CN113506144B - Clothing sales prediction method and system based on artificial intelligence and big data - Google Patents

Clothing sales prediction method and system based on artificial intelligence and big data Download PDF

Info

Publication number
CN113506144B
CN113506144B CN202111034433.5A CN202111034433A CN113506144B CN 113506144 B CN113506144 B CN 113506144B CN 202111034433 A CN202111034433 A CN 202111034433A CN 113506144 B CN113506144 B CN 113506144B
Authority
CN
China
Prior art keywords
clothing
style
heat
item
single item
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202111034433.5A
Other languages
Chinese (zh)
Other versions
CN113506144A (en
Inventor
郑妙春
李锦莲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NANTONGYOUYUAN ART DESIGN Co.,Ltd.
Original Assignee
Haimen Chuangrui Machinery Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Haimen Chuangrui Machinery Co ltd filed Critical Haimen Chuangrui Machinery Co ltd
Priority to CN202111034433.5A priority Critical patent/CN113506144B/en
Publication of CN113506144A publication Critical patent/CN113506144A/en
Application granted granted Critical
Publication of CN113506144B publication Critical patent/CN113506144B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to the technical field of artificial intelligence and clothing industry, in particular to a clothing sales prediction method and a clothing sales prediction system based on artificial intelligence and big data. The method comprises the steps of processing a clothing matching image through a clothing analysis neural network, obtaining clothing styles and characteristic vectors of each clothing single product, and constructing a style characteristic distribution map through the clothing styles and the characteristic vectors. And acquiring the clothing style heat and the clothing single item heat through the network platform data. And resetting the pixel value of the corresponding position of the style characteristic distribution diagram according to the clothing style heat and the clothing single item heat to obtain a clothing single item heat map of each clothing single item, and calibrating and updating the clothing single item heat map through the clustering center of each style characteristic distribution diagram to obtain a standard clothing single item heat map. And obtaining the predicted clothing sales amount according to the standard clothing single item heat map. The invention eliminates the information difference between shops and a network platform, considers the influence of the collocation relationship between the clothing single products on the sales volume and predicts the accurate clothing sales volume.

