Label determination method, apparatus and electronic device
By employing a multi-dimensional label analysis and comprehensive judgment method, the problem of poor accuracy in recipe labels caused by a single model was solved, thereby improving the accuracy and reliability of recipe labels.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NINGBO FOTILE KITCHEN WARE CO LTD
- Filing Date
- 2026-01-07
- Publication Date
- 2026-06-05
AI Technical Summary
Existing recipe labeling methods rely on a single model, resulting in poor accuracy of recipe labels and difficulty in handling diverse target text features.
Multi-dimensional label analysis is adopted. Multi-dimensional label mining is performed through a first preset label analysis model. Accuracy analysis is performed by combining the second preset label analysis model under each preset dimension. Based on the first analysis data and the indicator data of multiple second preset label analysis models, a comprehensive judgment is made to determine the target label.
It significantly improves the accuracy, reliability, and completeness of the coverage of recipe labels, ensuring the scientific and precise nature of label selection.
Smart Images
Figure CN122153680A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of computer technology, and in particular to a tag identification method, apparatus, and electronic device. Background Technology
[0002] With the widespread application of digital technology in the food industry, recipe tagging has become a key technology for achieving accurate recipe retrieval and efficient management. Currently, most mainstream recipe tagging methods rely on a single large model to complete the tag prediction task. This method trains a specific model to extract features and classify tags from the target text, directly outputting the prediction results to achieve basic tag annotation. However, this tagging approach completely entrusts all tag prediction to a single model. If the model suffers from training data bias or semantic understanding limitations, it struggles to effectively handle diverse target text features, resulting in poor accuracy of the determined recipe tags. Summary of the Invention This disclosure provides a label determination method, apparatus, and electronic device to at least solve the problem of poor accuracy of recipe labels in related technologies.
[0003] According to a first aspect of the present disclosure, a tag determination method is provided, comprising: Input the recipe to be tagged into the first preset tag analysis model, perform multi-dimensional tag analysis on the recipe to be tagged, and obtain at least one candidate recipe tag corresponding to each of the multiple preset dimensions of the recipe to be tagged. Each candidate recipe tag under each preset dimension is input into the second preset tag analysis model corresponding to each preset dimension. The accuracy of each candidate recipe tag is analyzed, and the first analysis data corresponding to each candidate recipe tag is output. The first analysis data indicates the probability that each candidate recipe tag belongs to the tag of the recipe to be tagged under the corresponding preset dimension. Based on the first analysis data and the first preset index data corresponding to each of the multiple second preset label analysis models, the target label is determined from the at least one candidate recipe label corresponding to each of the multiple preset dimensions. The first preset index data characterizes the analysis performance of the corresponding second preset label analysis model on the recipe label.
[0004] According to a second aspect of the present disclosure, a tag determining apparatus is provided, comprising: The candidate recipe tag acquisition module is used to input the recipe to be tagged into the first preset tag analysis model, perform multi-dimensional tag analysis on the recipe to be tagged, and obtain at least one candidate recipe tag corresponding to the recipe to be tagged in each of the multiple preset dimensions. The first analysis data acquisition module is used to input each candidate recipe label under each preset dimension into the second preset label analysis model corresponding to each preset dimension, analyze the accuracy of each candidate recipe label, and output the first analysis data corresponding to each candidate recipe label. The first analysis data indicates the probability that each candidate recipe label belongs to the label of the recipe to be labeled under the corresponding preset dimension. The target tag acquisition module is used to determine the target tag from the at least one candidate recipe tag corresponding to each of the multiple preset dimensions based on the first analysis data and the first preset index data corresponding to each of the multiple second preset tag analysis models. The first preset index data characterizes the analysis performance of the corresponding second preset tag analysis model on the recipe tag.
[0005] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the method as described in any one of the first aspects above.
[0006] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided such that, when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the method described in any of the first aspects of the present disclosure. According to a fifth aspect of the present disclosure, a computer program product including instructions is provided that, when run on a computer, causes the computer to perform the method described in any of the first aspects of the present disclosure.
[0007] The technical solutions provided by the embodiments of this disclosure bring at least the following beneficial effects: First, the recipes to be tagged are input into a first preset tag analysis model to perform multi-dimensional tag analysis, obtaining at least one candidate recipe tag for each of the multiple preset dimensions. This ensures extensive and in-depth tag mining of the recipes to be tagged, effectively generating a set of candidate recipe tags covering all preset dimensions, laying a comprehensive foundation for subsequent accurate selection. Second, for each candidate recipe tag under each preset dimension, a second preset tag analysis model dedicated to that preset dimension performs precise confidence assessment, specifically for analyzing and outputting first analysis data. This first analysis data indicates the probability that each candidate recipe tag belongs to the tag of the recipe to be tagged under the corresponding preset dimension. Finally, by combining the first preset index data representing the analysis performance of the recipe tags by the corresponding second preset tag analysis model with the real-time generated first analysis data, a comprehensive judgment is made to determine the target tag from at least one candidate recipe tag corresponding to each of the multiple preset dimensions. Based on the collaborative decision-making mechanism of the first analysis data and the first preset index data of each second preset tag analysis model, the accuracy, reliability, and completeness of the coverage dimensions of the target tag are significantly improved.
[0008] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description
[0009] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.
