CN114358821A

CN114358821A - Commodity detail feature extraction method and device, computer equipment and storage medium

Info

Publication number: CN114358821A
Application number: CN202111613543.7A
Authority: CN
Inventors: 马俊
Original assignee: Chuangyou Digital Technology Guangdong Co Ltd
Current assignee: Chuangyou Digital Technology Guangdong Co Ltd
Priority date: 2021-12-27
Filing date: 2021-12-27
Publication date: 2022-04-15
Anticipated expiration: 2041-12-27
Also published as: CN114358821B

Abstract

The application relates to a method and a device for extracting detailed features of commodities, computer equipment and a storage medium. The method comprises the following steps: acquiring commodity data to be processed corresponding to the target commodity details, wherein the commodity data to be processed carries the target commodity category; processing the commodity data to be processed to obtain corresponding commodity title data; inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category, and performing feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data; and extracting target commodity detail characteristic words from the commodity title data according to the target characteristic label marks, wherein the target commodity detail characteristic words are used for describing the commodity characteristics of the target commodity details. By adopting the method, the core characteristics of the target thin goods can be captured, and the user can quickly and intuitively know the target thin goods through the captured core characteristics.

Description

Commodity detail feature extraction method and device, computer equipment and storage medium

Technical Field

The present application relates to the field of computer technologies, and in particular, to a method and an apparatus for extracting detailed features of a commodity, a computer device, and a storage medium.

Background

In the field of commodity sales, commodities can be classified into large, medium, small and fine categories, the granularity of the categories is increased layer by layer, and the larger the granularity is, so that various users such as commodity category operators, buyers, consumer analysts and the like cannot really know the commodities.

The accurate positioning of commodity classification needs to be judged based on commodity keywords, and each type of positioning needs to rely on the extraction of commodity keywords (namely the extraction of commodity features), the feature extraction of commodities is generally performed through an LSTM (Long Short-Term Memory) at present, and the extracted commodity features are displayed for users to see, but due to the structural property of the LSTM, only the above information can be obtained, and the semantic dependency information cannot be obtained for a longer text, so that the granularity of the commodity features extracted in the mode is large, and the finer feature extraction of the commodities cannot be achieved.

Disclosure of Invention

In view of the above, it is desirable to provide a method, an apparatus, a computer device and a storage medium for extracting features of commodity details, which can capture core features of target details commodities and help a user quickly and intuitively recognize target details through the captured core features.

A method for extracting commodity fine feature comprises the following steps:

acquiring commodity data to be processed corresponding to the target commodity details, wherein the commodity data to be processed carries the target commodity category;

processing the commodity data to be processed to obtain corresponding commodity title data;

inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category, and performing feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data;

and extracting target commodity detail characteristic words from the commodity title data according to the target characteristic label marks, wherein the target commodity detail characteristic words are used for describing the commodity characteristics of the target commodity details.

In one embodiment, the method further comprises the following steps: acquiring a candidate commodity data set corresponding to the candidate commodity details, wherein the candidate commodity data set comprises at least one candidate commodity data, and the candidate commodity data carries candidate commodity categories; processing the candidate commodity data to obtain corresponding candidate commodity title data; inputting candidate commodity title data corresponding to the same candidate commodity category into the same initial commodity fine feature extraction model, and training the initial commodity fine feature extraction model to obtain a trained commodity fine feature extraction model; and establishing a mapping relation between the candidate commodity category and the matched trained commodity fine-class feature extraction model.

In one embodiment, inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category comprises the following steps: and inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category according to the mapping relation.

In one embodiment, inputting the candidate commodity title data corresponding to the same candidate commodity category into the same initial commodity fine feature extraction model, and training the initial commodity fine feature extraction model to obtain a trained commodity fine feature extraction model, includes: acquiring candidate commodity title data corresponding to the same candidate commodity category, wherein the candidate commodity title data carries an actual characteristic label mark; inputting the candidate commodity title data into the same initial commodity detail feature extraction model, and performing feature extraction on the candidate commodity title data through the initial commodity detail feature extraction model to obtain an output feature tag mark; calculating according to the actual characteristic label mark and the output characteristic label mark to obtain a training loss value; and adjusting the model parameters of the initial commodity fine feature extraction model according to the training loss value until the convergence condition is met, and obtaining the trained commodity fine feature extraction model.

In one embodiment, before obtaining the trained commodity fine-class feature extraction model, the method further includes: calculating according to the actual characteristic label mark and the output characteristic label mark to obtain an output accuracy rate and a model recall rate; and calculating to obtain a model evaluation value according to the output accuracy rate and the model recall rate, and executing the step of obtaining the trained commodity detail feature extraction model when the model evaluation value reaches a preset model evaluation threshold value.

In one embodiment, processing the data of the commodity to be processed to obtain corresponding data of a commodity title includes: acquiring name information and corresponding descriptive information of the commodity to be processed from the commodity data to be processed; dividing commodity name information and descriptive information to be processed to obtain a plurality of words; and extracting commodity title data corresponding to the commodity data to be processed from the plurality of words.

In one embodiment, the target feature tag flag includes a plurality of feature elements, and the extracting of the target item detail feature words from the item title data according to the target feature tag flag includes: determining invalid characteristic elements in the target characteristic label mark according to a preset rule; eliminating invalid characteristic elements in the target characteristic label mark to obtain an intermediate characteristic label mark; and extracting the target commodity detail feature words from the commodity title data according to the intermediate feature label marks.

In one embodiment, the target product detail type feature words comprise a plurality of detail type feature keywords, and the method further comprises: acquiring commodity sales data corresponding to various detailed characteristic keywords from to-be-processed commodity data; and screening the multiple fine characteristic keywords according to the commodity sales data to obtain target fine characteristic keywords, determining the target fine characteristic keywords as characteristic descriptive information of the commodities under the target commodity fine category, and displaying.