Description

Clothing sales prediction method and system based on artificial intelligence and big data
Technical Field
The invention relates to the technical field of artificial intelligence and clothing industry, in particular to a clothing sales prediction method and a clothing sales prediction system based on artificial intelligence and big data.
Background
There are many information-rich network platforms in the internet of modern society. For the clothing industry, information in a network platform has a certain guiding function, such as comments of electric commerce platforms such as Taobao and Jingdong; it is known that postings of community platforms such as the small red book and the like affect the sales volume of clothes.
For the clothing industry, the accurate prediction of the clothing sales volume of different styles can guide the commodity sales volume, reduce the loss of sales opportunities, reduce the stock backlog and improve the clothing sales volume. In the prior art, clothing sales can be predicted by constructing a mathematical model of the relationship between historical clothing sales social media data. However, the method ignores the difference between the clothes with different styles in the shops and the clothes in the social media, so that the error of the finally obtained prediction result is larger. And the influence of sales volume caused by matching relation among the clothing items is not considered, and reliable guide information cannot be provided for merchants.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a clothing sales prediction method and a clothing sales prediction system based on artificial intelligence and big data, and the adopted technical scheme is as follows:
the invention provides a clothing sales prediction method based on artificial intelligence and big data, which comprises the following steps:
obtaining a plurality of clothes matching images containing different clothes single product matches; sending the clothing matching image into a pre-trained clothing analysis neural network to output clothing style and the characteristic vector of each clothing single product; mapping the feature vectors into a two-dimensional space as coordinates, and constructing a style feature distribution map of each clothing single product by taking the corresponding clothing style as an element value;
acquiring clothing information and network platform user quantity in a plurality of network platforms; the clothing information comprises clothing style emotional tendency, clothing single item emotional tendency, clothing style evaluation praise number, clothing single item evaluation praise number and clothing single item sales volume; obtaining clothing style heat according to the clothing style evaluation praise number, the network platform user quantity and the clothing style emotional tendency; obtaining clothing single item popularity according to the single item evaluation praise number, the network platform user quantity, the clothing single item emotional tendency and the clothing single item sales volume;
obtaining a clustering center of the style characteristic distribution map; resetting the pixel value of the corresponding position of the style characteristic distribution diagram according to the clothing single item heat degree and the clothing style heat degree to obtain a clothing single item heat degree diagram of each clothing single item; calibrating and updating all the clothing single item heat maps according to the clustering center to obtain a standard clothing single item heat map;
and obtaining the predicted clothing sales volume according to the standard clothing single item heat map.
Further, the step of sending the clothing matching image into a pre-trained clothing analysis neural network to output the feature vector and the corresponding clothing style of each clothing single product comprises:
the clothing analysis neural network segments the clothing matching image through a first clothing segmentation sub-network to obtain a plurality of clothing single-product images; sending each clothing single-item image into a corresponding clothing analysis encoder to obtain the feature vector of each clothing single-item; and merging the characteristic vectors, sending the merged characteristic vectors into a full connection layer, and outputting the clothing style.
Further, the clothing analysis neural network further includes:
using a plurality of clothes matching images as a training set; obtaining the feature vectors and the clothing styles of a training set; constructing a distance matrix according to the distance between the feature vectors in the training set; constructing a standard matrix according to the difference of the clothing styles in the training set; adjusting a loss value of a loss function by a difference of the distance matrix and the standard matrix; and training the clothing analysis neural network according to the loss function.
Further, the training function includes:
Figure 100002_DEST_PATH_IMAGE002
wherein the content of the first and second substances,
Figure 100002_DEST_PATH_IMAGE004
matching the number of images to the garments in the training set,
Figure 100002_DEST_PATH_IMAGE006
is the number of the individual items of clothing,
Figure 100002_DEST_PATH_IMAGE008
is as follows
Figure 100002_DEST_PATH_IMAGE010
A matrix of the distances is generated from the distance matrix,
Figure 100002_DEST_PATH_IMAGE012
for the said standard matrix,
Figure 100002_DEST_PATH_IMAGE014
is the first in the training set
Figure 100002_DEST_PATH_IMAGE016
The matched image of the clothing belongs to the style of the clothing
Figure 100002_DEST_PATH_IMAGE018
The true probability of (a) of (b),
Figure 100002_DEST_PATH_IMAGE020
in the training set
Figure 430715DEST_PATH_IMAGE016
The matched image of the clothing belongs to the style of the clothing
Figure 232449DEST_PATH_IMAGE018
The probability of prediction of (a) is,
Figure 100002_DEST_PATH_IMAGE022
the number of the clothing style categories.
Further, the obtaining of the clothing style popularity according to the clothing style evaluation praise number, the network platform user quantity and the clothing style emotional tendency comprises:
taking the ratio of the clothing style evaluation praise number in each network platform to the user amount of the network platform as the clothing style evaluation heat; taking the product of the costume style evaluation heat and the costume style emotional tendency as a first initial costume style heat of the costume style on the network platform;
taking the ratio of the user quantity of each network platform to the maximum user quantity of the network platforms in all the network platforms as the quality evaluation value of the network platform; taking the product of the quality evaluation value and the first initial clothing style heat as a second initial clothing style heat; and accumulating the second initial clothing style heat corresponding to all the network platforms to obtain the clothing style heat.