[0010] Figure 1 This is a flowchart illustrating a label determination method according to an exemplary embodiment; Figure 2 This is a schematic diagram illustrating a process for obtaining at least one candidate recipe label corresponding to each of multiple preset dimensions of a recipe to be labeled, according to an exemplary embodiment. Figure 3 This is a schematic diagram illustrating a process for determining first preset indicator data according to an exemplary embodiment; Figure 4 This is a schematic diagram illustrating a process for determining a target label from at least one candidate recipe label corresponding to multiple preset dimensions, according to an exemplary embodiment. Figure 5 This is a schematic diagram illustrating another process for determining a target label from at least one candidate recipe label corresponding to multiple preset dimensions, according to an exemplary embodiment. Figure 6This is a schematic diagram illustrating another process for determining a target label from at least one candidate recipe label corresponding to multiple preset dimensions, according to an exemplary embodiment. Figure 7 This is a schematic diagram illustrating a process for determining a preset comparison threshold according to an exemplary embodiment; Figure 8 This is a schematic diagram illustrating a process of determining a target label analysis model and updating the target label analysis model to a first preset label analysis model, according to an exemplary embodiment. Figure 9 This is a block diagram illustrating a label determining device according to an exemplary embodiment; Figure 10 This is a block diagram illustrating an electronic device for tag identification according to an exemplary embodiment. Detailed Implementation
[0011] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.
[0012] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar different contents and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.
[0013] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.
[0014] Figure 1 This is a flowchart illustrating a label determination method according to an exemplary embodiment, such as... Figure 1 As shown, this label determination method is used in electronic devices such as servers and includes the following steps.
[0015] In step S101, the recipe to be tagged is input into the first preset tag analysis model, and multi-dimensional tag analysis is performed on the recipe to be tagged to obtain at least one candidate recipe tag corresponding to each of the multiple preset dimensions.
[0016] In a specific embodiment, such as Figure 2 As shown, the above method also includes: In step S201, keywords are extracted from the recipes to be tagged to obtain the target keywords for the recipes to be tagged.
[0017] In one specific embodiment, the above-mentioned keyword extraction of the recipe to be tagged to obtain the target keywords of the recipe to be tagged may include: inputting the recipe to be tagged into a preset keyword extraction model, extracting keywords from the recipe to be tagged, and obtaining the target keywords of the recipe to be tagged.
[0018] In one specific embodiment, the aforementioned preset keyword extraction model can be a deep learning network for extracting keywords from the recipe to be tagged, such as a convolutional neural network and a recurrent neural network.
[0019] In an optional embodiment, the aforementioned preset keyword extraction model can be trained based on multiple sample recipes to be labeled and the corresponding labeled keywords of each of the multiple sample recipes to be labeled.
[0020] In step S203, the recipe features are expanded based on the target keywords to obtain the expanded feature information of the recipe to be tagged.
[0021] In a specific embodiment, the above-mentioned recipe feature expansion based on target keywords to obtain extended feature information of the recipe to be tagged may include: inputting target keywords into a preset feature expansion model, expanding the recipe features of the recipe to be tagged, and generating extended feature information.
[0022] In a specific embodiment, the aforementioned preset feature expansion model can be a deep learning network for expanding the recipe features of the recipe to be tagged, such as a convolutional neural network and a recurrent neural network.
[0023] In an optional embodiment, the aforementioned preset feature expansion model can be trained based on the sample keywords corresponding to each of the multiple sample recipes to be labeled and the target expanded feature information corresponding to each of the multiple sample recipes to be labeled.
[0024] In a specific embodiment, the above-mentioned inputting the recipe to be tagged into a first preset tag analysis model and performing multi-dimensional tag analysis on the recipe to be tagged to obtain at least one candidate recipe tag corresponding to the recipe to be tagged in each of multiple preset dimensions includes: In step S205, the target keywords and extended feature information are input into the first preset tag analysis model. Based on the multi-dimensional tag analysis of the recipe to be tagged, the tags are extended to obtain at least one candidate recipe tag corresponding to each of the multiple preset dimensions of the recipe to be tagged.
[0025] In a specific embodiment, the aforementioned first preset label analysis model can be a deep learning network that expands the labels based on multi-dimensional label analysis of the recipe to be labeled, such as convolutional neural networks and recurrent neural networks.
[0026] In an optional embodiment, the first preset label analysis model can be trained based on the sample keywords, sample extended feature information, and at least one target candidate recipe label corresponding to each of the multiple sample recipes to be labeled in multiple preset dimensions.
[0027] For example, multiple preset dimensions may include at least two of the following dimensions: ingredient characteristics, processing attributes, and flavor components.
[0028] In the above embodiments, target keywords are obtained by extracting keywords from the recipe to be tagged, and extended feature information is obtained by expanding the recipe features based on the target keywords. Then, the target keywords and extended feature information are input into the first preset tag analysis model. Tag expansion is achieved while performing multi-dimensional tag analysis, which can deeply explore the core content of the recipe to be tagged, supplement the feature details of the recipe from more angles, and make the multi-dimensional tag analysis more comprehensive and in-depth. This greatly enriches the types and number of candidate recipe tags, effectively improves the coverage of recipe features by candidate recipe tags, and ensures that the obtained candidate recipe tags can more accurately and completely reflect the characteristics of the recipe to be tagged in multiple preset dimensions, significantly improving the comprehensiveness and accuracy of recipe tag annotation.
