CN114332477A

CN114332477A - Feature recognition model training method, article feature recognition method and article feature recognition device

Info

Publication number: CN114332477A
Application number: CN202111585717.3A
Authority: CN
Inventors: 王兴鹏; 史世睿; 李天浩
Original assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2021-12-22
Filing date: 2021-12-22
Publication date: 2022-04-12

Abstract

The disclosure provides a feature recognition model training method, an article feature recognition method and an article feature recognition device. The training method comprises the following steps: acquiring an article sample data set, wherein the article sample data set comprises a plurality of article samples, respectively inputting article image data, text data for describing articles and article style data in each article sample into an image feature extraction layer, a text feature extraction layer and a style feature extraction layer of an initial model, and respectively outputting article image features, text features and style features; inputting the article image characteristics and the text characteristics into a characteristic combination layer of the initial model, and outputting article combination characteristics; inputting the article combination characteristics and the style characteristics into a matching layer of the initial model, and outputting a matching result; and adjusting the model parameters of the initial model according to the matching result and the style label to obtain the trained feature recognition model.

Description

Feature recognition model training method, article feature recognition method and article feature recognition device

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a feature recognition model training method, an article feature recognition method and apparatus, an electronic device, a computer-readable storage medium, and a computer program product.

Background

The traditional mode of article style labeling mainly adopts manual labeling, and this kind of mode needs to set for a large amount of manual editing rules in advance, and has the problem that consume manpower, material resources, labeling efficiency are low, the commonality is poor, moreover, in the setting process of manual editing rules, because there is the difference of artificial subjective cognition, can lead to the difference of setting for editing rules, and then lead to the result difference of manual labeling style.

In the related art, the labeling of the style of the article is realized by adopting a machine learning algorithm, but the generalization of the machine learning algorithm to the strange article and the strange style is poor.

Disclosure of Invention

In view of the above, the present disclosure provides a feature recognition model training method, an article feature recognition method and apparatus, an electronic device, a computer-readable storage medium, and a computer program product.

A first aspect of the present disclosure provides a feature recognition model training method, including:

the method comprises the steps of obtaining an article sample data set, wherein the article sample data set comprises a plurality of article samples, each article sample comprises article image data, text data used for describing an article and article style data, and the article samples are provided with style labels;

inputting article image data into an image feature extraction layer of the initial model aiming at each article sample, and outputting article image features; inputting text data for describing an article into a text feature extraction layer of an initial model, and outputting text features; inputting article style data into a style feature extraction layer of the initial model, and outputting style features;

inputting the article image characteristics and the text characteristics into a characteristic combination layer of the initial model, and outputting article combination characteristics;

inputting the article combination characteristics and the style characteristics into a matching layer of the initial model, and outputting a matching result for representing the article combination characteristics and the style characteristics;

and adjusting the model parameters of the initial model according to the matching result and the style label to obtain the trained feature recognition model.

According to an embodiment of the present disclosure, inputting the item image feature and the text feature into an image combination layer of the initial model, and outputting the item combination feature comprises:

and splicing the article image features and the text features into article combination features by using the image combination layer.

According to the embodiment of the present disclosure, after the acquiring the article sample data set, the method further includes:

and generating an amplified article sample data set according to the article sample data set.

According to an embodiment of the present disclosure, generating an augmented article sample data set from an article sample data set comprises:

determining a first item data list similar to the item image data from the item database according to the item image data, wherein the first item data list comprises item image data of different items, text data for describing the items and item style data;

according to the text data for describing the article, determining a second article data list similar to the text data for describing the article from the article database, wherein the second article data list comprises article image data of different articles, text data for describing the article and article style data;

and generating an augmented article sample data set according to the first article data list and the second article data list.

According to the embodiment of the disclosure, the model parameters of the initial model include model parameters of an image feature extraction layer of the initial model, model parameters of a text feature extraction layer of the initial model, and model parameters of a style feature extraction layer of the initial model, and the model parameters of the initial model are adjusted according to the matching result and the style label to obtain a trained feature recognition model, including:

and adjusting the model parameters of the image feature extraction layer of the initial model, the text feature extraction layer of the initial model and the style feature extraction layer of the initial model according to the matching result and the style label to obtain the trained feature recognition model.

A second aspect of the present disclosure provides an article feature identification method, including:

acquiring article information of an article to be processed, wherein the article information comprises article image information and text information for describing the article;

inputting the article image information into an image feature extraction layer of a feature recognition model, and outputting article image features of an article to be processed, wherein the feature recognition model is obtained by training through a training method of the embodiment of the disclosure;

inputting text information for describing the article into a text feature extraction layer of a feature recognition model, and outputting text features of the article to be processed;

inputting the article image features and the text features into a feature combination layer of a feature recognition model, and outputting article combination features of the article to be processed;

inputting the description characteristics of the article combination characteristics and the candidate style into a matching layer of a characteristic identification model, and outputting a matching result for representing the article combination characteristics and the style characteristics, wherein the description characteristics of the candidate style are obtained by inputting article style information acquired from a candidate style database into a style characteristic extraction layer of the characteristic identification model;

and determining style characteristic information matched with the article combination characteristics of the article to be processed according to the matching result.

According to an embodiment of the present disclosure, the feature identification method further includes:

acquiring a plurality of item comment text data and a plurality of item title text data;

preprocessing the comment text data of the plurality of articles and the title text data of the plurality of articles to obtain a text data set for representing the style of the articles;

a candidate style database is generated from the text data set used to characterize the style of the item.

generating a vector data set for characterizing the style of the article from the text data set for characterizing the style of the article;

a candidate style database is generated from the vector data set for characterizing the style of the item.

A third aspect of the present disclosure provides a feature recognition model training apparatus, including: the device comprises a first acquisition module, a feature extraction module, a feature combination module, a matching module and an adjustment module. The system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring an article sample data set, the article sample data set comprises a plurality of article samples, each article sample comprises article image data, text data used for describing an article and article style data, and the article sample has a style label. The characteristic extraction module is used for inputting the article image data into an image characteristic extraction layer of the initial model and outputting article image characteristics aiming at each article sample; inputting text data for describing an article into a text feature extraction layer of an initial model, and outputting text features; and inputting the style data of the article into a style characteristic extraction layer of the initial model, and outputting style characteristics. And the characteristic combination module is used for inputting the article image characteristics and the text characteristics into the characteristic combination layer of the initial model and outputting article combination characteristics. And the matching module is used for inputting the article combination characteristics and the style characteristics into a matching layer of the initial model and outputting a matching result for representing the article combination characteristics and the style characteristics. And the adjusting module is used for adjusting the model parameters of the initial model according to the matching result and the style label to obtain the trained feature recognition model.