Further, obtaining the popularity of the clothing items according to the evaluation praise number of the items, the user quantity of the network platform and the emotional tendency of the clothing items and the sales volume of the clothing items:
taking the ratio of the clothing item evaluation praise number in each network platform to the user amount of the network platform as the clothing item evaluation heat; taking the product of the evaluation popularity of the clothing items and the emotional tendency of the clothing items as a first initial clothing item popularity of the clothing items on the network platform;
taking the ratio of the clothing single item sales volume to the user volume of the network platform as a sales volume evaluation value; taking the product of the quality evaluation value, the sales evaluation value and the first initial clothing single item heat as a second initial clothing single item heat; and accumulating the second initial clothing item heat corresponding to all the network platforms to obtain the clothing item heat.
Further, the step of resetting the pixel value of the position corresponding to the style characteristic distribution diagram according to the clothing item heat degree and the clothing style heat degree to obtain the clothing item heat degree diagram of each clothing item includes:
setting the pixel value at the corresponding position in the style characteristic distribution map as the product of the clothing single item heat and the clothing style heat; and processing each pixel point by utilizing a Gaussian convolution kernel to obtain the clothing single item heat map.
Further, the updating all the clothing item heat maps according to the cluster center calibration to obtain a standard clothing item heat map comprises:
obtaining the distance relation between the pixel point in each clothing single item heat map and all the clustering centers; obtaining a matching pixel point group according to the difference of the distance relation between different clothing single item heat maps; resetting the pixel points at the corresponding positions in the clothing single-item heat map according to the pixel value average value of the matched pixel point group; and traversing all the pixel points to complete updating, and obtaining a standard clothing single item heat map.
Further, the predicting clothing sales amount according to the standard clothing item heat map comprises:
obtaining the sales relation between future clothing single item sales volume and the corresponding heat value in the standard clothing single item heat map through historical data; sending the image of the commercially-paved and sold clothing single product into the clothing analysis neural network to obtain the characteristic vector of the commercially-sold clothing single product; and obtaining the on-sale heat value of the corresponding position in the standard clothing single item heat map through the on-sale clothing single item feature vector, and obtaining the predicted clothing sales volume according to the sales relationship and the on-sale heat value.
The invention also provides a clothing sales prediction system based on artificial intelligence and big data, which comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, and is characterized in that the processor realizes any step of the clothing sales prediction method based on artificial intelligence and big data when executing the computer program.
The invention has the following beneficial effects:
1. according to the embodiment of the invention, the clothing matching image is analyzed and processed through the clothing analysis neural network, and the clothing style corresponding to the characteristic vector of the clothing single product is obtained. And further obtaining corresponding heat through corresponding clothing information in the network platform. The clothing single item heat degree graph is obtained through combination of the characteristic vectors and the heat degrees, the clothing single item information is represented by the characteristic vectors, and the difference between shop information and a network platform is eliminated, so that the subsequent prediction of clothing sales volume is high in reliability and accuracy.
2. According to the embodiment of the invention, the heat map of the single clothing is constructed through the network platform data and the image information of the clothing matching image. The single clothing item heat map represents the sales heat of the current single clothing item, the standard single clothing item heat map after calibration and updating considers the matching situation of the single clothing item and combines the heat of different single clothing items, so that the standard single clothing item heat map can comprehensively represent the sales heat of the single clothing item.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a method for predicting clothing sales based on artificial intelligence and big data according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a clothing analysis neural network according to an embodiment of the present invention.
Detailed Description
In order to further illustrate the technical means and effects of the present invention adopted to achieve the predetermined invention purpose, the following detailed description will be given to a clothes sales forecasting method and system based on artificial intelligence and big data according to the present invention, and the detailed implementation, structure, features and effects thereof with reference to the accompanying drawings and the preferred embodiments. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The following describes a specific scheme of a clothing sales prediction method and system based on artificial intelligence and big data in detail with reference to the accompanying drawings.
Referring to fig. 1, a flow chart of a garment sales prediction method based on artificial intelligence and big data according to an embodiment of the present invention is shown, where the method includes:
step S1: obtaining a plurality of clothes matching images containing different clothes single product matches; sending the clothing matching images into a pre-trained clothing analysis neural network to output clothing styles and characteristic vectors of each clothing single product; and mapping the characteristic vectors into a two-dimensional space as coordinates, and constructing a style characteristic distribution map of each clothing single product by taking the corresponding clothing style as an element value.
And obtaining a plurality of clothes matching images containing different clothes single product matches from a clothes database consisting of big data. The matching of different clothes can form different clothes styles, such as Japanese, Korean, Chinese and the like. Therefore, each clothing matching image corresponds to a clothing style. And (4) sending the clothing matching images into a pre-trained clothing analysis neural network to output the clothing style and the characteristic vector of each clothing single product. In the embodiment of the invention, three clothes items, namely upper clothes, lower clothes and shoes are used for analysis. Referring to fig. 2, a schematic diagram of a clothing analysis neural network structure provided in an embodiment of the present invention is shown, where the clothing analysis neural network specifically includes:
1) and forming a training set by matching a plurality of clothes with the images. The labels of the training set are divided into two categories: apparel style labels and labels between different items of apparel.
2) The clothing analysis neural network segments the clothing matching image through the first clothing segmentation sub-network to obtain a plurality of clothing single-product images. The first clothing segmentation sub-network adopts a coding-decoding structure, a training set is input into a clothing segmentation encoder 101 to extract features, and a clothing single segmentation graph is output through a clothing segmentation decoder 102. And multiplying the clothes matching images by using the clothes single item segmentation image as a shade to obtain the upper garment, the lower garment and the corresponding clothes single item images. In the embodiment of the present invention, the first assembly segmentation subnetwork uses a central Mask network (central Mask), and in other embodiments, other networks may be selected according to specific situations.
3) And (4) sending the three clothing single-item images into corresponding three clothing analysis encoders to obtain the characteristic vector of each clothing single-item. The garment analysis encoder comprises a first garment analysis encoder 103, a second analysis encoder 104 and a third analysis encoder 105, corresponding to the upper garment, the lower garment and the trousers, respectively. The three garment analysis encoders are distributed in parallel and have the same structure. Three garment analysis encoders map corresponding garment images to three different feature spaces. The different feature spaces can guarantee the accuracy of subsequent garment style classification. And (4) merging the three characteristic vectors through the channels, sending the three characteristic vectors into the full-connection layer 106, and outputting the clothing style. In the embodiment of the present invention, the feature vector has a size of two rows and one column.
4) In order to ensure that the feature vectors meeting the same clothing style in the feature spaces of clothing single products in different areas are close and the feature vectors in different clothing styles are far apart, the network classification is accurate, and the loss value of the network needs to be adjusted in the training process. Setting batch size during training
Figure 843559DEST_PATH_IMAGE004
I.e. each time of input
Figure 56234DEST_PATH_IMAGE004
The matched images of the piece of clothing are used as a group of training groups. In the embodiments of the present invention
Figure 588847DEST_PATH_IMAGE004
Is 48. And obtaining the characteristic vector and the clothing style of each clothing matching image in the training set after network processing. Namely, three feature vectors and one clothing style can be obtained from each clothing matching image. Constructing a distance matrix according to the distance between the characteristic vectors in the training set, wherein the distance matrix comprises
Figure DEST_PATH_IMAGE024
Line of
Figure DEST_PATH_IMAGE026
The value of the column indicates the second
Figure 725430DEST_PATH_IMAGE024
Matching the image and the second image with the piece of clothing
Figure 659888DEST_PATH_IMAGE026
The piece of clothing matches the distance between the feature vectors of the images. And constructing a standard matrix according to the clothing style difference in the training set. In the embodiment of the present invention, if
Figure 810509DEST_PATH_IMAGE024
Matching the image and the second image with the piece of clothing
Figure 830417DEST_PATH_IMAGE026
The styles of the clothes matched with the images are the same, then the standard matrix is
Figure 98588DEST_PATH_IMAGE024
Line of
Figure 762918DEST_PATH_IMAGE026
The value of the column is 0; otherwise, it is 5. I.e. three distance matrices and one criterion matrix are obtained. The loss value of the loss function is adjusted by the difference of the distance matrix and the standard matrix. And continuously and iteratively updating parameters of the clothing analysis neural network by using a gradient descent method according to the loss function. The loss function comprises two parts, wherein the first part is distance measurement loss, the fact that the characteristic vectors of the single clothes with the same clothes style are close, the characteristic vectors with different clothes styles are far apart is guaranteed, and a distance measurement loss item is fitted by a mathematical modeling method and introduced into a network; the second part is the classification loss commonly used in the existing loss function, and the garment style classification accuracy is guaranteed. The loss function specifically includes:
Figure DEST_PATH_IMAGE028
wherein the content of the first and second substances,
Figure 379713DEST_PATH_IMAGE004
in the case of a batch size,
Figure 886918DEST_PATH_IMAGE008
is as follows
Figure 427621DEST_PATH_IMAGE010
A matrix of distances is formed by the distance matrix,
Figure 477616DEST_PATH_IMAGE012
is a standard matrix, and is characterized in that,
Figure 78362DEST_PATH_IMAGE014
is the first in the training set
Figure 72863DEST_PATH_IMAGE016
The matched image of a piece of clothing belongs to the style of clothing
Figure 417257DEST_PATH_IMAGE018
The true probability of (a) of (b),
Figure 69561DEST_PATH_IMAGE020
in the training set
Figure 106788DEST_PATH_IMAGE016
The matched image of a piece of clothing belongs to the style of clothing
Figure 323005DEST_PATH_IMAGE018
The prediction probability of (2).
Figure 346456DEST_PATH_IMAGE022
The number of garment style categories.
And mapping the characteristic vectors into a two-dimensional space as coordinates, and constructing a style characteristic distribution map of each clothing single product by taking the corresponding clothing style as an element value. Three style characteristic distribution maps corresponding to the upper garment, the lower garment and the shoes are obtained.
Step S2: obtaining clothing information in a plurality of network platforms; the clothing information comprises clothing style emotional tendency, clothing single item emotional tendency, clothing style evaluation praise number, clothing single item evaluation praise number and clothing single item sales volume; obtaining clothing style heat according to the clothing style evaluation praise number and the clothing style emotional tendency; acquiring clothing style heat according to the user quantity and the clothing style heat of the network platform; obtaining the popularity of the clothing single item according to the single item evaluation praise number and the emotional tendency of the clothing single item; and obtaining the heat of the single clothes according to the user quantity, the sales volume of the single clothes and the heat of the single clothes of the network platform.