[0029] In step S103, each candidate recipe label under each preset dimension is input into the second preset label analysis model corresponding to each preset dimension, the accuracy of each candidate recipe label is analyzed, and the first analysis data corresponding to each candidate recipe label is output.
[0030] In a specific embodiment, the first analysis data indicates the probability that each candidate recipe label belongs to the label of the recipe to be labeled under the corresponding preset dimension. The first analysis data is a numerical probability value in the interval [0,1].
[0031] In a specific embodiment, the aforementioned second preset label analysis model can be a deep learning network, such as a convolutional neural network or a recurrent neural network, that analyzes the accuracy of each candidate recipe label under its respective preset dimension.
[0032] In an optional embodiment, the second preset label analysis model described above can be trained based on the candidate recipe labels for each sample under each preset dimension and the target analysis data corresponding to each candidate recipe label.
[0033] In step S105, based on the first analysis data and the first preset index data corresponding to each of the multiple second preset label analysis models, the target label is determined from at least one candidate recipe label corresponding to each of the multiple preset dimensions.
[0034] In one specific embodiment, the aforementioned first preset index data characterizes the analysis performance of the corresponding second preset label analysis model on recipe labels.
[0035] In a specific embodiment, such as Figure 3 As shown, the aforementioned first preset indicator data is updated according to a first preset period, and the first preset indicator data is determined in the following manner: In step S301, when the first update time of the first preset indicator data is reached, the historical indicator data, the number of first tags, the number of second tags, and the number of third tags corresponding to the multiple second preset tag analysis models within the first preset period before the first update time are obtained.
[0036] In one specific embodiment, the first number of tags is the number of tags of the first historical candidate tags analyzed by each of the multiple second preset tag analysis models, the second number of tags is the number of tags of the first historical candidate tags that belong to the corresponding first historical recipe, and the third number of tags is the number of tags of the second historical candidate tags that belong to the corresponding second historical recipe.
[0037] In step S303, based on the number of first tags and the number of second tags, the first historical labeling accuracy rate of each of the multiple second preset tag analysis models within the first preset period before the first update time is determined.
[0038] In a specific embodiment, the above-mentioned determination of the first historical labeling accuracy rate of each of the multiple second preset label analysis models within the first preset period before the first update time based on the first label number and the second label number may include: determining the first historical labeling accuracy rate of each of the multiple second preset label analysis models within the first preset period before the first update time based on the formula: first historical labeling accuracy rate = second label number / first label number.
[0039] In step S305, based on the first historical labeling accuracy rate and the first preset adjustment coefficient, the first target coefficient corresponding to each of the multiple second preset label analysis models is determined.
[0040] In one specific embodiment, the aforementioned first preset adjustment coefficient can be a preset benchmark scaling factor used to constrain the value range of the first target coefficient. Specifically, the first preset adjustment coefficient can be 0.1, and can be determined from historical experience data. The aforementioned first target coefficient is used to constrain the growth rate of the first preset indicator data.
[0041] In a specific embodiment, determining the first target coefficient corresponding to each of the multiple second preset label analysis models based on the first historical labeling accuracy rate and the first preset adjustment coefficient may include: based on the formula Determine the first target coefficients corresponding to each of the multiple second preset label analysis models, where β i Let A be the first target coefficient corresponding to the i-th second preset label analysis model. i Let max(A) be the first historical labeling accuracy corresponding to the i-th second preset label analysis model. b A c ..., A y The first historical labeling accuracy is the highest among the first historical labeling accuracy rates corresponding to all the second preset labeling analysis models.
[0042] In step S307, the first preset indicator data is determined based on the first target coefficient, the number of third labels, and historical indicator data.
[0043] In a specific embodiment, determining the first preset indicator data based on the first target coefficient, the number of third labels, and historical indicator data may include: determining the first preset indicator data based on the formula: First preset indicator data = Historical indicator data + First target coefficient * Number of third labels.
[0044] In the above embodiments, by updating the first preset indicator data according to the first preset period, and obtaining relevant historical indicator data and label quantity information within the historical period when the update time is reached, the historical labeling performance of multiple second preset label analysis models can be dynamically evaluated based on real historical data. By calculating the first historical labeling accuracy rate and combining it with the first preset adjustment coefficient to determine the first target coefficient, the growth rate of the indicator data is effectively constrained. This allows the first preset indicator data to be adjusted in real time according to the changes in the actual analysis performance of the second preset label analysis model. This fully considers the historical performance of the model and avoids abnormal fluctuations in the indicator data through the adjustment mechanism. Ultimately, it ensures that the first preset indicator data accurately reflects the current analysis performance of the model, providing a scientific, reliable, and dynamically updated quantitative basis for subsequently determining model weights and target labels, and ensuring the rationality of model weight allocation and the accuracy of label selection during the label analysis process.
[0045] In a specific embodiment, such as Figure 4As shown, the above-mentioned determination of target tags from at least one candidate recipe tag corresponding to each of multiple preset dimensions, based on the first analysis data and the first preset indicator data corresponding to each of the multiple second preset tag analysis models, includes: In step S401, based on the first preset index data, the model weights corresponding to each of the multiple second preset label analysis models are determined.