According to an embodiment of the present disclosure, the feature combination module includes a stitching unit for stitching the article image feature and the text feature into an article combination feature using the image combination layer.

According to an embodiment of the present disclosure, the apparatus further includes: and the generating module is used for generating an amplified article sample data set according to the article sample data set.

According to an embodiment of the present disclosure, a generation module includes a first determination unit, a second determination unit, and a generation unit. The first determining unit determines a first item data list similar to the item image data from the item database according to the item image data, wherein the first item data list comprises item image data of different items, text data for describing the items and item style data. And the second determining unit is used for determining a second item data list similar to the text data for describing the item from the item database according to the text data for describing the item, wherein the second item data list comprises item image data of different items, the text data for describing the item and item style data. And the generating unit is used for generating an amplified article sample data set according to the first article data list and the second article data list.

According to an embodiment of the present disclosure, the generating unit includes a generating subunit configured to generate an augmented item sample data set according to a data intersection of the first item data list and the second item data list.

According to the embodiment of the disclosure, the adjusting module comprises an adjusting unit, and the adjusting unit is used for adjusting the model parameters of the image feature extraction layer of the initial model, the model parameters of the text feature extraction layer of the initial model and the model parameters of the style feature extraction layer of the initial model according to the matching result and the style label to obtain the trained feature recognition model.

A fourth aspect of the present disclosure provides an article feature identification device, including: the device comprises a second obtaining module, a first identification module, a second identification module, a third identification module, a fourth identification module and a determination module. The second acquisition module is used for acquiring the article information of the article to be processed, wherein the article information comprises article image information and text information used for describing the article. The first identification module is used for inputting the article image information into an image feature extraction layer of a feature identification model and outputting article image features of the article to be processed, wherein the feature identification model is obtained by training through the training method provided by the embodiment of the disclosure. And the second identification module is used for inputting the text information for describing the article into a text feature extraction layer of the feature identification model and outputting the text feature of the article to be processed. And the third identification module is used for inputting the article image characteristics and the text characteristics into a characteristic combination layer of the characteristic identification model and outputting article combination characteristics of the article to be processed. And the fourth identification module is used for inputting the item combination characteristics and the description characteristics of the candidate style into a matching layer of the characteristic identification model and outputting a matching result for representing the item combination characteristics and the description characteristics of the candidate style, wherein the description characteristics of the candidate style are obtained by inputting the item style information acquired from the candidate style database into a style characteristic extraction layer of the characteristic identification model. And the determining module is used for determining style characteristic information matched with the article combination characteristic of the article to be processed according to the matching result.

A fifth aspect of the present disclosure provides an electronic device, comprising: one or more processors; a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the above-described feature recognition model training method.

The sixth aspect of the present disclosure also provides a computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform the above-mentioned feature recognition model training method.

A seventh aspect of the present disclosure also provides a computer program product comprising a computer program which, when executed by a processor, implements the above-described feature recognition model training method.

According to an embodiment of the present disclosure, since an article sample data set including article image data, text data for describing an article, article style data and style tags is employed, the image data of the article and the text data for describing the article are respectively input into an image feature extraction layer of an initial model, a text feature extraction layer outputs image features and text features, then a feature combination layer outputs article combination features, the article combination features and style features extracted by a style feature extraction layer of the initial model are input into a matching layer, the feature recognition model is obtained through the technical means of adjusting the model parameters of the initial model by the output matching result and the style label, therefore, the technical problem that the generalization of the machine learning algorithm to strange objects and styles is poor in the related technology is at least partially solved, and the generalization of the identification to the strange objects and the strange styles is improved.

Drawings

The above and other objects, features and advantages of the present disclosure will become more apparent from the following description of embodiments of the present disclosure with reference to the accompanying drawings, in which:

FIG. 1 schematically illustrates an exemplary system architecture to which the feature recognition model training method of embodiments of the present disclosure may be applied;

FIG. 2 schematically illustrates a flow diagram of a method of feature recognition model training in accordance with an embodiment of the present disclosure;

fig. 3 schematically shows a flow chart of a method of amplifying a sample data set according to an embodiment of the present disclosure;

FIG. 4 schematically illustrates a feature recognition model architecture diagram according to an embodiment of the present disclosure;

FIG. 5 schematically illustrates a flow chart of an item feature identification method according to an embodiment of the present disclosure;

FIG. 6 schematically illustrates a flow chart of a candidate style database generation method according to an embodiment of the present disclosure;

FIG. 7 schematically illustrates a system architecture diagram for item feature identification in accordance with an embodiment of the present disclosure;

FIG. 8 schematically illustrates a block diagram of a feature recognition model training apparatus according to an embodiment of the present disclosure;

FIG. 9 schematically illustrates a block diagram of an item feature identification device, in accordance with an embodiment of the present disclosure; and

FIG. 10 schematically illustrates a block diagram of an electronic device suitable for implementing a feature recognition model training method according to an embodiment of the present disclosure.

Detailed Description

Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood that the description is illustrative only and is not intended to limit the scope of the present disclosure. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the disclosure. It may be evident, however, that one or more embodiments may be practiced without these specific details. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present disclosure.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The terms "comprises," "comprising," and the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.

All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It is noted that the terms used herein should be interpreted as having a meaning that is consistent with the context of this specification and should not be interpreted in an idealized or overly formal sense.

Where a convention analogous to "at least one of A, B and C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B and C" would include but not be limited to systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).