An apparatus for extracting features of commodity details, the apparatus comprising:

the acquisition module is used for acquiring to-be-processed commodity data corresponding to the target commodity category, and the to-be-processed commodity data carries the target commodity category;

the processing module is used for processing the commodity data to be processed to obtain corresponding commodity title data;

the input module is used for inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category, and performing feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data;

and the generating module is used for extracting target commodity detail characteristic words from the commodity title data according to the target characteristic label marks, wherein the target commodity detail characteristic words are used for describing the commodity characteristics of the target commodity details.

A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program:

A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:

According to the method, the device, the computer equipment and the storage medium for extracting the commodity detail feature, the target commodity title data under the target commodity detail is subjected to feature extraction through the target commodity detail feature extraction model matched with the target commodity category to obtain the target feature label mark, and the target commodity detail feature words for describing the commodity feature under the target commodity detail are extracted from the input commodity title data through the target feature label mark. The extracted feature words of the target commodity details can represent the core features of the target commodity details, and for users such as product operators, buyers, consumer analysts and the like, the users can quickly form relatively intuitive knowledge of a certain commodity details through the extracted feature words of the commodity details even if the users do not know the specific certain commodity details. And the commodity title data under the specific commodity details under different commodity categories are respectively and independently used for feature extraction by using different commodity detail feature extraction models, so that the problem of inaccurate model prediction caused by conflict among different commodity categories or conflict among some commodity title data of the commodity details under different commodity categories can be solved.

Drawings

FIG. 1 is a diagram illustrating an exemplary embodiment of an application of a method for extracting features of commodity subclasses;

FIG. 2 is a schematic flow chart of a method for extracting features of commodity details in one embodiment;

FIG. 3 is a schematic flow chart illustrating the steps of data processing of the commodity to be processed according to an embodiment;

FIG. 4 is a flowchart illustrating a step of extracting feature words from a detail category of a target commodity in one embodiment;

FIG. 5 is a flowchart illustrating a method for extracting detailed features of a commodity according to an embodiment;

FIG. 6 is a flowchart illustrating the training steps of the commodity fine class feature extraction model in one embodiment;

FIG. 7 is a flowchart illustrating the verification step of the item detail feature extraction model in one embodiment;

FIG. 8 is a flowchart illustrating a method for extracting detailed features of a commodity according to an embodiment;

FIG. 9 is a block diagram showing the structure of a device for extracting details of a commodity according to an embodiment;

FIG. 10 is a diagram showing an internal structure of a computer device in one embodiment;

FIG. 11 is a diagram illustrating an internal structure of a computer device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The method for extracting the commodity detail features can be applied to the application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices, and the server 104 may be implemented by an independent server or a server cluster formed by a plurality of servers.

Specifically, the terminal 102 acquires to-be-processed commodity data corresponding to a target commodity category, and sends the to-be-processed commodity data to the server 104, the server 104 acquires to-be-processed commodity data corresponding to the target commodity category, the to-be-processed commodity data carries the target commodity category, the to-be-processed commodity data is processed to obtain corresponding commodity title data, the commodity title data is input into a target commodity category feature extraction model matched with the target commodity category, feature extraction is performed on the commodity title data through the target commodity category feature extraction model to obtain a target feature tag mark corresponding to the commodity title data, and a target commodity detail feature word is extracted from the commodity title data according to the target feature tag mark, wherein the target commodity detail feature word is used for describing the commodity features of the target commodity category.

In another embodiment, the terminal 102 obtains to-be-processed commodity data corresponding to a target commodity category, the to-be-processed commodity data carries a target commodity category, the to-be-processed commodity data is processed to obtain corresponding commodity title data, the commodity title data is input into a target commodity category-matched target commodity category feature extraction model, feature extraction is performed on the commodity title data through the target commodity category feature extraction model to obtain a target feature tag mark corresponding to the commodity title data, and a target commodity category feature word is extracted from the commodity title data according to the target feature tag mark and is used for describing commodity features of the target commodity category.

In one embodiment, as shown in fig. 2, a method for extracting a detail feature of a commodity is provided, which is described by taking the method as an example of being applied to a terminal or a server in fig. 1, and includes the following steps:

step 202, to-be-processed commodity data corresponding to the target commodity subclass is obtained, and the to-be-processed commodity data carries the target commodity class.

The target commodity subclass is a commodity subclass to be predicted, and the commodity subclass belongs to one class of commodity classification, and commodities are generally classified into a major class, a middle class, a minor class, and a subclass. For example, the major group is skin care, the middle group is skin care, the minor group is basic care, and the minor group is facial mask, lotion, milky lotion, facial cleanser, etc.

The thin category belongs to the smallest level in the commodity classification, and the commodity thin category is a detailed distinction of commodity varieties, including the specifications, the flower colors, the grades and the like of the commodities, and more specifically embodies the characteristics of the commodities, for example, the commodity thin category can be 60-degree goblet five-grain liquor. That is, the thin-category features of the goods may enable the user to quickly develop an intuitive sense of all goods under the target goods thin-category.

Specifically, a target commodity subclass may be determined according to an actual business requirement, an actual application scenario, or an actual Product requirement, and all commodity data under the target commodity subclass is obtained as to-be-processed commodity data, where the commodity data includes commodity name information and descriptive information corresponding to the commodity, where the commodity name information may be composed of an SPU (Standard Product Unit) and a SKU (Stock Keeping Unit), the SPU information is a minimum Unit of commodity information aggregation, and is a set of reusable and easily-retrievable standardized information, and the set describes characteristics of a Product, and in colloquial terms, a commodity with the same attribute value and characteristics may be referred to as an SPU. The SKU is a physically inseparable stock keeping unit, which is handled according to different states of business and different management modes when in use, and is most commonly used in clothes and footwear. For example, "iPhone X" may determine that a product is an SPU. "iPhone X64G silver" is a SKU.