[0046] In a specific embodiment, determining the model weights corresponding to each of the multiple second preset label analysis models based on the first preset index data may include: based on the formula Determine the model weights corresponding to each of the multiple second-preset label analysis models, among which I represents the model weights corresponding to the i-th second preset label analysis model. i For the first preset index data corresponding to the i-th second preset label analysis model, max(I) b I c ...,I y The highest first preset index data is the first preset index data corresponding to each of the second preset label analysis models.
[0047] In step S403, based on the first analysis data and model weights, the target label is determined from at least one candidate recipe label corresponding to each of multiple preset dimensions.
[0048] In the above embodiments, by determining the model weights corresponding to each of the multiple second preset label analysis models based on the first preset index data, the differences in the performance of different second preset label analysis models in recipe label analysis can be fully considered. Each second preset label analysis model is assigned a weight value that matches its analysis capability. Then, when filtering labels in combination with the first analysis data, the label accuracy evaluation results output by each model can be scientifically weighted according to the model weights. This makes the analysis process consider both the accuracy of the candidate recipe labels themselves and the differences in the analysis performance of the corresponding models, thereby filtering out target labels with higher overall credibility from candidate recipe labels of multiple preset dimensions, effectively improving the scientificity, accuracy and reliability of label filtering.
[0049] In one specific embodiment, when each candidate recipe tag corresponds to one of multiple preset dimensions, such as... Figure 5 As shown, the above determination of target tags from at least one candidate recipe tag corresponding to each of multiple preset dimensions, based on the first analysis data and model weights, includes: In step S501, based on the first analysis data and model weights, the second analysis data corresponding to each candidate recipe label is determined.
[0050] In a specific embodiment, the above-mentioned determination of the second analysis data corresponding to each candidate recipe tag based on the first analysis data and model weights may include: determining the second analysis data corresponding to each candidate recipe tag based on the formula: second analysis data = first analysis data * model weights.
[0051] In step S503, based on the second analysis data and the preset comparison threshold, the target label is determined from at least one candidate recipe label corresponding to each of the multiple preset dimensions.
[0052] In one specific embodiment, determining the target label from at least one candidate recipe label corresponding to each of multiple preset dimensions based on the second analysis data and a preset comparison threshold may include: taking the candidate recipe labels whose second analysis data is greater than or equal to the preset comparison threshold as the target label.
[0053] In the above embodiments, by determining the second analysis data corresponding to each candidate recipe tag based on the first analysis data and model weights, the accuracy evaluation of the candidate tags in the corresponding preset dimensions can be combined with the differences in the analytical performance of the model itself. This allows the second analysis data to fully integrate both tag applicability and model credibility information, providing a more comprehensive evaluation basis for tag selection. Furthermore, by filtering the second analysis data of each candidate recipe tag using a preset comparison threshold, tags with better overall evaluation results can be accurately identified from candidate tags across multiple preset dimensions. This effectively eliminates interference from low-credibility candidate tags, ensuring that the determined target tags have both high descriptive accuracy in the corresponding dimensions and reliable support for model analytical performance, significantly improving the accuracy and effectiveness of target tag selection.
[0054] In one specific embodiment, when the target candidate label corresponds to at least two of a plurality of preset dimensions, the target candidate label is any one of the candidate recipe labels among at least one candidate recipe label corresponding to each of the plurality of preset dimensions, such as... Figure 6 As shown, the above-mentioned determination of the target label from at least one candidate recipe label corresponding to multiple preset dimensions, based on the first analysis data and model weights, also includes: In step S601, based on the second analysis data and the first preset index data corresponding to the target candidate label, the third analysis data corresponding to the target candidate label is determined.
[0055] In a specific embodiment, determining the third analysis data corresponding to the target candidate label based on the second analysis data and the first preset indicator data corresponding to the target candidate label may include: based on the formula The third analysis data corresponding to the target candidate label is determined, among which, The third analysis data corresponding to the target candidate labels. The second analysis data corresponding to the target candidate labels. The first preset index data of the analysis model for the second preset label corresponding to at least two preset dimensions of the target candidate label.
[0056] In a specific embodiment, determining the target label from at least one candidate recipe label corresponding to each of multiple preset dimensions based on the second analysis data and a preset comparison threshold includes: In step S603, the target label is determined from at least one candidate recipe label corresponding to each of the multiple preset dimensions, based on the third analysis data corresponding to the target candidate label, the second analysis data corresponding to other candidate labels, and the preset comparison threshold.
[0057] In one specific embodiment, the aforementioned other candidate tags are candidate recipe tags other than the target candidate tag among at least one candidate recipe tag corresponding to each of multiple preset dimensions.
[0058] In a specific embodiment, the above-mentioned determination of the target label from at least one candidate recipe label corresponding to multiple preset dimensions based on the third analysis data corresponding to the target candidate label, the second analysis data corresponding to other candidate labels, and the preset comparison threshold may include: taking other candidate labels whose second analysis data is greater than or equal to the preset comparison threshold, and the target candidate label whose third analysis data is greater than or equal to the preset comparison threshold as the target label.
[0059] In the above embodiments, when the target candidate label corresponds to at least two preset dimensions, the third analysis data is determined based on its second analysis data and the first preset index data. This allows for a deep integration of cross-dimensional label accuracy evaluation and model historical analysis performance, enabling the third analysis data to comprehensively reflect the label's overall adaptability across multiple dimensions. Furthermore, by combining the second analysis data of other candidate labels with preset comparison thresholds for screening, the comprehensive performance of different candidate labels in single-dimensional and cross-dimensional scenarios can be effectively considered. This eliminates interference from labels that perform well only in a single dimension but lack cross-dimensional adaptability, ensuring that the final target label not only reflects a high-accuracy description of the corresponding dimension but also meets the comprehensive applicability requirements of multi-dimensional related scenarios. This significantly improves the comprehensiveness, coordination, and accuracy of multi-dimensional label screening.