The embodiment of the disclosure provides a feature recognition model training method, which adopts an article sample data set comprising article image data, text data for describing articles, article style data and article style labels, and obtains a feature recognition model through training by respectively inputting the article image data and the text data for describing articles into an image feature extraction layer and a text feature extraction layer of an initial model, then inputting the image feature and the text feature into a feature combination layer to output article combination features, inputting the article combination features and the style features extracted by the style feature extraction layer of the initial model into a matching layer, and adjusting model parameters of the initial model through the output matching results and the article style labels.

FIG. 1 schematically illustrates an exemplary system architecture 100 to which a feature recognition model training method may be applied, according to an embodiment of the present disclosure. It should be noted that fig. 1 is only an example of a system architecture to which the embodiments of the present disclosure may be applied to help those skilled in the art understand the technical content of the present disclosure, and does not mean that the embodiments of the present disclosure may not be applied to other devices, systems, environments or scenarios.

As shown in fig. 1, the system architecture 100 according to this embodiment may include

terminal devices

101, 102, 103, a network 104 and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired and/or wireless communication links, and so forth.

The user may use the

terminal devices

101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like. The

terminal devices

101, 102, 103 may have installed thereon various communication client applications, such as a shopping-like application, a web browser application, a search-like application, an instant messaging tool, a mailbox client, and/or social platform software, etc. (by way of example only).

The

terminal devices

101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.

The server 105 may be a server providing various services, such as a background management server (for example only) providing support for websites browsed by users using the

terminal devices

101, 102, 103. The background management server may analyze and perform other processing on the received data such as the user request, and feed back a processing result (e.g., a webpage, information, or data obtained or generated according to the user request) to the terminal device.

It should be noted that the feature recognition model training method or the feature recognition method provided by the embodiments of the present disclosure may be generally executed by the server 105. Accordingly, the feature recognition model training apparatus or the feature recognition apparatus provided by the embodiments of the present disclosure may be generally disposed in the server 105. The feature recognition model training method or the feature recognition method provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the

terminal devices

101, 102, 103 and/or the server 105. Accordingly, the feature recognition model training apparatus or the feature recognition apparatus provided in the embodiments of the present disclosure may also be disposed in a server or a server cluster different from the server 105 and capable of communicating with the

terminal devices

101, 102, 103 and/or the server 105. Alternatively, the feature recognition model training method or the feature recognition method provided by the embodiment of the present disclosure may also be executed by the

terminal device

101, 102, or 103, or may also be executed by another terminal device different from the

terminal device

101, 102, or 103. Accordingly, the feature recognition model training apparatus or the feature recognition apparatus provided in the embodiments of the present disclosure may also be disposed in the

terminal device

101, 102, or 103, or disposed in another terminal device different from the

terminal device

101, 102, or 103.

For example, the item sample data set may be obtained by any one of the

terminal devices

101, 102, or 103 (for example, but not limited to, the terminal device 101), may be stored in any one of the

terminal devices

101, 102, or 103, or may be stored on an external storage device and may be imported into the terminal device 101. Then, the terminal device 101 may locally execute the feature recognition model training method provided by the embodiment of the present disclosure, or send the article sample data set to another terminal device, a server, or a server cluster, and execute the feature recognition model training method provided by the embodiment of the present disclosure by another terminal device, a server, or a server cluster that receives the article sample data set.

It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

FIG. 2 schematically shows a flow diagram of a feature recognition model training method according to an embodiment of the present disclosure.

As shown in fig. 2, the feature recognition model training method of this embodiment includes operations S210 to S270.

In operation S210, an article sample data set is obtained, where the article sample data set includes a plurality of article samples, each article sample includes article image data, text data for describing an article, and article style data, and the article sample has a style label.

According to an embodiment of the present disclosure, the item image data may include an item picture, which may include a model or a display shelf, a display case, etc. displaying the item, in addition to the item itself. Textual data describing the item may include textual data describing the color of the item, textual data describing the brand of the item, textual data describing the style of the item, and so forth, for example: red shirts, blue certain sports brand T-shirts, etc. The item style data may include sports styles, leisure styles, business styles, sweet styles, and the like. The style label can represent whether the article is matched with the style by adopting binary number, and when the article is matched with the style, the style label is 1; when the item does not match the style, the style label is 0. For example: blue the style of a sports brand T-shirt is sports style, then the style label is 1. The style of blue certain sports brand T-shirt is gentlewoman style, and the style label is 0.

In operation S220, for each item sample, item image data is input into the image feature extraction layer of the initial model, and item image features are output.

According to an embodiment of the present disclosure, for example: the article image data is a picture of a certain blue sports brand T-shirt, the article image data is input into an image feature extraction layer of an initial model, and the output article image features can comprise blue, a mark of the sports brand, the T-shirt and the like.

In operation S230, text data describing the article is input into a text feature extraction layer of the initial model, and text features are output.

According to an embodiment of the present disclosure, for example: the text data for describing the article is blue T-shirt of certain sports brand, the image data of the article is input into a text data extraction layer of an initial model, and the output text data features can include blue, name of the sports brand, T-shirt and the like.

In operation S240, the style data of the article is input into the style feature extraction layer of the initial model, and style features are output.

According to an embodiment of the present disclosure, for example: and if the style of the blue T-shirt of a certain sports brand is a sports style, the article style label is 1, which represents that the article is matched with the style, and the sample data is positive sample data. And inputting the style data of the article into a style characteristic extraction layer of the initial model, wherein the output style characteristic can be motion.

In operation S250, the item image feature and the text feature are input into the feature combination layer of the initial model, and the item combination feature is output.

According to the embodiment of the present disclosure, the image feature "blue, mark of sports brand, T-shirt" and the text feature "blue, name of sports brand, T-shirt" of the article are input into the feature combination layer, and the output article combination feature is "blue, mark of sports brand, name of sports brand, T-shirt".

In operation S260, the item combination feature and the style feature are input into a matching layer of the initial model, and a matching result for characterizing the item combination feature and the style feature is output.

According to the embodiment of the disclosure, the article combination feature "blue, the logo of the sports brand, the name of the sports brand, the T-shirt" and the style feature "sports" are input into the matching layer of the initial model, and the matching result for characterizing the article combination feature and the style feature is output, for example, the matching result is 0.8.

