A method, device, equipment and storage medium for predicting commodity categories

Abstract

The present application discloses a method, apparatus, device, and storage medium for predicting product categories, including: inputting preset product information into a pre-trained model for processing to obtain a product category corresponding to the preset product information; wherein the pre-trained model is an existing trained model for predicting product categories; performing distribution statistics on the product category corresponding to the preset product information to obtain a corresponding category distribution result, and performing distribution alignment processing on the preset product information according to the category distribution result; using the preset product information after distribution alignment to fine-tune the pre-trained model to obtain a target prediction model, so as to use the target prediction model to predict the product category of the product information to be predicted. The present application migrates the pre-trained model trained with a large-scale corpus to the target prediction model, avoiding a large amount of manual annotation processes and improving the efficiency and accuracy of product category prediction.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a method, device, equipment and storage medium for predicting product categories. Background Art

[0002] Big data is everywhere, especially in the rapidly developing e-commerce platforms, and data maintenance is particularly important. Among them, a large number of product categories are the top priorities of governance. Since each e-commerce platform has its own unique product categories, how to adaptively organize all products and sort out multi-level categories according to different platforms is the only way to improve the competitiveness and work efficiency of the platform. Most of the existing product category governance solutions combine "direct manual annotation" and "fine-tuning the model". However, the annotation of product categories is relatively complex, the cost of manual annotation is high and the efficiency is low, and the accuracy cannot be guaranteed at the same time.

[0003] Therefore, how to provide an efficient and accurate method for predicting product categories is a technical problem that those skilled in the art need to solve urgently. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and storage medium for predicting product categories, which can avoid a large number of manual annotation processes and improve the efficiency and accuracy of product category prediction. The specific solutions are as follows:

[0005] The first aspect of the present application provides a method for predicting product categories, including:

[0006] Inputting preset product information into a pre-trained model for processing to obtain a product category corresponding to the preset product information; wherein, the pre-trained model is an existing trained model for predicting product categories;

[0007] Performing distribution statistics on the product category corresponding to the preset product information to obtain a corresponding category distribution result, and performing distribution alignment processing on the preset product information according to the category distribution result;

[0008] Using the preset product information after distribution alignment to fine-tune the pre-trained model to obtain a target prediction model, so as to predict the product category of the product information to be predicted by using the target prediction model.

[0009] Optionally, before inputting the preset product information into the pre-trained model for processing, it further includes:

[0010] Obtaining third-party product information and corresponding product categories;

[0011] Clean the third-party product information and corresponding product categories by means of active learning, so as to fine-tune the pre-trained model using the cleaned data.

[0012] Optionally, the cleaning of the third-party product information and corresponding product categories by means of active learning includes:

[0013] Train the classifiers of the corresponding types of product categories respectively using the third-party product information in each product category;

[0014] Use the trained classifiers to predict the product categories of the corresponding third-party product information, and delete the third-party product information with a confidence level less than the first preset threshold.

[0015] Optionally, before training the classifiers of the corresponding types of product categories respectively using the third-party product information in each product category, it further includes:

[0016] Divide the third-party product information and corresponding product categories into a training set, a test set and a validation set, and screen the third-party product information from the training set to train the classifiers of the corresponding types of product categories.

[0017] Optionally, the product category prediction method further includes:

[0018] Verify the model effect of the trained classifiers of each product category by means of five-fold cross-validation.

[0019] Optionally, the cleaning of the third-party product information and corresponding product categories by means of active learning includes:

[0020] Use the query function to divide the third-party product information according to the types of product categories to obtain positive samples and negative samples corresponding to each product category respectively; wherein, the positive samples contain third-party product information with the same product category and corresponding type, and the negative samples contain third-party product information with other types of product categories;

[0021] Train the classifiers of the corresponding types of product categories respectively using the positive samples and the negative samples, and move the third-party product information with a confidence level greater than the second preset threshold in the negative samples during the training process into the positive samples to continue training until the classifier converges.

[0022] Optionally, the statistical analysis of the distribution of the product categories corresponding to the preset product information includes:

[0023] Determine the sampled product information from the preset product information, and verify the product category of the sampled product information, so as to perform distribution statistics on the product category corresponding to the verified sampled product information.

[0024] The second aspect of the present application provides a product category prediction device, including:

[0025] A preprocessing module, configured to input the preset product information into a pre-trained model for processing to obtain a product category corresponding to the preset product information; wherein, the pre-trained model is an existing trained model for predicting product categories;

[0026] A distribution alignment module, configured to perform distribution statistics on the product category corresponding to the preset product information to obtain a corresponding category distribution result, and perform distribution alignment processing on the preset product information according to the category distribution result;

[0027] A first fine-tuning module, configured to fine-tune the pre-trained model by using the preset product information after distribution alignment to obtain a target prediction model, so as to predict the product category of the product information to be predicted by using the target prediction model.

