no update prompt for the target sample data will be displayed.

11. The method according to claim 1, characterized in that, After labeling the sample data in the sample dataset, the method further includes: In response to receiving a statistical instruction, based on the correlation between sample identifiers and annotation results recorded in the annotation file, the number of sample identifiers corresponding to each annotation result is counted; Display the number of sample labels corresponding to each annotation result.

12. The method according to claim 1, characterized in that, The sample data with the target annotation results is also used for paginated display in the sample data processing interface; The step of filtering sample data with the target annotation results in the sample dataset includes: Based on the number of sample data to be displayed on the current page, a corresponding number of sample data with the target annotation results are selected from the sample dataset.

13. The method according to claim 1, characterized in that, The method is applied to a sample data processing system, which includes an initial data management module and a dataset management module. The initial data management module is used to store sample data, and the dataset management module is used to process the sample data. Before displaying at least a portion of the sample data in the sample dataset, the method further includes: In response to the selection of one or more sample data in the initial data management module, the selected sample data is imported into the dataset management module as sample data in the sample dataset. The dataset management module is used to display at least a portion of the sample data in the sample dataset.

14. The method according to claim 13, characterized in that, The process of importing the selected sample data into the dataset management module includes: The selected sample data is imported into the dataset management module from the storage path in the initial data management module. When displaying at least a portion of the sample data in the sample dataset, the dataset management module retrieves the sample data to be displayed from the initial data management module based on the storage path.

15. The method according to claim 14, characterized in that, After importing the storage path into the dataset management module, the method further includes: In the initial data management module, selected sample data is identified to distinguish between selected and unselected sample data in the initial data management module.

16. A sample data processing apparatus, characterized in that, The device includes: A display module is configured to, in response to receiving a first display instruction, display at least a portion of the sample data in the sample dataset, the sample dataset including multiple sample data, each sample data having a sample identifier, and in the sample dataset, the sample identifier of the sample data being divided into multiple sample identifier intervals; The association module is used to respond to the displayed sample data being labeled, and to record the association between the sample identifier and the labeling result in the labeling file. The labeling result represents the data category to which the sample data belongs, and the data category is divided according to the quality of the sample data. The labeling file includes interval statistical areas corresponding one-to-one with the sample identifier intervals, and each interval statistical area includes a result statistical area corresponding one-to-one with the labeling result. When recording the association between the sample identifier and the labeling result, the target sample identifier interval where the sample identifier is located is determined according to the divided sample identifier intervals, and the sample identifier is recorded in the result statistical area of ​​the labeling result to be associated within the interval statistical area of ​​the target sample identifier interval. The sample identifier in each result statistical area is stored in a contiguous storage area of ​​the storage medium. When reading the association between the sample identifier and the labeling result in the contiguous storage area, multiple associations between sample identifiers and labeling results can be read in a single reading process. The filtering module is used to respond to receiving a filtering instruction including target annotation results, and to filter sample data with target annotation results in the sample dataset according to the correlation between sample identifiers and annotation results recorded in the annotation file, and to process the filtered sample data, wherein processing the filtered sample data includes using the filtered sample data as model training data.

17. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the sample data processing method of any one of claims 1 to 15.

18. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer instructions for causing a computer to perform the sample data processing method according to any one of claims 1 to 15.

19. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the sample data processing method according to any one of claims 1 to 15.

Images