According to an embodiment of the present disclosure, the item combination feature and the style feature may be represented in a vector, and the matching layer of the initial model may be represented by calculating a distance between the item combination feature vector and the style feature vector, for example: euclidean distance, cosine similarity and the like, and determining a matching result for characterizing the combination characteristic and the style characteristic of the article.

In operation S270, model parameters of the initial model are adjusted according to the matching result and the style label, so as to obtain a trained feature recognition model.

According to the embodiment of the present disclosure, since the input is positive sample data and the style label is 1, the model parameters of the initial model may be adjusted according to the matching result 0.8 and the style label 1, after the adjustment is completed, the output matching result may be a numerical value of 0.9, 0.95 or closer to 1, and the closer to 1 the output matching result is, the higher the recognition accuracy of the feature recognition model is proved.

According to an embodiment of the present disclosure, since an item sample data set including item image data, text data for describing an item, item style data, and an item style label is employed, the image data of the article and the text data for describing the article are respectively input into an image feature extraction layer of an initial model, a text feature extraction layer outputs image features and text features, then a feature combination layer outputs article combination features, the article combination features and style features extracted by a style feature extraction layer of the initial model are input into a matching layer, the feature recognition model is obtained through the technical means of adjusting the model parameters of the initial model by the output matching result and the style label, therefore, the technical problem that the generalization of the machine learning algorithm to strange objects and styles is poor in the related technology is at least partially solved, and the generalization of the identification to the strange objects and the strange styles is improved.

According to the embodiment of the disclosure, both the article image feature and the text feature can be represented by vectors, and the article image feature vector and the text feature vector can be spliced into a one-dimensional vector by using the image combination layer, for example: the feature vector of the article image is (a, b, c), and the feature vector of the text is (e, f, g), then the feature vector of the article combination spliced together may be (a, b, c, e, f, g).

According to the embodiment of the present disclosure, the image combination layer may also be used to splice the feature vector of the article image and the feature vector of the text into a two-dimensional vector or matrix, which is not described herein again.

According to the embodiment of the disclosure, the article combination characteristics are obtained in a splicing mode, and the image characteristics and the text characteristics of the article can be combined for identifying the style matched with the article, so that the accuracy of model prediction is improved.

According to an embodiment of the present disclosure, after the article sample data set is acquired, the training method further includes:

According to the embodiment of the disclosure, the amplification of the article sample data set comprises the step of respectively carrying out depth representation on the picture data and the text data in the article sample data set through two modes of picture data enhancement and text data enhancement on the basis of the obtained article sample data set, so as to obtain the amplification article sample data set.

According to an embodiment of the present disclosure, for example: the article image data included in the article sample data may be a picture of a black sports shoe with a certain brand mark, the text data for describing the article may be a black sports shoe of a certain brand, the article style data is a sports style, and the article style label is 1. The article sample data may be amplified according to the common features of the article sample data, for example, the amplified article sample data may be article image data of red or green sports shoes with certain brand marks, text data and article style data for describing articles, article style labels, or may be article image data of various sports shoes with different colors, which do not have certain brand marks and have styles similar to the styles in the article sample data, text data and article style data for describing articles, style labels, and so on. Since the above-listed article sample data sets are all positive sample data matching the style of the article, the style label is 1.

According to the embodiment of the present disclosure, when performing model training, negative sample data that an article does not match with a style may also be adopted, for example: the negative sample data comprises article image data which can be a black sports shoe picture with a certain brand mark, text data used for describing the article can be a certain brand black sports shoe, article style data is a business style, and a style label is 0.

According to the embodiment of the disclosure, when the feature recognition model training method of the embodiment of the disclosure is implemented, the proportion range of the positive sample data and the negative sample data can be 1: 1-1: 5.

According to the embodiment of the disclosure, the article sample data set mainly comes from field expert data and a small amount of manually marked data, the data flow is limited, an amplified article sample data set is generated, and the initial model is trained by utilizing the amplified article sample data set and the article sample data, so that the accuracy of the identification and prediction of the feature identification model can be improved.

The method shown in fig. 2 is further described with reference to fig. 3-7 in conjunction with specific embodiments.

Fig. 3 schematically shows a flow chart of a method of generating an amplified sample data set according to an embodiment of the present disclosure.

As shown in fig. 3, generating the amplification sample data set of this embodiment includes operations S310 to S330.

In operation S310, a first item data list similar to the item image data is determined from the item database according to the item image data, wherein the first item data list includes item image data of different items, text data describing the items, and item style data.

According to an embodiment of the present disclosure, for example: taking the article image data as a picture of a pink dress with a lace as an example, it can be determined from the article database that the first article data list similar to the article image data may include a plurality of different articles, for example: red one-piece dress, half-length dress with lace and pink one-piece dress with wave points. Taking a red dress item in the first item data list as an example, the first item data list includes a picture of the red dress, for example: a picture of a girl wearing the red dress; text data describing the red dress, for example: red, one-piece dress; article style data, such as: sweet style and gentlewoman style.

In operation S320, a second item data list similar to the text data for describing the item is determined from the item database according to the text data for describing the item, wherein the second item data list includes item image data of different items, text data for describing the item, and item style data.

According to an embodiment of the present disclosure, for example: taking the text data of the pink dress with the lace as an example, from the article database, it can be determined that the second article data list similar to the text data for describing the article includes a plurality of different articles, for example: a half-length dress with a lace, a red one-piece dress with a lace and a pink one-piece dress with a wave point. Taking the example that the second article data list comprises the half-skirt with the lace as an example, the article image data included in the second article data list is a picture of the half-skirt with the lace, the text data used for describing the article is the lace and the half-skirt, and the style data of the article is a gentlewoman style and a sweet style.

In operation S330, an augmented item sample data set is generated according to the first item data list and the second item data list.

According to an embodiment of the present disclosure, an augmented article sample data set may be generated according to all data in the first article data list and the second article data list, and the generated augmented article sample data set includes article image data of a red dress, a half-length with a lace, a pink dress with a dot, a red dress with a lace, text data for describing an article, and style data of the article.

According to the embodiment of the disclosure, the item data lists are respectively generated through the item image data and the text data for describing the item, and the item sample data can be automatically amplified, so that the problems of low model prediction accuracy and poor generalization caused by the fact that the model training is only performed depending on the obtained sample data in the related technology are solved.