In the internet era, information guidance of network platforms can affect purchasing trends of customers. The embodiment of the invention obtains the clothing information in the plurality of network platforms through the big data technology and provides guidance information for the prediction of subsequent clothing sales volume. The clothing information comprises clothing style emotional tendency, clothing single item emotional tendency, clothing style evaluation praise number, clothing single item evaluation praise number and clothing single item sales volume. The clothing style in the clothing information matches the clothing style obtained from the image information in step S1.
In the embodiment of the invention, the emotional tendency of each evaluation in the latest month of the clothing style in the network platform is obtained through the text emotion analysis technology of the NLP, and the second evaluation is taken
Figure DEST_PATH_IMAGE030
The emotional tendency of each comment is
Figure DEST_PATH_IMAGE032
Each emotional tendency is divided into positive, neutral and negative, which are quantized to 1, 0 and-1, respectively. It should be noted that the text emotion analysis technology can be implemented by two methods, namely a neural network-based method and a corpus-based method, and in other embodiments, a suitable text emotion analysis technology can be freely selected.
The evaluation praise in the network platform represents the approval of the user to the evaluation, so the ratio of the clothing style evaluation praise in each network platform to the user quantity of the network platform is taken as the clothing style evaluation heat, and the second step is taken as the evaluation heat of the clothing style
Figure 885891DEST_PATH_IMAGE030
The clothes style of the individual evaluation is evaluated to have a heat degree of
Figure DEST_PATH_IMAGE034
. Combining all the clothing evaluation heat and emotion analysis results, the user tendency and the user heat of the clothing style on a network platform can be shown, so that the product of the clothing style evaluation heat and the clothing style emotion tendency is used as the first initial clothing style heat of the clothing style on the network platform, and the second initial clothing style heat is recorded
Figure 625177DEST_PATH_IMAGE010
The clothes style is
Figure 204057DEST_PATH_IMAGE016
The first initial clothing style heat on the individual network platform is
Figure DEST_PATH_IMAGE036
. Namely, it is
Figure DEST_PATH_IMAGE038
Wherein
Figure DEST_PATH_IMAGE040
Is shown as
Figure 578668DEST_PATH_IMAGE010
The clothes style is
Figure 582396DEST_PATH_IMAGE016
The set of all evaluations on each network platform.
Because the scales of the network platforms are different, the evaluation quality is also different, and therefore the ratio of the user quantity of each network platform to the maximum network platform user quantity in all the network platforms is used as the quality evaluation value of the network platform. The quality assessment value is used to indicate the quality scale of a network platform. And taking the quality evaluation value as the weight of the first initial clothing style heat, and taking the product of the quality evaluation value and the first initial clothing style heat as the second initial clothing style heat. One clothing style corresponds to one second initial clothing style heat in each network platform. And accumulating the second initial clothing style heat corresponding to all the network platforms to obtain the clothing style heat.
The sales volume of the clothing single item can more intuitively show the heat of the clothing single item, so that the method is similar to the clothing style heat obtaining method, and the method for obtaining the heat of the clothing single item by considering the sales volume of the clothing single item specifically comprises the following steps:
and taking the ratio of the evaluation praise number of the clothing single in each network platform to the user quantity of the network platform as the evaluation popularity of the clothing single. And taking the product of the evaluation heat of the clothing single product and the emotional tendency of the clothing single product as the first initial clothing single product heat of the clothing single product on the network platform. And taking the ratio of the clothing single product sales volume to the user volume of the network platform as a sales volume evaluation value. And taking the quality evaluation value and the pin quantity evaluation value as weights of the first initial clothing item heat degree, and taking the product of the quality evaluation value, the pin quantity evaluation value and the first initial clothing item heat degree as a second initial clothing item heat degree. And accumulating the second initial clothing item heat corresponding to all the network platforms to obtain the clothing item heat.
In the embodiment of the invention, the value ranges of the clothing style heat and the clothing single item heat are normalized to
Figure DEST_PATH_IMAGE042
Closer to 1 indicates higher heat.
Step S3: obtaining a clustering center of the style characteristic distribution diagram; resetting the pixel value of the corresponding position of the style characteristic distribution diagram according to the heat degree of the clothing single item and the heat degree of the clothing style, and obtaining a heat degree diagram of the clothing single item of each clothing single item; and updating all the clothing single item heat maps according to the calibration of the clustering center to obtain a standard clothing single item heat map.
And introducing the clothing style heat and the clothing item heat obtained in the step S2 into the style characteristic distribution map, and setting the pixel value at the corresponding position in the style characteristic distribution map as the product of the clothing item heat and the clothing style heat. And processing each pixel point by using a Gaussian convolution kernel to obtain a clothing single item heat map. The position of the clothing single item is represented by the characteristic vector in the clothing single item heat map, so that the difference between shop information and network platform information is eliminated.
According to the prior knowledge, the sales volume of the clothing items with high fitting degree can influence each other, for example, when the sales volume of an upper garment is larger, the sales volume of a lower garment and shoes which are fitted with the upper garment with the highest fitting degree can be promoted. Therefore, the clothing single item heat degree map is updated according to the heat degrees of different clothing single items, so that the accurate heat degrees of different types of clothing can be obtained, and an accurate sales prediction result can be obtained.