[0060] In a specific embodiment, such as Figure 7 As shown, the aforementioned preset comparison threshold is updated according to a first preset period, and the preset comparison threshold is determined in the following manner: In step S701, when the first update time reaches the preset comparison threshold, the historical comparison threshold, historical target coefficient, number of third labels and number of fourth labels corresponding to multiple second preset label analysis models within the first preset period before the first update time are obtained.
[0061] In one specific embodiment, the third number of tags refers to the number of recipe tags belonging to the corresponding second historical recipe among the second historical candidate tags analyzed by multiple second preset tag analysis models, and the fourth number of tags refers to the number of tags of the second historical candidate tags. The initial value of the historical target coefficient can be 1.
[0062] In step S703, based on the number of third and fourth tags, the second historical labeling accuracy of multiple second preset tag analysis models within a first preset period before the first update time is determined.
[0063] In a specific embodiment, determining the second historical labeling accuracy of multiple second preset label analysis models within a first preset period before the first update time based on the number of third labels and the number of fourth labels may include: determining the second historical labeling accuracy of multiple second preset label analysis models within a first preset period before the first update time based on the formula: second historical labeling accuracy = number of third labels / number of fourth labels.
[0064] In step S705, the second target coefficient is determined based on the second historical marking accuracy rate, the historical target coefficient, and the preset accuracy rate.
[0065] In one specific embodiment, the aforementioned second target coefficient is used to constrain the update magnitude of the preset comparison threshold.
[0066] In a specific embodiment, determining the second target coefficient based on the second historical marking accuracy rate, the historical target coefficient, and the preset accuracy rate may include: based on a formula. Determine the second objective coefficient.
[0067] In step S707, a preset comparison threshold is determined based on the historical comparison threshold, the second target coefficient, the second historical marking accuracy, and the preset accuracy.
[0068] In a specific embodiment, determining the preset comparison threshold based on the historical comparison threshold, the second target coefficient, the second historical marking accuracy rate, and the preset accuracy rate may include: preset comparison threshold = historical comparison threshold + second target coefficient * (second historical marking accuracy rate - preset accuracy rate), thus obtaining the preset comparison threshold.
[0069] In the above embodiments, by updating the preset comparison threshold according to the first preset period, information such as the historical comparison threshold, historical target coefficient, and number of related labels within the historical period is obtained when the update time is reached. The threshold benchmark can be dynamically adjusted based on the model's actual historical labeling performance. By calculating the second historical labeling accuracy and combining the historical target coefficient with the preset accuracy, the second target coefficient is determined, effectively constraining the update range of the preset comparison threshold and avoiding unreasonable fluctuations. This allows the preset comparison threshold to reflect the actual labeling level of multiple second preset label analysis models within the historical period in real time. This provides a scientific, reasonable, and adaptive evaluation benchmark for subsequent target label selection based on the second or third analysis data, ensuring that low-confidence candidate labels can be effectively filtered during the label selection process while fully considering the dynamic performance of the model. This significantly improves the accuracy, stability, and reliability of the target label determination process.
[0070] In one specific embodiment, the first preset label analysis model is updated according to a second preset period. After the labels are determined after the second preset period, such as Figure 8 As shown, the above method also includes: In step S801, when the second update time of the first preset label analysis model is reached, the second preset index data corresponding to each of the multiple second preset label analysis models, the number of fifth labels and the number of sixth labels corresponding to the first preset label analysis model within the second preset period before the second update time are obtained.
[0071] In a specific embodiment, the number of fifth tags is the number of tags of the third historical candidate tags output by the first preset tag analysis model, and the number of sixth tags is the number of tags of the recipe tags belonging to the corresponding third historical recipe among the third historical candidate tags.
[0072] In step S803, based on the number of fifth and sixth tags, the third historical tagging accuracy of the first preset tag analysis model within the second preset period before the second update time is determined.
[0073] In a specific embodiment, determining the third historical labeling accuracy of the first preset label analysis model within a second preset period before the second update time based on the number of fifth labels and the number of sixth labels may include: determining the third historical labeling accuracy of the first preset label analysis model within a second preset period before the second update time based on the formula: third historical labeling accuracy = number of sixth labels / number of fifth labels.
[0074] In step S805, the third target coefficient is determined based on the second preset index data, the third historical marking accuracy rate, the second preset adjustment coefficient, the third preset adjustment coefficient, and the preset accuracy rate.
[0075] In one specific embodiment, the second preset adjustment coefficient serves as a constant benchmark for calculating the third target coefficient. The third preset adjustment coefficient constrains the impact of updates to the second preset indicator data within a preset time period on the third target coefficient. The third target coefficient controls the update frequency of the first preset label analysis model. For example, the second preset adjustment coefficient can be 0.6, determined from historical experience data. The third preset adjustment coefficient can be 0.5, also determined from historical experience data.