According to an embodiment of the present disclosure, generating an augmented item sample data set from a first item data list and a second item data list comprises:

and generating an amplified article sample data set according to the data intersection of the first article data list and the second article data list.

According to an embodiment of the present disclosure, for example: the first item data list comprises a red dress, a half-length dress with lace and a pink dress with dots. The second article data list comprises a half-length dress with a lace, a red one-piece dress with a lace and a pink one-piece dress with a dot. The generated sample data set of the augmented article comprises a picture of the article image data as a pink half skirt with a lace, and the text data for describing the article comprises: bud thread lace, pink, dress, half-length, red, the style of article includes: gentlewoman style and sweet style.

According to the embodiment of the disclosure, the article data lists are respectively generated through the article image data and the text data for describing the article, and then the intersection of the two data lists is taken, so that the sample data is automatically amplified, and the accuracy of the amplified sample data is improved.

According to an embodiment of the present disclosure, the style tag may be represented by a binary number, and when the style tag is 1, it represents that the article in the article sample data matches the style; when the style label is 0, it indicates that the article in the article sample data does not match the style.

According to the embodiment of the disclosure, for example, after positive article sample data of an article and a style are input into an initial model, the output matching result is 0.8, and the style label is 1, the output matching result can be changed by adjusting model parameters of an image feature extraction layer, a text feature extraction layer and a style feature extraction layer in the initial model.

According to an embodiment of the present disclosure, a difference between the matching result and the actual result may be calculated using a formula shown in formula (one).

The XOR is an exclusive-or function, when 2 numbers with values of 0 or 1 are equal, the return value of the XOR function is 0, otherwise, the return value is 1;

(Item_i，Style_j)_truethe matching relation between the actual article and the style is 0 or 1, 1 represents correspondence, and 0 represents no correspondence;

(Item_i，Style_j)_predthe matching relation between the article and the style predicted by the algorithm is represented by floating point numbers in a value range of 0-1, and the larger the numerical value is, the higher the matching degree between the article and the style is considered by the algorithm;

i (-) is an indicator function where it returns 0 when the input value is less than 0 and returns 1 when the input value is greater than 0.

According to the embodiment of the disclosure, a plurality of matching results and the style label output after model parameters are adjusted for many times are input into the loss function, and when the change of the loss function approaches zero, the matching degree of the image features, the text features and the style features extracted by the image feature extraction layer, the text feature extraction layer and the style feature extraction layer in the initial model is the highest, which indicates that the prediction accuracy of the feature recognition model is the highest, and then the training of the initial model is completed, so that the trained feature recognition model is obtained.

According to the embodiment of the disclosure, the image feature extraction layer, the text feature extraction layer and the style feature extraction layer of the initial model are adjusted through the matching result output by the matching layer and the style label, so that a supervised model training process is realized, and the accuracy of model identification prediction is improved.

FIG. 4 schematically shows a feature recognition model architecture diagram according to an embodiment of the present disclosure.

As shown in fig. 4, the feature recognition model of this embodiment includes three layers of architectures: the bottom layer comprises an image feature extraction layer, a text feature extraction layer and a style feature extraction layer which are respectively used for extracting image features, text features and style features in the sample data of the article. The intermediate layer includes a feature combination layer for combining the image features with the text features to form article combination features. The top layer comprises a matching layer used for matching the combination characteristics and style characteristics of the articles, outputting a matching result and adjusting the model parameters of the bottom layer according to the matching result and the style labels.

Fig. 5 schematically illustrates a flow chart of an item feature identification method according to an embodiment of the present disclosure.

As shown in fig. 5, the article feature identification method of this embodiment includes operations S510 to S560.

In operation S510, item information of an item to be processed is acquired, wherein the item information includes item image information and text information for describing the item.

According to an embodiment of the present disclosure, for example: the article image information included in the article information with the processed article is a dark blue picture of the jeans coat with a fur collar and a certain brand special pattern, and the text information used for describing the article is the dark blue fur collar jeans coat of a certain brand.

In operation S520, the article image information is input into the image feature extraction layer of the feature recognition model, and the article image features of the article to be processed are output, wherein the feature recognition model is obtained by training through the training method according to the embodiment of the disclosure.

According to the embodiment of the disclosure, the image of the jeans coat with the deep blue band fur collar and the special pattern of a certain brand is input into the image feature extraction layer of the feature recognition model obtained by training with the training method provided by the embodiment of the disclosure, and the output article image features may include: dark blue, fur collar, jeans coat, certain brand-specific pattern, which may be the pattern of clover, for example.

In operation S530, text information describing the article is input into a text feature extraction layer of the feature recognition model, and text features of the article to be processed are output.

According to the embodiment of the disclosure, the text information of a certain brand of deep blue fur collar jeans coat is input into the text feature extraction layer of the feature recognition model, and the output text features of the article to be processed may include: dark blue, certain brand, fur collar, jeans coat.

In operation S540, the article image feature and the text feature are input into the feature combination layer of the feature recognition model, and the article combination feature of the article to be processed is output.

According to the embodiment of the disclosure, the article image feature "dark blue, fur collar, cowboy coat, a certain brand specific pattern, for example, a pattern of clover" and the text feature "dark blue, a certain brand, fur collar, cowboy coat" are input into the feature combination layer of the feature recognition model, and the article combination feature of the article to be processed, which may be "dark blue, fur collar, cowboy coat, pattern of clover, a certain brand", is output.

In operation S550, the item combination feature and the description feature of the candidate style obtained by inputting the item style information acquired from the candidate style database into the style feature extraction layer of the feature recognition model are input into the matching layer of the feature recognition model, and a matching result for characterizing the item combination feature and the style feature is output.

According to an embodiment of the present disclosure, for example: and inputting the descriptive characteristics of the article combination characteristics of 'dark blue, fur collar, cowboy coat and clover pattern, a certain brand' and the descriptive characteristics of the candidate style, for example, the descriptive characteristics of the candidate style are leisure style, into a matching layer of the feature recognition model, and outputting a matching result.

In operation S560, style feature information that the commodity style information of the to-be-processed commodity matches the item combination feature of the to-be-processed item is determined according to the matching result.