And the pixel values in the style characteristic distribution diagram represent the clothing style, the style characteristic distribution diagram is clustered according to the pixel values, the number of the clustering clusters is set as the clothing style category number, the distance formula is the difference value of the pixel values, and each style characteristic distribution diagram obtains a plurality of clustering centers. Each cluster center corresponds to a clothing style category. Because the feature spaces of the feature vectors corresponding to different clothing items are different, the clustering centers of the feature distribution maps of different styles are different, and in order to correctly represent the collocation correlation among different clothing items, the clothing item heat map needs to be calibrated and updated according to the position relationship of the clustering centers, which specifically comprises the following steps:
the positions of the pixel points in the clothing single item heat map represent a clothing single item. And obtaining the distance relation between the pixel point in each clothing single item heat map and all the clustering centers. Obtaining a matching pixel point group according to the difference of distance relations between different clothing single item heat maps; resetting the pixel points at the corresponding positions in the clothing single item heat map according to the pixel value mean value of the matched pixel point group; and traversing all the pixel points to complete updating, and obtaining a standard clothing single item heat map.
In the embodiment of the present invention, the distance relationship is expressed by euclidean distance, which specifically includes:
Figure DEST_PATH_IMAGE044
wherein, in the step (A),
Figure DEST_PATH_IMAGE046
is a single item of clothing
Figure 23742DEST_PATH_IMAGE006
The distance relationship in the heat map of the item of apparel of type 1,
Figure 214552DEST_PATH_IMAGE018
in order to be of the kind of the style of clothes,
Figure 970018DEST_PATH_IMAGE022
the number of the garment style categories is,
Figure DEST_PATH_IMAGE048
in the style of clothes
Figure 438040DEST_PATH_IMAGE018
In the first place
Figure DEST_PATH_IMAGE050
A cluster center point in the heat map of the single garment,
Figure DEST_PATH_IMAGE052
is a single item of clothing
Figure 407877DEST_PATH_IMAGE006
With the style of clothes
Figure 351562DEST_PATH_IMAGE018
In the first place
Figure DEST_PATH_IMAGE054
And the Euclidean distance of the clustering center points in the clothing single item heat map.
In an embodiment of the invention, an objective function is constructed
Figure DEST_PATH_IMAGE056
Obtaining a set of matched pixels, wherein
Figure DEST_PATH_IMAGE058
Is a single item of clothing
Figure DEST_PATH_IMAGE060
In the first place
Figure DEST_PATH_IMAGE062
Distance relation in a heat map of an individual garment,
Figure DEST_PATH_IMAGE064
is a single item of clothing
Figure DEST_PATH_IMAGE066
In the first place
Figure DEST_PATH_IMAGE068
Distance relation in a heat map of a garment single item. And the target function searches matching pixel points in the other two clothing item heat maps to form a matching pixel point group through the difference of the distance relation between the pixel points in each clothing item heat map and the clustering center.Averaging the pixel values in the Gaussian hot spot range corresponding to the matched pixel group, and resetting the first step by using the average value
Figure 176298DEST_PATH_IMAGE062
Seed and second
Figure 623460DEST_PATH_IMAGE068
Traversing pixel values in the heat map of the clothing single item
Figure 970390DEST_PATH_IMAGE054
And updating all pixel points in the heat maps of the two kinds of clothing single products to the heat maps of the other two kinds of clothing single products. Respectively using the second method
Figure 401372DEST_PATH_IMAGE062
Heat map of single garment and its second
Figure 108428DEST_PATH_IMAGE068
And the heat map of the single garment type is used for updating the distribution maps of the other two single garment types.
Step S4: and obtaining the predicted clothing sales amount according to the standard clothing single item heat map.
Obtaining the sales relation between the future clothing single item sales volume and the corresponding heat value in the standard clothing single item heat map through historical data; sending the image of the commercially-paved single clothing item into a clothing analysis neural network to obtain a characteristic vector of the single clothing item on sale; and obtaining the on-sale heat value of the corresponding position in the standard clothing single item heat map through the on-sale clothing single item feature vector, and obtaining the predicted clothing sales volume according to the sales relation and the on-sale heat value.
In the embodiment of the invention, the sales relationship is fitted through the clothing heat map of a reference month in historical data and the clothing sales volume of a reference month in the next month, the sales relationship is fitted through a mathematical modeling method, and the least square method is adopted for fitting, and the method specifically comprises the following steps:
Figure DEST_PATH_IMAGE070
wherein, in the step (A),
Figure DEST_PATH_IMAGE072
in order to predict the amount of sales of the garment,
Figure 597047DEST_PATH_IMAGE030
is the heat value in the clothing heat map,
Figure DEST_PATH_IMAGE074
、
Figure DEST_PATH_IMAGE076
in order to fit the parameters to the image,
Figure DEST_PATH_IMAGE078
。
in summary, in the embodiments of the present invention, the clothing matching image is processed through the clothing analysis neural network, the clothing style and the feature vector of each clothing item are obtained, and the style feature distribution map is constructed through the clothing style and the feature vector. And acquiring the clothing style heat and the clothing single item heat through the network platform data. And resetting the pixel value of the corresponding position of the style characteristic distribution diagram according to the clothing style heat and the clothing single item heat to obtain a clothing single item heat map of each clothing single item, and calibrating and updating the clothing single item heat map through the clustering center of each style characteristic distribution diagram to obtain a standard clothing single item heat map. And obtaining the predicted clothing sales amount according to the standard clothing single item heat map.
The invention also provides a clothing sales prediction system based on artificial intelligence and big data, which comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, and is characterized in that the processor realizes any step of the clothing sales prediction method based on artificial intelligence and big data when executing the computer program.
It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (10)