The data of the commodities to be processed carry the categories of the target commodities, wherein the categories of the target commodities are large categories in the categories of the commodities, that is, the categories of the commodities to which the target commodities related to the data of the commodities to be processed belong can be determined according to the categories of the target commodities.

And step 204, processing the commodity data to be processed to obtain corresponding commodity title data.

The commodity title data is descriptive information formed by titles of the commodity data to be processed, the commodity title data can be commodity name information, and the format of the commodity title data can be as follows: the SPU title + SKU title may be extracted from the to-be-processed commodity data, and specifically, the SPU title, and the SKU title may be extracted from the to-be-processed commodity data, and the commodity title data may be composed of the extracted SPU title, and SKU title.

In one embodiment, as shown in FIG. 3, step 204 comprises:

step 302, the name information of the commodity to be processed and the corresponding descriptive information are obtained from the data of the commodity to be processed.

And 304, segmenting the commodity name information and the descriptive information to be processed to obtain a plurality of words.

And step 306, extracting the commodity title data corresponding to the commodity data to be processed from the plurality of words.

The commodity data to be processed includes commodity name information to be processed and corresponding commodity descriptive information, where the commodity descriptive information is information describing relevant commodities, such as information describing colors, specifications, and the like of the commodities. Specifically, to-be-processed commodity name information and corresponding descriptive information in to-be-processed commodity data are obtained, the to-be-processed commodity name information and the descriptive information are segmented through a segmentation technology, the segmentation technology can be a JIEBA segmentation technology, specifically, the to-be-processed commodity name information and the descriptive information are segmented to obtain a plurality of words, an SPU title, an SPU subheader and an SKU title can be further extracted from the plurality of words, and finally, the commodity title data are formed through the extracted SPU title, the SPU subheader and the SKU title.

For example, the name information of the commodity to be processed and the corresponding descriptive information are "the skin care and concealing capabilities of BB cosmetics pregnant face for isolating and nursing baby in the united states of Belli pregnant woman special sunscreen cream" and are strong ", and the name information and the descriptive information of the commodity to be processed are divided to obtain a plurality of words: "USA, Beli, pregnant woman, Special, sunscreen, isolation, lactation, available, BB, cosmetics, pregnancy, face, skin care, concealer, power, strength". The SPU titles extracted from the various terms are: "the facial skin care concealer of BB cosmetics available for isolating lactation with sunscreen cream special for pregnant women in the United states", SPU secondary title is: "sunscreen isolation foundation concealer 2 in 1", SKU title: color classification, namely skin color, and finally, the obtained commodity title data are as follows: "American Belli pregnant woman special sunscreen cream isolation lactation could use BB cosmetics pregnant face skin care concealer sunscreen isolation foundation concealer 2 in 1 color classification: skin color".

And step 206, inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category, and performing feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data.

In the present embodiment, the target product detail feature extraction model is a model for extracting a product detail feature in the input product title data, and has a correspondence relationship with the target product category, and the product detail feature can be understood as a feature of a product under the detail classification category.

The target commodity category is a large category in the commodity classification, each large category comprises a plurality of commodity details, the data volume of the commodity detail data set of each commodity detail is large, the commodity detail data between each large category possibly conflict, if the commodity detail data are placed in the same model for training, the problems of conflict of model feature extraction, inaccurate model prediction and the like occur, and therefore in order to avoid the problems, the commodity detail feature extraction model is not suitable for training one commodity detail feature extraction model by each commodity detail, the commodity detail data of all the commodity details under each large category are used for training the commodity detail feature extraction model corresponding to the same large category, and the trained commodity detail feature extraction model is universal for the commodity detail data of all the commodity details under the large category. That is, the trained commodity detail feature extraction model can take each matched commodity detail data under the large category as input, and the input is predicted through the commodity detail feature extraction model. The item detail data may be item title data.

Further, inputting the product title data into the target product detail feature extraction model, and performing feature extraction on the input product title data through the target product detail feature extraction model to obtain a target feature tag corresponding to the product title data, where the target feature tag is a stack of words containing tag tags and does not have specific description information, for example, the target feature tag may be a stack of words containing BIO tags and does not contain specific description information, such as: the target feature tag is labeled "BIOOOOOOOO OBIBIBIBIOOBIOOBIIOBIIIBIBI".

In an implementation manner of this embodiment, a structure of the target commodity subclass feature extraction model may use a structure of Bert + Crf, where Bert is an acronym of "Bidirectional Encoder retrieval from transforms," and is a Bidirectional coding characterization model using a transform as a main frame, and the Bert model structure may obtain context semantic information and support text input of maximum 512 characters, and a network layer of a Crf conditional random field is superimposed on the structure, so that the target commodity subclass feature extraction model can mine multiple features having an interaction relationship from input commodity title data, construct a dependency relationship among the multiple features, can learn a feature tag more accurately, and finally output a target feature tag matching the input commodity title data.

And 208, extracting target commodity detail feature words from the commodity title data according to the target feature label marks, wherein the target commodity detail feature words are used for describing commodity features of the target commodity details.

The target feature tag output by the target commodity detail feature extraction model is only a stack of words containing the tag, does not have specific description information, and cannot directly know the commodity detail feature corresponding to the target commodity detail from the target feature tag, so that the output target feature tag needs to be converted to obtain the target commodity detail feature words containing the specific description information.