According to the embodiment of the disclosure, a plurality of style features can be obtained after a plurality of style information of the article are acquired from the candidate style database and input into the style feature extraction layer of the feature recognition model. And inputting the article combination characteristics and the style characteristics into a matching layer of the characteristic identification model, sequencing the obtained matching results, and if the article to be processed only retains one style, adopting the style with the highest matching result as article style information of the article to be processed. If the to-be-processed article can reserve N styles, the styles corresponding to the N matching results can be sequentially taken from top to bottom from the sorted matching results as the article style information of the to-be-processed article.

According to the embodiment of the disclosure, the image feature and the text feature of the article matched with the style can be identified through the feature identification model, the image feature and the text feature of the article are combined to form the article combination feature, the style feature is extracted from the candidate style database by using the style feature extraction layer in the feature identification model, the style feature matched with the article combination feature is determined, the style of the article is determined, and the technical effect of automatically determining the style of a new article is achieved.

Fig. 6 schematically shows a flowchart of a candidate style database generation method according to an embodiment of the present disclosure.

As shown in fig. 6, the candidate style database generation method of this embodiment includes: s610 to S630.

In operation S610, a plurality of item review text data and a plurality of item title text data are acquired.

According to embodiments of the present disclosure, the item review text data may include that the article of clothing is suitable for home, travel, or any leisure situation, comfortable to wear, and the like. Item title text data may include home, travel must, comfort, and the like.

In operation S620, the text data of the comments on the plurality of items and the text data of the titles of the plurality of items are preprocessed, so as to obtain a text data set for characterizing the style of the item.

According to embodiments of the present disclosure, preprocessing may include cleaning the data, such as word cutting, removal of stop words, removal of punctuation marks, removal of feature symbols, filtering of high frequency words, filtering of low frequency words, and so forth. The preprocessing further comprises the step of carrying out clustering calculation on the washed data to obtain a text data set for representing the style of the article. For example: leisure, sports, and the like.

In operation S630, a candidate style database is generated from the text data set for characterizing the style of the item.

According to an embodiment of the present disclosure, a text data set for characterizing a style of an item may be stored in a database, generating a candidate style database.

According to the embodiment of the disclosure, the comment data and the title data of the article are acquired, and the text data set for representing the style of the article is obtained through data preprocessing. The problem that a candidate style database is lacked or is difficult to automatically generate in the machine learning process is solved.

According to an embodiment of the present disclosure, the method of generating a candidate style database further comprises:

a vector data set for characterizing the style of the item is generated from the text data set for characterizing the style of the item.

According to the embodiment of the disclosure, the data in the candidate style database can be represented by the vector, and when the style is matched for a new article, the style data can be inquired from the candidate style database through vector retrieval, so that the matching time can be saved.

In order to compare the prediction times of the vector search and the normal search, the complexity of the prediction times of the two search methods can be expressed by the formulas (two) and (three).

cost ═ I | J | d (two)

Wherein t is iteration times, | J | is the number of styles, k is the number of clustering centers, | I | is the number of articles, c is the number of finding the nearest clustering centers, and d is the vector dimension of the model training result.

Equation (two) represents the predicted temporal complexity of a normal search, and equation (three) represents the predicted temporal complexity of a brute-force search.

Since t and c are both constants, and k is a number that is much smaller than | J | and much smaller than | I |, the calculation result of equation (three) is much smaller than that of equation (two), i.e., the prediction period of the vector search algorithm than the normal search algorithm.

FIG. 7 schematically illustrates a system architecture diagram for item feature identification in accordance with an embodiment of the present disclosure.

As shown in fig. 7, the item feature recognition system includes three parts, namely a model training module, a candidate style database generation module and a model prediction module.

And the model training module is used for amplifying the sample data by acquiring the field expert data and the manual marking data, and training the initial model by using the amplified sample data to obtain the feature recognition model.

And the candidate style database generation module is used for obtaining the candidate style database according to the method for generating the candidate style database in the embodiment of the disclosure by acquiring the item comment data and the item title data.

The model prediction module inputs the article image data of the article to be processed into the image feature extraction layer to extract the image features of the article to be processed; extracting text features of the article to be processed from the text data of the article to be processed, which is used for describing the article; and combining the image characteristics and the text characteristics of the article to be processed by using the article characteristic combination layer to obtain article combination characteristics. And matching the article combination characteristics with the description characteristics of the candidate styles extracted by the style characteristic extraction layer on the matching layer to determine the style of the article to be processed.

FIG. 8 schematically shows a block diagram of a feature recognition model training apparatus according to an embodiment of the present disclosure.

As shown in fig. 8, the feature recognition model training apparatus 800 includes: a first obtaining module 810, a feature extracting module 820, a feature combining module 830, a matching module 840, and an adjusting module 850.

The first obtaining module 810 is configured to obtain an article sample data set, where the article sample data set includes a plurality of article samples, each article sample includes article image data, text data used for describing an article, and article style data, and the article sample has an article style label.

A feature extraction module 820, configured to, for each item sample, input item image data into an image feature extraction layer of the initial model, and output item image features; inputting text data for describing an article into a text feature extraction layer of an initial model, and outputting text features; and inputting the style data of the article into a style characteristic extraction layer of the initial model, and outputting style characteristics.

And the feature combination module 830 is configured to input the article image features and the text features into the feature combination layer of the initial model, and output the article combination features.

And the matching module 840 is used for inputting the item combination characteristics and the style characteristics into the matching layer of the initial model and outputting the matching result for representing the item combination characteristics and the style characteristics.

And the adjusting module 850 is used for adjusting the model parameters of the initial model according to the matching result and the style label to obtain the trained feature recognition model.

According to an embodiment of the present disclosure, the feature combination module 830 includes a stitching unit for stitching the article image feature and the text feature into an article combination feature using the image combination layer.

Fig. 9 schematically illustrates a block diagram of an item feature identification device according to an embodiment of the present disclosure.

As shown in fig. 9, the article characteristic identification apparatus of this embodiment includes a second obtaining module 910, a first identifying module 920, a second identifying module 930, a third identifying module 940, a fourth identifying module 950, and a determining module 960.

A second obtaining module 910, configured to obtain item information of an item to be processed, where the item information includes item image information and text information used for describing the item.