1. A clothing sales prediction method based on artificial intelligence and big data is characterized by comprising the following steps:
obtaining a plurality of clothes matching images containing different clothes single product matches; sending the clothing matching image into a pre-trained clothing analysis neural network to output clothing style and the characteristic vector of each clothing single product; mapping the feature vectors into a two-dimensional space as coordinates, and constructing a style feature distribution map of each clothing single product by taking the corresponding clothing style as an element value;
acquiring clothing information and network platform user quantity in a plurality of network platforms; the clothing information comprises clothing style emotional tendency, clothing single item emotional tendency, clothing style evaluation praise number, clothing single item evaluation praise number and clothing single item sales volume; obtaining clothing style heat according to the clothing style evaluation praise number, the network platform user quantity and the clothing style emotional tendency; obtaining clothing single item popularity according to the single item evaluation praise number, the network platform user quantity, the clothing single item emotional tendency and the clothing single item sales volume;
clustering the style characteristic distribution map according to the pixel values to obtain a clustering center of the style characteristic distribution map; resetting the pixel value of the corresponding position of the style characteristic distribution diagram according to the clothing single item heat degree and the clothing style heat degree to obtain a clothing single item heat degree diagram of each clothing single item; calibrating and updating all the clothing single item heat maps according to the clustering center to obtain a standard clothing single item heat map;
and obtaining the predicted clothing sales volume according to the standard clothing single item heat map.
2. The method of claim 1, wherein the step of sending the clothing matching images into a pre-trained clothing analysis neural network to output the feature vector and the corresponding clothing style of each clothing item comprises:
the clothing analysis neural network segments the clothing matching image through a first clothing segmentation sub-network to obtain a plurality of clothing single-product images; sending each clothing single-item image into a corresponding clothing analysis encoder to obtain the feature vector of each clothing single-item; and merging the characteristic vectors, sending the merged characteristic vectors into a full connection layer, and outputting the clothing style.
3. The artificial intelligence and big data based clothing sales prediction method of claim 2, wherein the clothing analysis neural network further comprises:
using a plurality of clothes matching images as a training set; obtaining the feature vectors and the clothing styles of a training set; constructing a distance matrix according to the distance between the feature vectors in the training set; constructing a standard matrix according to the difference of the clothing styles in the training set; adjusting a loss value of a loss function by a difference of the distance matrix and the standard matrix; and training the clothing analysis neural network according to the loss function.
4. The artificial intelligence and big data based clothing sales prediction method of claim 3, wherein the loss function comprises:
Figure DEST_PATH_IMAGE002
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE004
matching the number of images to the garments in the training set,
Figure DEST_PATH_IMAGE006
is the number of the individual items of clothing,
Figure DEST_PATH_IMAGE008
is as follows
Figure DEST_PATH_IMAGE010
A matrix of the distances is generated from the distance matrix,
Figure DEST_PATH_IMAGE012
for the said standard matrix,
Figure DEST_PATH_IMAGE014
is the first in the training set
Figure DEST_PATH_IMAGE016
The matched image of the clothing belongs to the style of the clothing
Figure DEST_PATH_IMAGE018
The true probability of (a) of (b),
Figure DEST_PATH_IMAGE020
in the training set
Figure 103634DEST_PATH_IMAGE016
The matched image of the clothing belongs to the style of the clothing
Figure 833824DEST_PATH_IMAGE018
The probability of prediction of (a) is,
Figure DEST_PATH_IMAGE022
the number of the clothing style categories.
5. The artificial intelligence and big data based clothing sales prediction method of claim 1, wherein the obtaining of clothing style enthusiasm according to the clothing style evaluation praise number, the network platform user amount and the clothing style emotional tendency comprises:
taking the ratio of the clothing style evaluation praise number in each network platform to the user amount of the network platform as the clothing style evaluation heat; taking the product of the costume style evaluation heat and the costume style emotional tendency as a first initial costume style heat of the costume style on the network platform;
taking the ratio of the user quantity of each network platform to the maximum user quantity of the network platforms in all the network platforms as the quality evaluation value of the network platform; taking the product of the quality evaluation value and the first initial clothing style heat as a second initial clothing style heat; and accumulating the second initial clothing style heat corresponding to all the network platforms to obtain the clothing style heat.
6. The artificial intelligence and big data based clothing sales prediction method of claim 5, wherein the clothing item popularity is obtained according to the item rating praise number, the network platform user quantity, the clothing item emotional tendency and the clothing item sales quantity:
taking the ratio of the clothing item evaluation praise number in each network platform to the user amount of the network platform as the clothing item evaluation heat; taking the product of the evaluation popularity of the clothing items and the emotional tendency of the clothing items as a first initial clothing item popularity of the clothing items on the network platform;
taking the ratio of the clothing single item sales volume to the user volume of the network platform as a sales volume evaluation value; taking the product of the quality evaluation value, the sales evaluation value and the first initial clothing single item heat as a second initial clothing single item heat; and accumulating the second initial clothing item heat corresponding to all the network platforms to obtain the clothing item heat.
7. The method for predicting clothing sales based on artificial intelligence and big data according to claim 1, wherein the step of resetting the pixel values of the positions corresponding to the style characteristic distribution map according to the clothing item popularity and the clothing style popularity to obtain the clothing item popularity map of each clothing item comprises the steps of:
setting the pixel value at the corresponding position in the style characteristic distribution map as the product of the clothing single item heat and the clothing style heat; and processing each pixel point by utilizing a Gaussian convolution kernel to obtain the clothing single item heat map.
8. The artificial intelligence and big data based clothing sales prediction method of claim 1, wherein the updating all the clothing item heat maps according to the cluster center calibration to obtain the standard clothing item heat map comprises:
obtaining the distance relation between the pixel point in each clothing single item heat map and all the clustering centers; obtaining a matching pixel point group according to the difference of the distance relation between different clothing single item heat maps; resetting the pixel points at the corresponding positions in the clothing single-item heat map according to the pixel value average value of the matched pixel point group; and traversing all the pixel points to complete updating, and obtaining a standard clothing single item heat map.
9. The artificial intelligence and big data based clothing sales prediction method of claim 1, wherein the predicting clothing sales amount according to the standard clothing item heat map comprises:
obtaining the sales relation between future clothing single item sales volume and the corresponding heat value in the standard clothing single item heat map through historical data; sending the image of the commercially-paved and sold clothing single product into the clothing analysis neural network to obtain the characteristic vector of the commercially-sold clothing single product; and obtaining the on-sale heat value of the corresponding position in the standard clothing single item heat map through the on-sale clothing single item feature vector, and obtaining the predicted clothing sales volume according to the sales relationship and the on-sale heat value.
10. A clothing sales prediction system based on artificial intelligence and big data, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the steps of the method according to any one of claims 1 to 9.
CN202111034433.5A 2021-09-03 2021-09-03 Clothing sales prediction method and system based on artificial intelligence and big data Active CN113506144B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111034433.5A CN113506144B (en) 2021-09-03 2021-09-03 Clothing sales prediction method and system based on artificial intelligence and big data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111034433.5A CN113506144B (en) 2021-09-03 2021-09-03 Clothing sales prediction method and system based on artificial intelligence and big data