Specifically, since the length of the target feature tag label is the same as the text length of the product title data, that is, the target feature tag label output by the target product detail feature extraction model corresponds to the input product title data one by one, the target product detail feature word can be extracted from the product title data according to the target feature tag label, specifically, the specific feature word corresponding to the effective element in the target feature tag label in the product title data is obtained, the extracted specific feature word forms the target product detail feature word, and the target product detail feature word is used for describing the product feature of the target product detail and contains specific description information, so that the product feature corresponding to the target product detail can be intuitively known through the target product detail feature word.

In one embodiment, as shown in FIG. 4, the target feature tag label includes a plurality of feature elements, and step 208 includes:

and 402, determining invalid characteristic elements in the target characteristic label mark according to a preset rule.

And step 404, eliminating invalid feature elements in the target feature tag mark to obtain an intermediate feature tag mark.

And 406, extracting the target commodity detail feature words from the commodity title data according to the intermediate feature label marks.

The target characteristic label mark comprises a plurality of characteristic elements, invalid characteristic elements in the target characteristic label mark can be determined according to a preset rule, the invalid characteristic elements are removed, effective characteristic elements are left to form an intermediate characteristic label mark, matched keywords are extracted from commodity title data according to the effective characteristic elements in the intermediate characteristic label mark, and target commodity detail characteristic words are formed. The preset rule can be set in advance according to actual business requirements, actual product requirements or actual application scenes.

For example, the target feature tag flags are: BIOOOBIOOBI comprises characteristic elements of three types, wherein the characteristic elements are B, I, O respectively, invalid characteristic elements in the target characteristic label mark are determined to be O according to preset rules, all O in the target characteristic label mark are removed, and the obtained intermediate characteristic label mark is: BI, the title data of the goods are: and finally, extracting the target commodity fine characteristic words from the commodity title data according to the intermediate characteristic label mark, wherein the color of the scarf of the cotton-flax tea card salt lake is purple: cotton, linen, scarf, purple.

In the method for extracting the fine feature of the commodity, the feature extraction model of the fine feature of the target commodity matched with the category of the target commodity is used for extracting the feature of the commodity title data under the fine category of the target commodity to obtain a target feature tag mark, and the target feature tag mark is used for extracting the fine feature words of the target commodity for describing the feature of the commodity under the fine category of the target commodity from the input commodity title data. The extracted feature words of the target commodity details can represent the core features of the target commodity details, and for users such as product operators, buyers, consumer analysts and the like, the users can quickly form relatively intuitive knowledge of a certain commodity details through the extracted feature words of the commodity details even if the users do not know the specific certain commodity details. And the commodity title data under the specific commodity details under different commodity categories are respectively and independently used for feature extraction by using different commodity detail feature extraction models, so that the problem of inaccurate model prediction caused by conflict among different commodity categories or conflict among some commodity title data of the commodity details under different commodity categories can be solved.

In one embodiment, as shown in fig. 5, the method for extracting the item detail feature further includes:

step 502, a candidate commodity data set corresponding to the candidate commodity category is obtained, wherein the candidate commodity data set comprises at least one candidate commodity data, and the candidate commodity data carries the candidate commodity category.

Step 504, the candidate commodity data is processed to obtain corresponding candidate commodity title data.

Step 506, inputting the candidate commodity title data corresponding to the same candidate commodity category into the same initial commodity fine feature extraction model, and training the initial commodity fine feature extraction model to obtain a trained commodity fine feature extraction model.

Specifically, all candidate commodity data under the candidate commodity category are obtained to form a candidate commodity data set, each candidate commodity data carries a candidate commodity category, the candidate commodity data are processed, candidate commodity name information and commodity descriptive information in the candidate commodity data can be obtained, the candidate commodity name information and the commodity descriptive information are segmented through a segmentation technology to obtain a plurality of words, and then the commodity title data are extracted from the plurality of words.

The commodity category is a major category in the commodity category, each major category comprises a plurality of commodity details, the data volume of the commodity detail data set of each commodity detail is large, the commodity detail data between each major category possibly conflict, and if the commodity detail data are placed in the same model for training, the problems of conflict, inaccurate model prediction and the like in model feature extraction occur.

That is, the commodity title data corresponding to the specific commodity details in different major categories are respectively trained by using a single commodity detail feature extraction model, and the commodity title data corresponding to the specific commodity details in the same major category are trained by using the same commodity detail feature extraction model. Therefore, candidate commodity title data corresponding to the same candidate commodity category can be input into the same initial commodity fine feature extraction model, the initial fine feature extraction model is trained, and the specific training process is explained in detail in the following embodiments, so that the trained commodity fine feature extraction model is obtained.

And step 508, establishing a mapping relation between the candidate commodity category and the matched trained commodity subclass feature extraction model.

In this embodiment, inputting the commodity title data into the target commodity detail feature extraction model matched with the target commodity category includes: and inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category according to the mapping relation.

After the trained commodity fine feature extraction model is obtained, a mapping relation between the candidate commodity category and the matched trained commodity fine feature extraction model is established, and after the commodity title data corresponding to the commodity data to be processed is obtained, the commodity title data can be input into the target commodity fine feature extraction model matched with the target commodity category according to the established mapping relation.

For example, the candidate commodity data set may be divided into 13 candidate commodity categories, each candidate commodity category includes a plurality of candidate commodity details, and 13 candidate commodity categories — total 313 candidate commodity details, in the technical solution of the present application, a separate commodity detail feature extraction model needs to be trained for all candidate commodity data under each candidate commodity category, so that finally, 13 commodity detail feature extraction models corresponding to the 13 candidate commodity categories are trained in total, and are used for predicting commodity detail feature words in the commodity data under each candidate commodity category.