[0076] In a specific embodiment, determining the third target coefficient based on the second preset indicator data, the third historical labeling accuracy rate, the second preset adjustment coefficient, the third preset adjustment coefficient, and the preset accuracy rate may include: first, determining the standard deviation of all second preset indicator data and the sum of all second preset indicator data based on the second preset indicator data corresponding to each second preset label analysis model; then, based on the formula... Determine the third objective coefficient.
[0077] In step S807, based on the third target coefficient and the second preset index data, a target label analysis model is determined from multiple second preset label analysis models, and the target label analysis model is updated to the first preset label analysis model.
[0078] In a specific embodiment, the above-mentioned determination of the target label analysis model from multiple second preset label analysis models based on the third target coefficient and the second preset index data, and updating the target label analysis model to the first preset label analysis model may include: firstly, based on the second preset index data corresponding to each second preset label analysis model, determining the sum of the index data of the second preset index data of all second preset label analysis models; then determining at least one candidate label analysis model whose second preset index data is ≥ the third target coefficient * the sum of index data; and finally, taking the candidate label analysis model with the highest second preset index data as the target label analysis model.
[0079] In the above embodiments, by updating the first preset label analysis model according to the second preset period, obtaining the relevant historical label quantity and calculating the third historical labeling accuracy when the update time is reached, the analysis performance can be dynamically evaluated based on the actual labeling performance of the model within the historical period. The third target coefficient is determined by combining the second preset index data and the preset adjustment coefficient, which effectively controls the update frequency of the model and avoids analysis deviations caused by over-updating or under-updating. This allows the first preset label analysis model to adapt to changes in the label analysis scenario in real time, accurately selects the target label analysis model with better overall performance from multiple second preset label analysis models for updating, and continuously improves the accuracy and effectiveness of the model in the process of multi-dimensional label analysis and expansion.
[0080] Figure 9 This is a block diagram illustrating a label determining device according to an exemplary embodiment. (Refer to...) Figure 9 The device includes: The candidate recipe tag acquisition module 910 is used to input the recipe to be tagged into the first preset tag analysis model, perform multi-dimensional tag analysis on the recipe to be tagged, and obtain at least one candidate recipe tag corresponding to each of the multiple preset dimensions. The first analysis data acquisition module 920 is used to input each candidate recipe label under each preset dimension into the second preset label analysis model corresponding to each preset dimension, analyze the accuracy of each candidate recipe label, and output the first analysis data corresponding to each candidate recipe label. The target label acquisition module 930 is used to determine the target label from at least one candidate recipe label corresponding to each of multiple preset dimensions based on the first analysis data and the first preset indicator data corresponding to each of the multiple second preset label analysis models.
[0081] In an optional embodiment, the first preset indicator data is updated according to a first preset period, and the first preset indicator data is determined by the following module: The first tag quantity acquisition module is used to acquire, upon reaching the first update time of the first preset indicator data, the historical indicator data, the number of first tags, the number of second tags, and the number of third tags corresponding to the multiple second preset tag analysis models within the first preset period before the first update time. The first historical labeling accuracy acquisition module is used to determine the first historical labeling accuracy of each of the multiple second preset label analysis models within a first preset period before the first update time, based on the number of first labels and the number of second labels. The first target coefficient acquisition module is used to determine the first target coefficient corresponding to each of the multiple second preset label analysis models based on the first historical labeling accuracy rate and the first preset adjustment coefficient. The first preset indicator data acquisition module is used to determine the first preset indicator data based on the first target coefficient, the number of third labels, and historical indicator data.
[0082] In an optional embodiment, the target tag acquisition module 930 includes: The model weight determination unit is used to determine the model weights corresponding to each of the multiple second preset label analysis models based on the first preset index data. The target label determination unit is used to determine the target label from at least one candidate recipe label corresponding to each of multiple preset dimensions based on the first analysis data and model weights.
[0083] In an optional embodiment, when each candidate recipe label corresponds to one of multiple preset dimensions, the target label determination unit includes: The second analysis data determination subunit is used to determine the second analysis data corresponding to each candidate recipe tag based on the first analysis data and model weights; The first target label determination subunit is used to determine the target label from at least one candidate recipe label corresponding to multiple preset dimensions based on the second analysis data and preset comparison thresholds.
[0084] In an optional embodiment, when the target candidate label corresponds to at least two of the multiple preset dimensions, the target candidate label is any one of the at least one candidate recipe label corresponding to each of the multiple preset dimensions, and the target label determination unit further includes: The third analysis data determination subunit is used to determine the third analysis data corresponding to the target candidate label based on the second analysis data and the first preset indicator data corresponding to the target candidate label. The first sub-unit for determining the target label mentioned above includes: The second target label determination subunit is used to determine the target label from at least one candidate recipe label corresponding to multiple preset dimensions based on the third analysis data corresponding to the target candidate label, the second analysis data corresponding to other candidate labels, and a preset comparison threshold.
[0085] In an optional embodiment, the preset comparison threshold is updated according to a first preset period, and the preset comparison threshold is determined by the following module: The second tag quantity acquisition module is used to acquire, at the first update time when the preset comparison threshold is reached, the historical comparison threshold, historical target coefficient, and the number of third and fourth tags corresponding to multiple second preset tag analysis models within the first preset period before the first update time; The second historical labeling accuracy acquisition module is used to determine the second historical labeling accuracy of multiple second preset label analysis models within a first preset period before the first update time, based on the number of third and fourth labels. The second target coefficient acquisition module is used to determine the second target coefficient based on the second historical marking accuracy rate, the historical target coefficient, and the preset accuracy rate. The preset comparison threshold acquisition module is used to determine the preset comparison threshold based on the historical comparison threshold, the second target coefficient, the second historical marking accuracy rate, and the preset accuracy rate.