[0028] The third aspect of the present application provides an electronic device, the electronic device includes a processor and a memory; wherein the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the foregoing product category prediction method.

[0029] The fourth aspect of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are loaded and executed by a processor, the foregoing product category prediction method is implemented.

[0030] In the present application, first input the preset product information into a pre-trained model for processing to obtain a product category corresponding to the preset product information; wherein, the pre-trained model is an existing trained model for predicting product categories; then perform distribution statistics on the product category corresponding to the preset product information to obtain a corresponding category distribution result, and perform distribution alignment processing on the preset product information according to the category distribution result; finally, fine-tune the pre-trained model by using the preset product information after distribution alignment to obtain a target prediction model, so as to predict the product category of the product information to be predicted by using the target prediction model. It can be seen that the present application migrates the pre-trained model trained with a large amount of corpus to the target prediction model, that is, migrates the product category information to the target prediction model, avoiding a large amount of manual annotation processes, and improving the efficiency and accuracy of product category prediction. Description of the Drawings

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on the provided drawings.

[0032] Figure 1 It is a flowchart of a commodity category prediction method provided by this application;

[0033] Figure 2 It is a schematic diagram of a specific commodity category prediction method provided by this application;

[0034] Figure 3 It is a specific data cleaning process diagram provided by this application;

[0035] Figure 4 It is a schematic diagram of the structure of a commodity category prediction device provided by this application;

[0036] Figure 5 It is a structural diagram of an electronic device for commodity category prediction provided by this application. Specific embodiments

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0038] Most of the existing commodity category governance solutions combine "direct manual annotation" and "fine-tuning the model". However, the category annotation of commodities is relatively complex, the cost of manual annotation is high and the efficiency is low, and the accuracy rate cannot be guaranteed at the same time. In view of the above technical defects, this application provides a commodity category prediction solution, which can avoid a large number of manual annotation processes and improve the efficiency and accuracy of commodity category prediction.

[0039] Figure 1 It is a flowchart of a commodity category prediction method provided by an embodiment of this application. Refer to Figure 1 As shown, the commodity category prediction method includes:

[0040] S11: Input the preset commodity information into the pre-trained model for processing to obtain the commodity category corresponding to the preset commodity information; wherein, the pre-trained model is an existing trained model for predicting commodity categories.

[0041] In this embodiment, the preset product information is input into a pre-trained model for processing to obtain a product category corresponding to the preset product information. The pre-trained model is an existing trained model for predicting product categories. Specifically, the pre-trained model (PTM) refers to a model trained using a large amount of text that has appeared in people's lives, enabling the model to learn the probability distribution of each word or character in this text, and thus building a model that conforms to this text distribution. The label of the corpus of the language model is its context, which determines that the language model can be trained using a large-scale corpus almost without limit. These large-scale corpora enable the pre-trained model to obtain powerful capabilities. Through the pre-trained model, that is, the language model trained using a large-scale corpus, downstream category governance tasks can exhibit better effects. In addition, based on the pre-trained model, various classification and regression NLP tasks can be connected downstream.

[0042] Furthermore, to ensure the transfer effect, in addition to using the existing open-source pre-trained model trained on a large-scale public corpus, it is also necessary to fine-tune the pre-trained model on third-party publicly available product category data. Fine-tuning refers to the process of applying a pre-trained model to one's own dataset and adapting the parameters to one's own dataset. Specifically, in this embodiment, third-party product information and corresponding product categories are first obtained, and then the third-party product information and corresponding product categories are cleaned through active learning to fine-tune the pre-trained model using the cleaned data. Since there are common NLP problems such as inconsistent annotations and ambiguous product names in the public corpus. In the above process, after obtaining the third-party data, the annotation corpus is first cleaned using the idea of active learning.

[0043] There are two data cleaning schemes in this embodiment. In one embodiment, the third-party product information and corresponding product categories are first divided into a training set, a test set, and a validation set, that is, the dataset is split into train, test, and valid, where valid remains fixed, and train and test are selected according to the scheme. And third-party product information is screened from the training set to train the classifiers for corresponding types of product categories. Then, the third-party product information in each product category is used to train the classifiers for corresponding types of product categories respectively; finally, the trained classifiers are used to predict the product categories of the corresponding third-party product information, and the third-party product information with a confidence level less than the first preset threshold is deleted. Through active learning, the unconfident samples in the model training are removed, and the product categories after the model is labeled are continuously iterated, thereby improving the accuracy of product category prediction.