In one embodiment, as shown in FIG. 6, step 506 includes:

step 602, obtaining candidate commodity title data corresponding to the same candidate commodity category, where the candidate commodity title data carries an actual feature tag mark.

And step 604, inputting the candidate commodity title data into the same initial commodity detail feature extraction model, and performing feature extraction on the candidate commodity title data through the initial commodity detail feature extraction model to obtain an output feature tag mark.

And 606, calculating to obtain a training loss value according to the actual characteristic label mark and the output characteristic label mark.

And 608, adjusting model parameters of the initial commodity fine feature extraction model according to the training loss value until a convergence condition is met, and obtaining the trained commodity fine feature extraction model.

Specifically, all candidate commodity title data under the same candidate commodity category are obtained, the candidate commodity title data can be obtained by processing the candidate commodity data under the same candidate commodity category, each candidate commodity title data carries a corresponding actual feature tag mark, and the actual feature tag mark is a correct feature tag mark of the candidate commodity title data, is also a stack of words containing the feature tag, and does not include any specific determined description information.

Further, all candidate commodity title data under the same candidate commodity major category are input into the same initial commodity detail feature extraction model, and the initial commodity detail feature extraction model calculates the candidate commodity title data to obtain an output feature tag mark. The output feature tag label is also a word containing a feature tag, and does not include specific description information.

Calculating to obtain a model training loss value of one iterative training through an output feature label mark output by an initial commodity fine feature extraction model and an actual feature label mark carried by input candidate commodity title data, obtaining a trained commodity fine feature extraction model when the model training loss value reaches a preset convergence condition by verifying whether the model training loss value reaches the preset convergence condition, and continuously adjusting model parameters of the initial commodity fine feature extraction model according to the training loss value until the preset convergence condition is met when the model training loss value does not reach the preset convergence condition to obtain the trained commodity fine feature extraction model. The preset convergence condition may be determined according to an actual service requirement, an actual product requirement, or an actual application scenario, for example, when the training loss value is no longer changed, it may be determined that the preset convergence condition is reached.

In one embodiment, as shown in fig. 7, before obtaining the trained feature extraction model for the commodity subclass, the method further includes:

and step 702, calculating according to the actual characteristic label mark and the output characteristic label mark to obtain the output accuracy rate and the model recall rate.

And step 704, calculating to obtain a model evaluation value according to the output accuracy rate and the model recall rate.

And 706, when the model evaluation value reaches a preset model evaluation threshold value, executing the step of obtaining the trained commodity fine feature extraction model.

Wherein, the model output accuracy rate can be also called as: the model precision ratio is an index for reflecting the output accuracy of the commodity fine feature extraction model, the higher the model output accuracy ratio is, the higher the output accuracy of the commodity fine feature extraction model is, the higher the model recall ratio is, the index for reflecting whether the commodity fine feature extraction model is predicted to be complete or not, and the higher the model recall ratio is, the more complete the commodity fine feature extraction model is predicted to be.

The output accuracy rate and the model recall rate can be obtained by calculating an actual characteristic tag mark and an output characteristic tag mark, specifically, the number of output entities is determined by characteristic elements in the output characteristic tag mark, the number of actual entities is determined by characteristic elements in the actual characteristic tag mark, the number of predicted correct entities is determined according to the number of the output entities and the number of the actual entities, the output accuracy rate is obtained by calculating according to the number of the output entities and the number of the predicted correct entities, and the model recall rate is obtained by calculating according to the number of the actual entities and the number of the predicted correct entities.

Further, after the output accuracy rate and the model recall rate are obtained, a model evaluation value can be obtained through calculation of the output accuracy rate and the model recall rate, the model evaluation value is an index used for evaluating a model effect of the commodity fine feature extraction model in training, the larger the model evaluation value is, the better the model effect of the commodity fine feature extraction model is, and conversely, the smaller the model evaluation value is, the worse the model effect of the commodity fine feature extraction model is, and whether the commodity fine feature extraction model achieves the best model effect can be determined through the model evaluation value.

The model evaluation value is calculated according to the output accuracy rate and the model recall rate, and can be specifically calculated according to the following formula:

F＝1+ln(precision)*ln(recall)/(ln(precision*recall))

wherein F represents a model evaluation value, Precision represents a model output accuracy rate, and Recall represents a model Recall rate.

For example, the actual feature tag of the product title data a is bibio, and the output feature tag output by the product detail feature extraction model is BIBIBI, wherein the number of actual entities determined by the feature elements in the actual feature tag is 2, which are BI BIs respectively, while the number of output entities determined by the feature elements in the output feature tag is 3, which are BI respectively, and only 1 entity is calculated from the actual entity number and the output entity number, so that the output accuracy rate is calculated from the output entity number and the predicted correct entity number: 1/3, the model recall rate is calculated according to the actual entity number and the predicted correct entity number: 1/2. Further, the model evaluation value F may be calculated from the above formula as 1+ ln (1/3) × ln (1/2)/(ln (1/3 × 1/2)).

And finally, determining whether the trained commodity fine feature extraction model achieves the optimal model effect by judging whether the model evaluation value reaches a preset model evaluation threshold value, wherein the preset model evaluation threshold value can be obtained in advance according to actual business requirements, actual product requirements or actual application scenes, specifically, obtaining the preset model evaluation threshold value, judging whether the model evaluation value reaches the preset model evaluation threshold value, if the model evaluation value reaches the preset model evaluation threshold value, indicating that the trained commodity fine feature extraction model achieves the optimal model effect, and outputting the trained commodity fine feature extraction model. On the contrary, if the model evaluation value does not reach the preset model evaluation threshold value, the trained commodity fine feature extraction model meets the convergence condition but does not reach the optimal model effect, and the convergence condition can be modified to re-adjust the model parameters and re-train the model until the commodity fine feature extraction model which can meet the convergence condition and reach the optimal model effect is obtained.