[0086] In an optional embodiment, the above-described apparatus further includes: The target keyword acquisition module is used to extract keywords from the recipes to be tagged, and obtain the target keywords for the recipes to be tagged. The extended feature information acquisition module is used to extend the recipe features of the recipe to be tagged based on the target keywords, so as to obtain the extended feature information of the recipe to be tagged. The aforementioned candidate recipe tag acquisition module 910 includes: The candidate recipe tag acquisition unit is used to input the target keywords and extended feature information into the first preset tag analysis model. Based on the multi-dimensional tag analysis of the recipe to be tagged, the tag is expanded to obtain at least one candidate recipe tag corresponding to each of the multiple preset dimensions of the recipe to be tagged.
[0087] In an optional embodiment, the first preset label analysis model is updated according to a second preset period. After the labels are determined after the second preset period, the device further includes: The third tag quantity acquisition module is used to acquire, when the second update time of the first preset tag analysis model is reached, the second preset indicator data corresponding to each of the multiple second preset tag analysis models, and the fifth tag quantity and sixth tag quantity corresponding to the first preset tag analysis model within the second preset period before the second update time; The third historical labeling accuracy acquisition module is used to determine the third historical labeling accuracy of the first preset label analysis model within the second preset period before the second update time, based on the number of fifth and sixth labels. The third target coefficient acquisition module is used to determine the third target coefficient based on the second preset indicator data, the third historical labeling accuracy rate, the second preset adjustment coefficient, the third preset adjustment coefficient, and the preset accuracy rate. The target label analysis model determination module is used to determine the target label analysis model from multiple second preset label analysis models based on the third target coefficient and the second preset index data, and update the target label analysis model to the first preset label analysis model.
[0088] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.
[0089] Figure 10 This is a block diagram illustrating an electronic device for tag identification according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows: Figure 10As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a tag identification method. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.
[0090] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements. In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the tag determination method as described in the embodiments of this disclosure.
[0091] In an exemplary embodiment, a computer-readable storage medium is also provided, wherein when the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the tag determination method of the present disclosure embodiments.
[0092] In an exemplary embodiment, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the tag determination method in the embodiments of this disclosure.
[0093] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.
[0094] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.
[0095] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.
Claims
1. A label identification method, characterized in that, The method includes: Input the recipe to be tagged into the first preset tag analysis model, perform multi-dimensional tag analysis on the recipe to be tagged, and obtain at least one candidate recipe tag corresponding to each of the multiple preset dimensions of the recipe to be tagged. Each candidate recipe tag under each preset dimension is input into the second preset tag analysis model corresponding to each preset dimension. The accuracy of each candidate recipe tag is analyzed, and the first analysis data corresponding to each candidate recipe tag is output. The first analysis data indicates the probability that each candidate recipe tag belongs to the tag of the recipe to be tagged under the corresponding preset dimension. Based on the first analysis data and the first preset index data corresponding to each of the multiple second preset label analysis models, the target label is determined from the at least one candidate recipe label corresponding to each of the multiple preset dimensions. The first preset index data characterizes the analysis performance of the corresponding second preset label analysis model on the recipe label.
2. The method according to claim 1, characterized in that, The first preset indicator data is updated according to a first preset period, and the first preset indicator data is determined in the following way: When the first update time of the first preset indicator data is reached, the historical indicator data, the number of first tags, the number of second tags, and the number of third tags corresponding to each of the plurality of second preset tag analysis models within the first preset period before the first update time are obtained. The number of first tags is the number of tags of the first historical candidate tags analyzed by each of the plurality of second preset tag analysis models. The number of second tags is the number of tags of the first historical candidate tags belonging to the corresponding first historical recipes. The number of third tags is the number of tags of the second historical candidate tags belonging to the corresponding second historical recipes. Based on the first number of tags and the second number of tags, determine the first historical labeling accuracy rate of each of the plurality of second preset label analysis models within the first preset period before the first update time; Based on the first historical labeling accuracy rate and the first preset adjustment coefficient, the first target coefficient corresponding to each of the plurality of second preset label analysis models is determined. The first preset adjustment coefficient is used to constrain the value range of the first target coefficient, and the first target coefficient is used to constrain the growth rate of the first preset indicator data. Based on the first target coefficient, the number of third labels, and the historical indicator data, the first preset indicator data is determined.
3. The method according to claim 1, characterized in that, The step of determining the target label from the at least one candidate recipe label corresponding to each of the multiple preset dimensions based on the first analysis data and the first preset indicator data corresponding to each of the multiple second preset label analysis models includes: Based on the first preset index data, determine the model weights corresponding to each of the plurality of second preset label analysis models. Based on the first analysis data and the model weights, the target label is determined from at least one candidate recipe label corresponding to each of the multiple preset dimensions.
4. The method according to claim 3, characterized in that, When each candidate recipe tag corresponds to one of the plurality of preset dimensions, determining the target tag from the at least one candidate recipe tag corresponding to each of the plurality of preset dimensions based on the first analysis data and the model weights includes: Based on the first analysis data and the model weights, the second analysis data corresponding to each candidate recipe tag is determined; Based on the second analysis data and the preset comparison threshold, the target label is determined from at least one candidate recipe label corresponding to each of the multiple preset dimensions.