Publications (2)

Publication Number Publication Date
CN113506144A CN113506144A (en) 2021-10-15
CN113506144B true CN113506144B (en) 2021-12-03

Family

ID=78016263

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111034433.5A Active CN113506144B (en) 2021-09-03 2021-09-03 Clothing sales prediction method and system based on artificial intelligence and big data

Country Status (1)

Country Link
CN (1) CN113506144B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112598486B (en) * 2021-01-07 2023-08-11 开封大学 Marketing accurate screening push system based on big data and intelligent internet of things

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7958047B2 (en) * 2005-02-04 2011-06-07 The Invention Science Fund I Virtual credit in simulated environments
CN109034255A (en) * 2018-08-02 2018-12-18 深圳码隆科技有限公司 The prediction technique and device of clothes sales volume
CN110110181B (en) * 2019-05-09 2020-12-11 湖南大学 Clothing matching recommendation method based on user style and scene preference

Also Published As

Publication number Publication date
CN113506144A (en) 2021-10-15

Similar Documents

Publication Publication Date Title
US11823059B2 (en) Generating a personalized preference ranking network for providing visually-aware item recommendations
CN104504055B (en) The similar computational methods of commodity and commercial product recommending system based on image similarity
US20190073335A1 (en) Using artificial intelligence to determine a size fit prediction
CN106846122B (en) Commodity data processing method and device
WO2019233077A1 (en) Ranking of business object
CN108694606B (en) Client portrait and service pushing method based on association rule
US12002052B2 (en) Feature subspace isolation and disentanglement in merchant embeddings
CN111028029B (en) Off-line commodity recommendation method and device and electronic equipment
CN111274472A (en) Information recommendation method and device, server and readable storage medium
CN113506144B (en) Clothing sales prediction method and system based on artificial intelligence and big data
CN112115358A (en) Personalized recommendation method using multi-hop path features in knowledge graph
Aziz Customer Segmentation basedon Behavioural Data in E-marketplace
US11062377B1 (en) Fit prediction
CN114741603A (en) Mixed collaborative filtering recommendation algorithm based on user clustering and commodity clustering
CN115456707A (en) Method and device for providing commodity recommendation information and electronic equipment
CN110555719B (en) Commodity click rate prediction method based on deep learning
CN116777528A (en) Commodity information recommendation method and device, computer equipment and storage medium
Bhargavi et al. Comparative study of consumer purchasing and decision pattern analysis using pincer search based data mining method
CN111639989B (en) Commodity recommendation method and readable storage medium
CN112650935B (en) Intelligent information processing method applied to e-commerce cloud service environment and cloud server
CN114219522A (en) Customer consumption behavior prediction method and device, electronic equipment and storage medium
Shukla et al. Total revenue prediction of a sports management application: Grook using machine learning models
CN115905887B (en) Big data intelligent decision system based on data processing
Sudarsan et al. E-commerce Website with Image Search and Price Prediction
CN112307298A (en) Method and device for generating personal brand label

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20220310

Address after: 226100 Fuxing Road, Yuet Lai Town, Haimen, Nantong, Jiangsu, 20

Patentee after: NANTONGYOUYUAN ART DESIGN Co.,Ltd.

Address before: 226131 group 10, Yuelai village, Yuelai Town, Haimen City, Nantong City, Jiangsu Province

Patentee before: Haimen chuangrui Machinery Co.,Ltd.