Therefore, the final step of training the commodity fine feature extraction model and the evaluation of the model effect can ensure the prediction accuracy of the finally trained commodity fine feature extraction model, namely the prediction accuracy of the commodity fine feature words predicted by the commodity fine feature extraction model.

In one embodiment, as shown in fig. 8, the target product detail feature words include a plurality of detail feature keywords, and the method for extracting the product detail feature further includes:

and step 802, acquiring commodity sales data corresponding to each detail characteristic keyword from the to-be-processed commodity data.

And 804, screening the multiple fine characteristic keywords according to the commodity sales data to obtain target fine characteristic keywords, determining the target fine characteristic keywords as characteristic descriptive information of the commodities under the target commodity fine category, and displaying the characteristic descriptive information.

After the target commodity detail characteristic words are obtained, the plurality of detail characteristic keywords in the target commodity detail characteristic words can be arranged in a descending order according to the sales volume corresponding to the target commodity detail characteristic words, a preset number of detail characteristic keywords are respectively taken out to be displayed, and the display can be drawn into icon display.

Specifically, commodity sales data corresponding to each fine characteristic keyword is obtained from commodity data to be processed, the commodity sales data of each fine characteristic keyword are sorted in a descending order, the fine characteristic keywords 20 before sale are extracted and determined as target fine characteristic keywords, and finally the target fine characteristic keywords are determined as characteristic descriptive information of commodities under the target commodity fine category and displayed.

In a specific embodiment, a method for extracting detailed features of a commodity is provided, which specifically comprises the following steps:

1. and acquiring a candidate commodity data set corresponding to the candidate commodity details, wherein the candidate commodity data set comprises at least one candidate commodity data, and the candidate commodity data carries the candidate commodity category.

2. And processing the candidate commodity data to obtain corresponding candidate commodity title data.

3. And inputting the candidate commodity title data corresponding to the same candidate commodity category into the same initial commodity fine feature extraction model, and training the initial commodity fine feature extraction model to obtain a trained commodity fine feature extraction model.

And 3-1, obtaining candidate commodity title data corresponding to the same candidate commodity category, wherein the candidate commodity title data carries an actual characteristic label mark.

And 3-2, inputting the candidate commodity title data into the same initial commodity detail feature extraction model, and performing feature extraction on the candidate commodity title data through the initial commodity detail feature extraction model to obtain an output feature tag mark.

And 3-3, calculating to obtain a training loss value according to the actual characteristic label mark and the output characteristic label mark.

And 3-4, adjusting model parameters of the initial commodity fine feature extraction model according to the training loss value until a convergence condition is met, and obtaining the trained commodity fine feature extraction model.

Before obtaining the trained commodity fine feature extraction model, the method further comprises the following steps: and calculating according to the actual characteristic label mark and the output characteristic label mark to obtain an output quasi-accuracy rate and a model recall rate, calculating according to the output accuracy rate and the model recall rate to obtain a training loss value model evaluation value, and executing the step of obtaining a trained commodity detail characteristic extraction model when the model evaluation value reaches a preset model evaluation threshold value.

4. And establishing a mapping relation between the candidate commodity category and the matched trained commodity fine-class feature extraction model.

5. And acquiring to-be-processed commodity data corresponding to the target commodity details, wherein the to-be-processed commodity data carries the target commodity category.

6. And processing the commodity data to be processed to obtain corresponding commodity title data.

And 6-1, acquiring name information of the commodity to be processed and corresponding descriptive information from the data of the commodity to be processed.

And 6-2, segmenting the name information and the descriptive information of the commodity to be processed to obtain a plurality of words.

And 6-3, extracting commodity title data corresponding to the to-be-processed commodity data from the plurality of words.

7. And inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category according to the mapping relation, and performing feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data.

8. And extracting target commodity detail characteristic words from the commodity title data according to the target characteristic label marks, wherein the target commodity detail characteristic words are used for describing the commodity characteristics of the target commodity details.

8-1, the target feature tag mark comprises a plurality of feature elements, and invalid feature elements in the target feature tag mark are determined according to a preset rule.

And 8-2, removing the invalid characteristic elements in the target characteristic label mark to obtain an intermediate characteristic label mark.

And 8-3, extracting the target commodity detail feature words from the commodity title data according to the intermediate feature label marks.

9. And acquiring commodity sales data corresponding to each detail characteristic keyword from the to-be-processed commodity data.

10. And screening the multiple fine characteristic keywords according to the commodity sales data to obtain target fine characteristic keywords, determining the target fine characteristic keywords as characteristic descriptive information of the commodities under the target commodity fine category, and displaying.

It should be understood that, although the steps in the above-described flowcharts are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in the above-described flowcharts may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of performing the sub-steps or the stages is not necessarily sequential, but may be performed alternately or alternatingly with other steps or at least a portion of the sub-steps or stages of other steps.

In one embodiment, as shown in fig. 9, there is provided an apparatus 900 for extracting characteristics of commodity details, including: an acquisition module 902, a processing module 904, an input module 906, and a generation module 908, wherein:

the obtaining module 902 is configured to obtain to-be-processed commodity data corresponding to the target commodity category, where the to-be-processed commodity data carries the target commodity category.

The processing module 904 is configured to process the commodity data to be processed to obtain corresponding commodity title data.

The input module 906 is configured to input the commodity title data into a target commodity detail feature extraction model matched with the target commodity category, and perform feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data.