5. The method according to claim 4, characterized in that, When the target candidate label corresponds to at least two of the plurality of preset dimensions, the target candidate label is any one of the at least one candidate recipe label corresponding to each of the plurality of preset dimensions. The step of determining the target label from the at least one candidate recipe label corresponding to each of the plurality of preset dimensions based on the first analysis data and the model weights further includes: Based on the second analysis data corresponding to the target candidate label and the first preset indicator data, the third analysis data corresponding to the target candidate label is determined; The step of determining the target label from at least one candidate recipe label corresponding to each of the plurality of preset dimensions based on the second analysis data and preset comparison thresholds includes: Based on the third analysis data corresponding to the target candidate label, the second analysis data corresponding to other candidate labels, and the preset comparison threshold, the target label is determined from the at least one candidate recipe label corresponding to each of the multiple preset dimensions. The other candidate labels are candidate recipe labels other than the target candidate label among the at least one candidate recipe labels corresponding to each of the multiple preset dimensions.
6. The method according to claim 5, characterized in that, The preset comparison threshold is updated according to a first preset period, and the preset comparison threshold is determined in the following manner: In the case of reaching the first update time of the preset comparison threshold, the historical comparison threshold, historical target coefficient, number of third tags and number of fourth tags corresponding to the plurality of second preset tag analysis models are obtained within the first preset period before the first update time. The number of third tags is the number of recipe tags belonging to the corresponding second historical recipe among the second historical candidate tags analyzed by the plurality of second preset tag analysis models, and the number of fourth tags is the number of tags of the second historical candidate tags. Based on the number of third tags and the number of fourth tags, determine the second historical labeling accuracy of the multiple second preset tag analysis models within the first preset period before the first update time; Based on the second historical marking accuracy rate, the historical target coefficient, and the preset accuracy rate, a second target coefficient is determined, which is used to constrain the update range of the preset comparison threshold. The preset comparison threshold is determined based on the historical comparison threshold, the second target coefficient, the second historical marking accuracy, and the preset accuracy.
7. The method according to claim 1, characterized in that, The method further includes: Keyword extraction is performed on the recipes to be tagged to obtain the target keywords for the recipes to be tagged; Based on the target keywords, the recipe to be tagged is extended to obtain the extended feature information of the recipe to be tagged. The step of inputting the recipe to be tagged into a first preset tag analysis model, performing multi-dimensional tag analysis on the recipe to be tagged, and obtaining at least one candidate recipe tag corresponding to the recipe to be tagged in each of the multiple preset dimensions includes: The target keywords and the extended feature information are input into the first preset tag analysis model. Based on the multi-dimensional tag analysis of the recipe to be tagged, the tags are extended to obtain at least one candidate recipe tag corresponding to the recipe to be tagged in each of the multiple preset dimensions.
8. The method according to claim 1, characterized in that, The first preset label analysis model is updated according to a second preset period. After the labels are determined after the second preset period, the method further includes: When the second update time of the first preset tag analysis model is reached, the second preset index data corresponding to each of the plurality of second preset tag analysis models, the number of fifth tags and the number of sixth tags corresponding to the first preset tag analysis model in the second preset period before the second update time are obtained. The number of fifth tags is the number of tags of the third historical candidate tags output by the first preset tag analysis model, and the number of sixth tags is the number of tags of the recipe tags belonging to the corresponding third historical recipes among the third historical candidate tags. Based on the number of the fifth label and the number of the sixth label, determine the third historical labeling accuracy of the first preset label analysis model in the second preset period before the second update time; Based on the second preset indicator data, the third historical labeling accuracy rate, the second preset adjustment coefficient, the third preset adjustment coefficient, and the preset accuracy rate, a third target coefficient is determined. The second preset adjustment coefficient is a constant benchmark for calculating the third target coefficient. The third preset adjustment coefficient is used to constrain the degree of influence of the update of the second preset indicator data within a preset time on the third target coefficient. The third target coefficient is used to control the update frequency of the first preset label analysis model. Based on the third target coefficient and the second preset index data, a target label analysis model is determined from the plurality of second preset label analysis models, and the target label analysis model is updated to the first preset label analysis model.
9. A label determining device, characterized in that, include: The candidate recipe tag acquisition module is used to input the recipe to be tagged into the first preset tag analysis model, perform multi-dimensional tag analysis on the recipe to be tagged, and obtain at least one candidate recipe tag corresponding to the recipe to be tagged in each of the multiple preset dimensions. The first analysis data acquisition module is used to input each candidate recipe label under each preset dimension into the second preset label analysis model corresponding to each preset dimension, analyze the accuracy of each candidate recipe label, and output the first analysis data corresponding to each candidate recipe label. The first analysis data indicates the probability that each candidate recipe label belongs to the label of the recipe to be labeled under the corresponding preset dimension. The target tag acquisition module is used to determine the target tag from the at least one candidate recipe tag corresponding to each of the multiple preset dimensions based on the first analysis data and the first preset index data corresponding to each of the multiple second preset tag analysis models. The first preset index data characterizes the analysis performance of the corresponding second preset tag analysis model on the recipe tag.
10. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the tag determination method as described in any one of claims 1 to 8.