The first identification module 920 is configured to input the article image information into an image feature extraction layer of a feature identification model, and output the article image features of the article to be processed, where the feature identification model is obtained by training through a training method provided in an embodiment of the present disclosure.

A second recognition module 930, configured to input the text information for describing the article into the text feature extraction layer of the feature recognition model, and output the text feature of the article to be processed.

And a third identification module 940, configured to input the article image feature and the text feature into a feature combination layer of the feature identification model, and output an article combination feature of the to-be-processed article.

A fourth recognition module 950, configured to input the item combination feature and the description feature of the candidate style into a matching layer of the feature recognition model, and output a matching result for characterizing the item combination feature and the description feature of the candidate style, where the description feature of the candidate style is obtained by inputting the item style information acquired from the candidate style database into a style feature extraction layer of the feature recognition model.

A determining module 960, configured to determine style characteristic information matched with the article combination characteristic of the article to be processed according to the matching result. It should be noted that, the embodiments of the apparatus portion of the present disclosure correspond to the same or similar embodiments of the method portion of the present disclosure, and the detailed description of the present disclosure is omitted here.

Any of the modules, units, sub-units, or at least part of the functionality of any of them according to embodiments of the present disclosure may be implemented in one module. Any one or more of the modules, units and sub-units according to the embodiments of the present disclosure may be implemented by being split into a plurality of modules. Any one or more of the modules, units, sub-units according to the embodiments of the present disclosure may be implemented at least partially as a hardware circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented in any other reasonable manner of hardware or firmware by integrating or packaging a circuit, or in any one of three implementations of software, hardware, and firmware, or in any suitable combination of any of them. Alternatively, one or more of the modules, units, sub-units according to embodiments of the disclosure may be at least partially implemented as computer program modules, which, when executed, may perform the corresponding functions.

For example, any plurality of the first obtaining module 810, the feature extracting module 820, the feature combining module 830, the matching module 840, the adjusting module 850 or the second obtaining module 910, the first identifying module 920, the second identifying module 930, the third identifying module 940, the fourth identifying module 950, and the determining module 960 may be combined into one module/unit/sub-unit to be implemented, or any one of the modules/units/sub-units may be split into a plurality of modules/units/sub-units. Alternatively, at least part of the functionality of one or more of these modules/units/sub-units may be combined with at least part of the functionality of other modules/units/sub-units and implemented in one module/unit/sub-unit. According to an embodiment of the present disclosure, at least one of the first obtaining module 810, the feature extracting module 820, the feature combining module 830, the matching module 840, the adjusting module 850 or the second obtaining module 910, the first identifying module 920, the second identifying module 930, the third identifying module 940, the fourth identifying module 950, and the determining module 960 may be implemented at least partially as a hardware circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented by hardware or firmware such as any other reasonable manner of integrating or packaging a circuit, or implemented by any one of three implementations of software, hardware, and firmware, or by a suitable combination of any of them. Alternatively, at least one of the first obtaining module 810, the feature extracting module 820, the feature combining module 830, the matching module 840, the adjusting module 850 or the second obtaining module 910, the first identifying module 920, the second identifying module 930, the third identifying module 940, the fourth identifying module 950 and the determining module 960 may be at least partially implemented as a computer program module, which when executed, may perform a corresponding function.

Fig. 10 schematically shows a block diagram of an electronic device adapted to implement the above described method according to an embodiment of the present disclosure. The electronic device shown in fig. 10 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present disclosure.

As shown in fig. 10, an electronic device 1000 according to an embodiment of the present disclosure includes a processor 1001 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM)1002 or a program loaded from a storage section 1008 into a Random Access Memory (RAM) 1003. Processor 1001 may include, for example, a general purpose microprocessor (e.g., a CPU), an instruction set processor and/or associated chipset, and/or a special purpose microprocessor (e.g., an Application Specific Integrated Circuit (ASIC)), among others. The processor 1001 may also include onboard memory for caching purposes. The processor 1001 may comprise a single processing unit or a plurality of processing units for performing the different actions of the method flow according to embodiments of the present disclosure.

In the RAM 1003, various programs and data necessary for the operation of the electronic apparatus 1000 are stored. The processor 1001, ROM 1002, and RAM 1003 are connected to each other by a bus 1004. The processor 1001 performs various operations of the method flow according to the embodiments of the present disclosure by executing programs in the ROM 1002 and/or the RAM 1003. Note that the programs may also be stored in one or more memories other than the ROM 1002 and the RAM 1003. The processor 1001 may also perform various operations of the method flows according to embodiments of the present disclosure by executing programs stored in the one or more memories.

Electronic device 1000 may also include an input/output (I/O) interface 1005, the input/output (I/O) interface 1005 also being connected to bus 1004, according to an embodiment of the present disclosure. The system 1000 may also include one or more of the following components connected to the I/O interface 1005: an input section 1006 including a keyboard, a mouse, and the like; an output section 1007 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage portion 1008 including a hard disk and the like; and a communication section 1009 including a network interface card such as a LAN card, a modem, or the like. The communication section 1009 performs communication processing via a network such as the internet. The driver 1010 is also connected to the I/O interface 1005 as necessary. A removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 1010 as necessary, so that a computer program read out therefrom is mounted into the storage section 1008 as necessary.

According to embodiments of the present disclosure, method flows according to embodiments of the present disclosure may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable storage medium, the computer program containing program code for performing the method illustrated by the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication part 1009 and/or installed from the removable medium 1011. The computer program performs the above-described functions defined in the system of the embodiment of the present disclosure when executed by the processor 1001. The systems, devices, apparatuses, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the present disclosure.

The present disclosure also provides a computer-readable storage medium, which may be contained in the apparatus/device/system described in the above embodiments; or may exist separately and not be assembled into the device/apparatus/system. The computer-readable storage medium carries one or more programs which, when executed, implement the method according to an embodiment of the disclosure.