[0044] In addition, to ensure the cleaning effect, the model effect of the trained classifier for each product category can also be verified through five-fold cross-validation. Five-fold cross-validation means splitting the labeled data into five parts, with each part serving as the validation set in turn, and the remaining four parts serving as the training set. The model is trained with the training set and the prediction effect of the model is observed with the validation set. Observing the effects of the five validation sets can often more fully illustrate the true effect of the model than a single result.

[0045] In another embodiment, the query function is first used to divide the third-party product information according to the product category types to obtain the positive samples and negative samples corresponding to each product category. The query function (QueryFunction) is the query function in active learning. The process of updating the machine learning model is as follows: the model is updated by incremental learning or re-learning, so as to integrate the manually labeled data into the machine learning model and improve the model effect. Among them, the positive samples contain third-party product information with the same product category and corresponding types, and the negative samples contain third-party product information with other product categories. Then, the positive samples and the negative samples are respectively used to train the classifier for the corresponding product category, and the third-party product information with a confidence level greater than the second preset threshold in the negative samples during the training process is moved into the positive samples for continued training until the classifier converges. That is, first randomly select the positive samples P (Positive), and then select a small amount of data U (Unlabelled) from the remaining data as the negative samples. Then train the classifier to predict the unselected U, add the ones with higher confidence levels to P, and put the remaining data back into U. Repeat the above process multiple times until convergence. For details, refer to Figure 3 as shown.

[0046] S12: Perform distribution statistics on the product category corresponding to the preset product information to obtain the corresponding category distribution result, and perform distribution alignment processing on the preset product information according to the category distribution result.

[0047] In this embodiment, the distribution statistics of the product categories corresponding to the preset product information are performed to obtain the corresponding category distribution results, and the preset product information is processed for distribution alignment according to the category distribution results. To improve the processing efficiency, only a small part of the product information and categories need to be statistically distributed. Therefore, before that, a small part of all product category data is randomly sampled, and this part of the data is manually verified, and only the category distribution after manual verification is statistically counted. According to this distribution, the distribution alignment of all product data is performed. Specifically, the sampled product information is determined from the preset product information, and the product categories of the sampled product information are verified, so as to perform the distribution statistics on the product categories corresponding to the sampled product information after verification. It can be understood that sampling means extracting a part of the data. Often, because the global data volume is too large to exceed the computing power of humans or machines, a part of the data needs to be extracted for manual marking or model training. There are many sampling methods, which are mainly flexibly selected according to the purpose of sampling. At the same time, in the actual scenario, the amount of sampled data is often determined according to the number of categories and the amount of manual work, and this embodiment does not limit this.

[0048] It should be noted that distribution alignment means that the category distribution after reduction of each category of the data to be aligned is approximated to the expected distribution. However, in actual operation, due to the long-tail nature of the product category distribution, categories with too little data are not conducive to the model learning and training relevant information. Therefore, for the tail categories, no reduction is required.

[0049] S13: The pre-trained model is fine-tuned using the preset product information after distribution alignment to obtain a target prediction model, so as to predict the product category of the product information to be predicted using the target prediction model.

[0050] In this embodiment, the pre-trained model is fine-tuned using the preset product information after distribution alignment to obtain a target prediction model, so as to predict the product category of the product information to be predicted using the target prediction model. This process is also to re-fine-tune the pre-trained language model with the aligned data, and repeat the above steps until convergence. So far, the transfer learning is completed, reducing the labor cost to the lowest, without the need for a large amount of manual direct annotation of the local data, and the model effect can be well transferred to the local data. It can be understood that transfer learning is to transfer the trained model parameters to a new model to help the new model training. Considering that most data or tasks are correlated, transfer learning can share the learned model parameters (which can also be understood as the knowledge learned by the model) to the new model in a certain way, thus accelerating and optimizing the learning efficiency of the model without starting from scratch like most networks.

[0051] It can be seen that in the embodiment of the present application, the preset commodity information is first input into a pre-trained model for processing to obtain a commodity category corresponding to the preset commodity information; wherein, the pre-trained model is an existing trained model for predicting commodity categories; then, distribution statistics are performed on the commodity category corresponding to the preset commodity information to obtain a corresponding category distribution result, and distribution alignment processing is performed on the preset commodity information according to the category distribution result; finally, the pre-trained model is fine-tuned using the preset commodity information after distribution alignment to obtain a target prediction model, so as to use the target prediction model to predict the commodity category of the commodity information to be predicted. The embodiment of the present application migrates the pre-trained model trained with a large-scale corpus to the target prediction model, that is, the commodity category information is migrated into the target prediction model, avoiding a large number of manual annotation processes and improving the efficiency and accuracy of commodity category prediction.