The generating module 908 is configured to extract target product detail feature words from the product title data according to the target feature tag, where the target product detail feature words are used to describe product features of the target product detail.

In one embodiment, the article detail feature extraction apparatus 900 obtains a candidate article data set corresponding to a candidate article detail, where the candidate article data set includes at least one candidate article data, and the candidate article data carries a candidate article category, processes the candidate article data to obtain corresponding candidate article title data, inputs the candidate article title data corresponding to the same candidate article category into the same initial article detail feature extraction model, trains the initial article detail feature extraction model to obtain a trained article detail feature extraction model, establishes a mapping relationship between the candidate article category and the trained article detail feature extraction model, and the input module 906 inputs the article title data into a target article detail feature extraction model matched with the target article category according to the mapping relationship.

In one embodiment, the commodity detail feature extraction device 900 obtains candidate commodity title data corresponding to the same candidate commodity category, where the candidate commodity title data carries an actual feature tag flag, inputs the candidate commodity title data into the same initial commodity detail feature extraction model, performs feature extraction on the candidate commodity title data through the initial commodity detail feature extraction model to obtain an output feature tag flag, calculates a training loss value according to the actual feature tag flag and the output feature tag flag, and adjusts a model parameter of the initial commodity detail feature extraction model according to the training loss value until a convergence condition is satisfied to obtain a trained commodity detail feature extraction model.

In one embodiment, the commodity detail feature extraction apparatus 900 calculates an output quasi-accuracy and a model recall ratio according to the actual feature tag labels and the output feature tag labels, calculates a training loss value model evaluation value according to the output accuracy and the model recall ratio, and performs the step of obtaining the trained commodity detail feature extraction model when the model evaluation value reaches a preset model evaluation threshold value.

In one embodiment, the processing module 904 obtains the name information of the commodity to be processed and the corresponding descriptive information from the commodity data to be processed, divides the name information of the commodity to be processed and the descriptive information to obtain a plurality of words, and extracts the commodity title data corresponding to the commodity data to be processed from the plurality of words.

In an embodiment, the target feature tag mark includes a plurality of feature elements, the generating module 908 determines an invalid feature element in the target feature tag mark according to a preset rule, removes the invalid feature element in the target feature tag mark to obtain an intermediate feature tag mark, and extracts a target commodity detail feature word from the commodity title data according to the intermediate feature tag mark.

In one embodiment, the commodity detail feature extraction device 900 obtains commodity sales data corresponding to each detail feature keyword from the commodity data to be processed, obtains a target detail feature keyword by screening from a plurality of detail feature keywords according to the commodity sales data, determines the target detail feature keyword as feature descriptive information of a commodity under the target commodity detail, and displays the feature descriptive information.

For specific limitations of the commodity detail feature extraction device, reference may be made to the above limitations of the commodity detail feature extraction method, and details are not described here. All or part of the modules in the commodity detail feature extraction device can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in fig. 10. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer equipment is used for storing the target commodity detail feature extraction model. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to realize a commodity detail feature extraction method.

In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 11. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to realize a commodity detail feature extraction method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.

Those skilled in the art will appreciate that the configurations shown in fig. 10 or 11 are merely block diagrams of some configurations relevant to the present disclosure, and do not constitute a limitation on the computing devices to which the present disclosure may be applied, and that a particular computing device may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program: the method comprises the steps of obtaining commodity data to be processed corresponding to target commodity details, wherein the commodity data to be processed carries a target commodity category, processing the commodity data to be processed to obtain corresponding commodity title data, inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category, carrying out feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data, extracting target commodity detail feature words from the commodity title data according to the target feature tag mark, wherein the target commodity detail feature words are used for describing commodity features of the target commodity details.

In one embodiment, the processor, when executing the computer program, further performs the steps of: the method comprises the steps of obtaining a candidate commodity data set corresponding to a candidate commodity category, wherein the candidate commodity data set comprises at least one candidate commodity data, the candidate commodity data carries a candidate commodity category, processing the candidate commodity data to obtain corresponding candidate commodity title data, inputting the candidate commodity title data corresponding to the same candidate commodity category into the same initial commodity category feature extraction model, training the initial commodity category feature extraction model to obtain a trained commodity category feature extraction model, and establishing a mapping relation between the candidate commodity category and the matched trained commodity category feature extraction model.

In one embodiment, the processor, when executing the computer program, further performs the steps of: and inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category according to the mapping relation.

In one embodiment, the processor, when executing the computer program, further performs the steps of: obtaining candidate commodity title data corresponding to the same candidate commodity category, wherein the candidate commodity title data carries an actual feature tag mark, inputting the candidate commodity title data into the same initial commodity detail feature extraction model, performing feature extraction on the candidate commodity title data through the initial commodity detail feature extraction model to obtain an output feature tag mark, calculating according to the actual feature tag mark and the output feature tag mark to obtain a training loss value, adjusting model parameters of the initial commodity detail feature extraction model according to the training loss value until a convergence condition is met, and obtaining a trained commodity detail feature extraction model.

In one embodiment, the processor, when executing the computer program, further performs the steps of: and calculating according to the actual characteristic label mark and the output characteristic label mark to obtain an output quasi-accuracy rate and a model recall rate, calculating according to the output accuracy rate and the model recall rate to obtain a training loss value model evaluation value, and executing the step of obtaining a trained commodity detail characteristic extraction model when the model evaluation value reaches a preset model evaluation threshold value.

In one embodiment, the processor, when executing the computer program, further performs the steps of: the method comprises the steps of obtaining name information of commodities to be processed and corresponding descriptive information from the commodity data to be processed, segmenting the name information of the commodities to be processed and the descriptive information to obtain a plurality of words, and extracting commodity title data corresponding to the commodity data to be processed from the words.