According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples may include, but are not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

For example, according to embodiments of the present disclosure, a computer-readable storage medium may include the ROM 1002 and/or the RAM 1003 described above and/or one or more memories other than the ROM 1002 and the RAM 1003.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Those skilled in the art will appreciate that various combinations and/or combinations of features recited in the various embodiments and/or claims of the present disclosure can be made, even if such combinations or combinations are not expressly recited in the present disclosure. In particular, various combinations and/or combinations of the features recited in the various embodiments and/or claims of the present disclosure may be made without departing from the spirit or teaching of the present disclosure. All such combinations and/or associations are within the scope of the present disclosure.

The embodiments of the present disclosure have been described above. However, these examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments are described separately above, this does not mean that the measures in the embodiments cannot be used in advantageous combination. The scope of the disclosure is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be devised by those skilled in the art without departing from the scope of the present disclosure, and such alternatives and modifications are intended to be within the scope of the present disclosure.

Claims

1. A feature recognition model training method comprises the following steps:

acquiring an article sample data set, wherein the article sample data set comprises a plurality of article samples, each article sample comprises article image data, text data for describing an article and article style data, and the article samples are provided with style labels;

inputting the article image data into an image feature extraction layer of an initial model and outputting article image features aiming at each article sample; inputting the text data for describing the article into a text feature extraction layer of the initial model, and outputting text features; inputting the article style data into a style feature extraction layer of the initial model, and outputting style features;

inputting the article image features and the text features into a feature combination layer of the initial model, and outputting article combination features;

and adjusting the model parameters of the initial model according to the matching result and the style label to obtain a trained feature recognition model.

2. The training method of claim 1, wherein the item image features and the text features are input into an image combination layer of the initial model, and outputting item combination features comprises:

and splicing the article image features and the text features into the article combination features by utilizing the image combination layer.

3. The training method of claim 1, wherein after acquiring the sample set of article data, further comprising:

4. The training method of claim 3, wherein said generating an augmented article sample data set from said article sample data set comprises:

determining a first item data list similar to the item image data from an item database according to the item image data, wherein the first item data list comprises item image data of different items, text data for describing the items and item style data;

and generating the amplified article sample data set according to the first article data list and the second article data list.

5. The training method of claim 4, wherein said generating the augmented item sample dataset from the first item data list and the second item data list comprises:

and generating the amplified article sample data set according to the data intersection of the first article data list and the second article data list.

6. The training method according to claim 1, wherein the model parameters of the initial model include model parameters of an image feature extraction layer of the initial model, model parameters of a text feature extraction layer of the initial model, and model parameters of a style feature extraction layer of the initial model, and the adjusting the model parameters of the initial model according to the matching result and the style label to obtain the trained feature recognition model includes:

and adjusting the model parameters of the image feature extraction layer of the initial model, the model parameters of the text feature extraction layer of the initial model and the model parameters of the style feature extraction layer of the initial model according to the error between the matching result and the style label to obtain the trained feature recognition model.

7. An article feature identification feature recognition method comprises the following steps:

inputting the article image information into an image feature extraction layer of a feature recognition model, and outputting article image features of the article to be processed, wherein the feature recognition model is obtained by training through the training method of any one of claims 1 to 6;

inputting the text information for describing the article into a text feature extraction layer of the feature recognition model, and outputting the text feature of the article to be processed;

inputting the article image features and the text features into a feature combination layer of the feature recognition model, and outputting article combination features of the article to be processed;

inputting the item combination characteristics and the description characteristics of the candidate style into a matching layer of the characteristic identification model, and outputting a matching result for representing the item combination characteristics and the description characteristics of the candidate style, wherein the description characteristics of the candidate style are obtained by inputting item style information acquired from a candidate style database into a style characteristic extraction layer of the characteristic identification model;

8. The method of claim 7, further comprising:

preprocessing the plurality of item comment text data and the plurality of item title text data to obtain a text data set for representing the style of the item;

and generating the candidate style database according to the text data set for representing the style of the article.

9. The method of claim 8, further comprising:

generating a vector data set for representing the style of the article according to the text data set for representing the style of the article;

and generating the candidate style database according to the vector data set for characterizing the style of the article.

10. A feature recognition model training apparatus comprising:

the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the article sample data set comprises a plurality of article samples, each article sample comprises article image data, text data for describing an article and article style data, and the article sample has a style label;

the feature extraction module is used for inputting the article image data into an image feature extraction layer of an initial model and outputting article image features aiming at each article sample; inputting the text data for describing the article into a text feature extraction layer of the initial model, and outputting text features; inputting the article style data into a style feature extraction layer of the initial model, and outputting style features;

the feature combination module is used for inputting the article image features and the text features into a feature combination layer of the initial model and outputting article combination features;

the matching module is used for inputting the article combination characteristics and the style characteristics into a matching layer of the initial model and outputting a matching result for representing the article combination characteristics and the style characteristics;

and the adjusting module is used for adjusting the model parameters of the initial model according to the matching result and the style label to obtain a trained feature recognition model.

11. An article feature identification device comprising:

the second acquisition module is used for acquiring article information of the article to be processed, wherein the article information comprises article image information and text information used for describing the article;

a first recognition module, configured to input the article image information into an image feature extraction layer of a feature recognition model, and output an article image feature of the article to be processed, where the feature recognition model is obtained by training according to the training method of any one of claims 1 to 6;

the second identification module is used for inputting the text information for describing the article into a text feature extraction layer of the feature identification model and outputting the text feature of the article to be processed;

the third identification module is used for inputting the article image characteristics and the text characteristics into a characteristic combination layer of the characteristic identification model and outputting article combination characteristics of the article to be processed;

a fourth identification module, configured to input the item combination feature and the description feature of the candidate style into a matching layer of the feature identification model, and output a matching result for characterizing the item combination feature and the description feature of the candidate style, where the description feature of the candidate style is obtained by inputting item style information acquired from a candidate style database into a style feature extraction layer of the feature identification model;

and the determining module is used for determining style characteristic information matched with the article combination characteristic of the article to be processed according to the matching result.

12. An electronic device, comprising:

one or more processors;

a storage device for storing one or more programs,

wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method of any of claims 1-6 or 7-9.

13. A computer readable storage medium having stored thereon executable instructions which, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 6 or 7 to 9.

14. A computer program product comprising a computer program which, when executed by a processor, implements a method according to any one of claims 1 to 6 or 7 to 9.