[0052] See Figure 4 As shown, the embodiment of the present application also correspondingly discloses a commodity category prediction device, including:

[0053] A preprocessing module 11, configured to input preset commodity information into a pre-trained model for processing to obtain a commodity category corresponding to the preset commodity information; wherein, the pre-trained model is an existing trained model for predicting commodity categories;

[0054] A distribution alignment module 12, configured to perform distribution statistics on the commodity category corresponding to the preset commodity information to obtain a corresponding category distribution result, and perform distribution alignment processing on the preset commodity information according to the category distribution result;

[0055] A first fine-tuning module 13, configured to fine-tune the pre-trained model using the preset commodity information after distribution alignment to obtain a target prediction model, so as to use the target prediction model to predict the commodity category of the commodity information to be predicted.

[0056] It can be seen that in the embodiments of the present application, the preset product information is first input into a pre-trained model for processing to obtain the product category corresponding to the preset product information; wherein, the pre-trained model is an existing trained model for predicting product categories; then, the distribution statistics of the product category corresponding to the preset product information are performed to obtain the corresponding category distribution result, and the preset product information is subjected to distribution alignment processing according to the category distribution result; finally, the pre-trained model is fine-tuned using the preset product information after distribution alignment to obtain a target prediction model, so as to predict the product category of the product information to be predicted using the target prediction model. The embodiments of the present application transfer the pre-trained model trained with a large-scale corpus to the target prediction model, that is, the product category information is transferred to the target prediction model, avoiding a large number of manual annotation processes and improving the efficiency and accuracy of product category prediction.

[0057] In some specific embodiments, the product category prediction device further includes:

[0058] An acquisition module, configured to acquire third-party product information and corresponding product categories;

[0059] A cleaning module, configured to perform data cleaning on the third-party product information and corresponding product categories by means of active learning;

[0060] A second fine-tuning module, configured to fine-tune the pre-trained model using the cleaned data.

[0061] In some specific embodiments, the cleaning module specifically includes a first cleaning sub-module and a second cleaning sub-module, wherein the first cleaning sub-module includes:

[0062] A first data partitioning unit, configured to partition the third-party product information and corresponding product categories into a training set, a test set, and a validation set, and screen the third-party product information from the training set to train the classifier for the corresponding type of product category;

[0063] A first training unit, configured to train the classifier for the corresponding type of product category using the third-party product information in each product category respectively;

[0064] A prediction unit, configured to predict the product category of the corresponding third-party product information using the trained classifier;

[0065] A deletion unit, configured to delete the third-party product information with a confidence level less than a first preset threshold;

[0066] A validation unit, configured to verify the model effect of the trained classifier for each product category by means of five-fold cross-validation;

[0067] The second cleaning sub-module includes:

[0068] A second data partitioning unit, configured to use a query function to partition the third-party commodity information according to the types of commodity categories, and respectively obtain positive samples and negative samples corresponding to each commodity category; wherein, the positive samples contain third-party commodity information with the same commodity category and belonging to the corresponding type, and the negative samples contain third-party commodity information with other types of commodity categories;

[0069] A second training unit, configured to respectively use the positive samples and the negative samples to train the classifiers for the corresponding types of commodity categories, and move the third-party commodity information with a confidence level greater than a second preset threshold in the negative samples during the training process into the positive samples for continued training until the classifier converges.

[0070] In some specific embodiments, the distribution alignment module is further configured to determine sampled commodity information from the preset commodity information, and verify the commodity categories of the sampled commodity information, so as to perform distribution statistics on the commodity categories corresponding to the verified sampled commodity information.

[0071] Furthermore, an embodiment of the present application also provides an electronic device. Figure 5 It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment, and the content in the figure cannot be considered as any limitation to the scope of use of the present application.

[0072] Figure 5 It is a schematic structural diagram of an electronic device 20 provided by an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the commodity category prediction method disclosed in any of the foregoing embodiments.

[0073] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows is any communication protocol applicable to the technical solution of the present application, and no specific limitation is imposed on it here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application requirements, and no specific limitation is made here.