In one embodiment, the target feature tag flag comprises a plurality of feature elements, and the processor when executing the computer program further performs the steps of: and determining invalid feature elements in the target feature tag mark according to a preset rule, removing the invalid feature elements in the target feature tag mark to obtain an intermediate feature tag mark, and extracting the target commodity detail feature words from the commodity title data according to the intermediate feature tag mark.

In one embodiment, the target product detail feature words comprise a plurality of detail feature keywords, and the processor executes the computer program to further implement the following steps: and acquiring commodity sales data corresponding to each fine characteristic keyword from the commodity data to be processed, screening a plurality of fine characteristic keywords according to the commodity sales data to obtain target fine characteristic keywords, determining the target fine characteristic keywords as characteristic descriptive information of the commodities under the target commodity fine category, and displaying the characteristic descriptive information.

In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of: the method comprises the steps of obtaining commodity data to be processed corresponding to target commodity details, wherein the commodity data to be processed carries a target commodity category, processing the commodity data to be processed to obtain corresponding commodity title data, inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category, carrying out feature extraction on the commodity title data through the target commodity detail feature extraction model to obtain a target feature tag mark corresponding to the commodity title data, extracting target commodity detail feature words from the commodity title data according to the target feature tag mark, wherein the target commodity detail feature words are used for describing commodity features of the target commodity details.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. 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 a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. A method for extracting commodity fine class features, comprising the following steps:

acquiring commodity data to be processed corresponding to the target commodity category, wherein the commodity data to be processed carries the target commodity category;

and extracting target commodity detail feature words from the commodity title data according to the target feature label marks, wherein the target commodity detail feature words are used for describing commodity features of the target commodity details.

2. The method of claim 1, further comprising:

acquiring a candidate commodity data set corresponding to a candidate commodity category, wherein the candidate commodity data set comprises at least one candidate commodity data, and the candidate commodity data carries the candidate commodity category;

processing the candidate commodity data to obtain corresponding candidate commodity title data;

inputting candidate commodity title data corresponding to the same candidate commodity category into the same initial commodity fine feature extraction model, and training the initial commodity fine feature extraction model to obtain a trained commodity fine feature extraction model;

establishing a mapping relation between the candidate commodity category and the matched trained commodity fine-category feature extraction model;

the inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category comprises the following steps:

and inputting the commodity title data into a target commodity detail feature extraction model matched with the target commodity category according to the mapping relation.

3. The method as claimed in claim 2, wherein the step of inputting the candidate commodity title data corresponding to the same candidate commodity category into the same initial commodity fine-category feature extraction model and training the initial commodity fine-category feature extraction model to obtain the trained commodity fine-category feature extraction model comprises:

obtaining candidate commodity title data corresponding to the same candidate commodity category, wherein the candidate commodity title data carries an actual characteristic label mark;

inputting the candidate commodity title data into the same initial commodity detail feature extraction model, and performing feature extraction on the candidate commodity title data through the initial commodity detail feature extraction model to obtain an output feature tag mark;

calculating to obtain a training loss value according to the actual characteristic label mark and the output characteristic label mark;

and adjusting the model parameters of the initial commodity fine feature extraction model according to the training loss value until a convergence condition is met, and obtaining a trained commodity fine feature extraction model.

4. The method of claim 3, wherein the obtaining the trained commodity fine class feature extraction model comprises:

calculating according to the actual characteristic label mark and the output characteristic label mark to obtain an output accuracy rate and a model recall rate;

calculating to obtain a model evaluation value according to the output accuracy rate and the model recall rate;

and when the model evaluation value reaches a preset model evaluation threshold value, executing the step of obtaining the trained commodity fine feature extraction model.

5. The method according to any one of claims 1 to 4, wherein the processing the data of the commodity to be processed to obtain corresponding data of a title of the commodity comprises:

acquiring name information and corresponding descriptive information of the commodity to be processed from the commodity data to be processed;

segmenting the name information of the commodity to be processed and the descriptive information to obtain a plurality of words;

and extracting commodity title data corresponding to the commodity data to be processed from the words.

6. The method according to any one of claims 1 to 4, wherein the target feature tag label includes a plurality of feature elements, and the extracting a target item detail feature word from the item title data according to the target feature tag label, the target item detail feature word being used for describing an item feature of the target item detail, includes:

determining invalid feature elements in the target feature tag mark according to a preset rule;

removing the invalid characteristic elements in the target characteristic label mark to obtain an intermediate characteristic label mark;

and extracting target commodity detail feature words from the commodity title data according to the intermediate feature label marks.

7. The method according to any one of claims 1 to 4, wherein the target commodity detail feature words comprise a plurality of detail feature keywords, and the method further comprises:

acquiring commodity sales data corresponding to each fine characteristic keyword from the to-be-processed commodity data;

and screening a target fine characteristic keyword from the plurality of fine characteristic keywords according to the commodity sales data, determining the target fine characteristic keyword as characteristic descriptive information of the commodity under the target commodity fine category, and displaying the characteristic descriptive information.

8. An apparatus for extracting features of commodity details, the apparatus comprising:

the system comprises an acquisition module, a classification module and a classification module, wherein the acquisition module is used for acquiring to-be-processed commodity data corresponding to a target commodity category, and the to-be-processed commodity data carries the target commodity category;

and the generation module is used for extracting target commodity detail characteristic words from the commodity title data according to the target characteristic label marks, wherein the target commodity detail characteristic words are used for describing the commodity characteristics of the target commodity details.

9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method of any of claims 1 to 7 are implemented when the computer program is executed by the processor.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.