[0074] In addition, as a carrier for storing resources, the memory 22 can be a read-only memory, a random access memory, a magnetic disk, an optical disc, etc. The resources stored thereon can include an operating system 221, a computer program 222, data 223, etc. The storage method can be temporary storage or permanent storage.

[0075] Among them, the operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20, so as to enable the processor 21 to perform operations and processing on the massive data 223 in the memory 22. It can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the commodity category prediction method executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks. The data 223 can include sample data collected by the electronic device 20.

[0076] Furthermore, an embodiment of the present application also discloses a storage medium in which a computer program is stored. When the computer program is loaded and executed by a processor, the steps of the commodity category prediction method disclosed in any of the foregoing embodiments are implemented.

[0077] In this specification, the various embodiments are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0078] Finally, it should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0079] The above has introduced in detail the method, device, equipment and storage medium for predicting product categories provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for predicting product categories, characterized in that, Including: Inputting preset product information into a pre-trained model for processing to obtain a product category corresponding to the preset product information; wherein, the pre-trained model is an existing trained model for predicting product categories; Performing distribution statistics on the product category corresponding to the preset product information to obtain a corresponding category distribution result, and performing distribution alignment processing on the preset product information according to the category distribution result; Using the preset product information after distribution alignment to fine-tune the pre-trained model to obtain a target prediction model, so as to use the target prediction model to predict the product category of the product information to be predicted; Before inputting the preset product information into the pre-trained model for processing, it further includes: Obtaining third-party product information and corresponding product categories; Performing data cleaning on the third-party product information and corresponding product categories in an active learning manner, so as to use the cleaned data to fine-tune the pre-trained model; The performing data cleaning on the third-party product information and corresponding product categories in an active learning manner includes: Respectively using the third-party product information in each product category to train the classifier for the corresponding product category; Using the trained classifier to predict the product category of the corresponding third-party product information, and deleting the third-party product information with a confidence level less than a first preset threshold.

2. The commodity category prediction method according to claim 1, characterized in that Before respectively using the third-party product information in each product category to train the classifier for the corresponding product category, it further includes: Dividing the third-party product information and corresponding product categories into a training set, a test set and a validation set, and screening the third-party product information from the training set to train the classifier for the corresponding product category.

3. The commodity category prediction method according to claim 1, wherein It further includes: Verifying the model effect of the trained classifier for each product category by means of five-fold cross-validation.

4. The method for predicting a product category according to claim 1, characterized in that The performing data cleaning on the third-party product information and corresponding product categories in an active learning manner includes: Using a query function to divide the third-party product information according to the product category types to respectively obtain a positive sample and a negative sample corresponding to each product category; wherein, the positive sample contains third-party product information with the same product category and belonging to the corresponding type, and the negative sample contains third-party product information with other product categories; Respectively using the positive sample and the negative sample to train the classifier for the corresponding product category, and moving the third-party product information with a confidence level greater than a second preset threshold in the negative sample during the training process into the positive sample for continued training until the classifier converges.

5. The commodity category prediction method according to any one of claims 1 to 4, characterized in that, The performing distribution statistics on the product category corresponding to the preset product information includes: Determining sampled product information from the preset product information, and verifying the product category of the sampled product information, so as to perform distribution statistics on the product category corresponding to the verified sampled product information.

6. A commodity category prediction device, characterized in that, Including: A preprocessing module, configured to input preset product information into a pre-trained model for processing to obtain a product category corresponding to the preset product information; wherein, the pre-trained model is an existing trained model for predicting product categories; A distribution alignment module, configured to perform distribution statistics on the product categories corresponding to the preset product information to obtain corresponding category distribution results, and perform distribution alignment processing on the preset product information according to the category distribution results; A first fine-tuning module, configured to fine-tune the pre-trained model by using the preset product information after distribution alignment to obtain a target prediction model, so as to predict the product category of the product information to be predicted by using the target prediction model; Wherein, before inputting the preset product information into the pre-trained model for processing, it further includes: Obtaining third-party product information and corresponding product categories; Performing data cleaning on the third-party product information and corresponding product categories by means of active learning, so as to fine-tune the pre-trained model by using the cleaned data; The performing data cleaning on the third-party product information and corresponding product categories by means of active learning includes: Respectively training the classifiers of the corresponding types of product categories by using the third-party product information in each product category; Predicting the product categories of the corresponding third-party product information by using the trained classifiers, and deleting the third-party product information with a confidence level less than a first preset threshold.

7. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the product category prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, For storing computer-executable instructions, when the computer-executable instructions are loaded and executed by a processor, the product category prediction method according to any one of claims 1 to 5 is implemented.

Images