E-commerce platform commodity recommendation system based on deep learning

Abstract

The invention discloses an e-commerce platform commodity recommendation system based on deep learning, and belongs to the technical field of e-commerce, and the system comprises a commodity consumption data collection module which is used for collecting historical clothes consumption data of a preset user on an e-commerce platform, and analyzing the characteristics of the clothes consumption data of the preset user; the browsing data analysis module responds to the historical clothes consumption data acquired by the commodity consumption data acquisition module and is used for acquiring clothes browsed by a preset user before corresponding clothes consumption and analyzing clothes characteristics of the clothes; according to the method, the deep learning technology is utilized, the consumption records of the preset user on the platform in different time periods are collected, the commodity and clothes characteristics corresponding to the consumption records are obtained, the favorite commodity and clothes can be more accurately recommended to the preset user in the future time period, and the click rate of the preset user on each commodity and clothes is collected, so that the user experience is improved. Commodity favored by a preset user can be better locked, so that consumption transaction is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and in particular to an e-commerce platform product recommendation system based on deep learning. Background Art

[0002] In today's e-commerce environment, personalized product recommendation systems have become a key tool for improving user experience and increasing sales. They are an important part of e-commerce platforms. They analyze users' purchase history, favorite products, browsing history and other data to provide users with personalized and accurate recommended products, improve user shopping experience, increase purchase conversion rate and platform revenue. With the development and popularization of deep learning technology, more and more e-commerce companies are beginning to adopt product recommendation systems based on deep learning.

[0003] Under the existing technology, the product recommendation system can be divided into user-based collaborative filtering algorithms, item-based collaborative filtering algorithms, and content-based recommendation algorithms;

[0004] The user-based collaborative filtering algorithm analyzes the user's historical behavior data, establishes a user similarity model, and recommends products purchased by other users who are highly similar to the user;

[0005] The collaborative filtering algorithm based on items builds a product similarity model by analyzing the relevant attributes of the products and the user's purchase history, and recommends other products that are highly similar to the products purchased by the user in the past.

[0006] Content-based recommendation algorithms recommend products that meet users' interests by analyzing the attributes of products and users' interests and preferences.

[0007] The recommendation method of existing e-commerce platforms is still to intelligently recommend products based on users' historical behaviors and preferences. Although this can improve shopping experience and user stickiness, there is still the problem of inaccurate recommendation results, which sometimes causes users to feel troubled. In addition, over-reliance on recommendation systems can also cause users to be sold unnecessary products and even disrupt shopping decisions. Summary of the invention

[0008] In view of the above problems existing in the existing electronic commerce technology field, the present invention is proposed.

[0009] Therefore, one of the objects of the present invention is to provide an e-commerce platform product recommendation system based on deep learning, which utilizes deep learning technology to collect consumption records of preset users on the platform at different time periods, and by obtaining product and clothing features corresponding to the consumption records, can more accurately recommend desired products and clothing to preset users in future time periods, and, based on the product and clothing browsing records at different time periods in the past, by collecting the click rate of each product and clothing by preset users, can better lock in the products desired by preset users, which is beneficial to facilitating consumer transactions.

[0010] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0011] The e-commerce platform product recommendation system based on deep learning includes:

[0012] The commodity consumption data collection module is used to collect the historical clothing consumption data of the preset user on the e-commerce platform and analyze the characteristics of the clothing consumption data of the preset user;

[0013] A browsing data analysis module, which responds to the historical clothing consumption data collected by the commodity consumption data collection module, and is used to obtain the clothing browsed by the preset user before the corresponding clothing consumption, and analyze its clothing characteristics;

[0014] A fusion data analysis unit, configured to analyze characteristic differences of corresponding clothing according to the return data and product evaluation information corresponding to the preset user's historical clothing consumption data, wherein the characteristic differences include clothing pattern cutting differences, clothing style differences, and clothing color differences, and the fusion data analysis unit includes a calculation module, a determination module, and a time acquisition module;

[0015] The calculation module calculates a return rate based on the return data corresponding to the preset user's historical clothing consumption data and the product evaluation feedback information, and calculates the preset user's historical consumption times corresponding to different clothing characteristics based on the return rate;

[0016] The determination module is based on the preset user historical consumption times corresponding to different clothing features calculated by the calculation module, and is used to preset a critical threshold value based on the return rate accounting for 20% of the corresponding historical consumption times. When the return rate corresponding to a certain clothing feature exceeds the critical threshold value, the system determines that the return rate of the clothing corresponding to the clothing feature is high, and prohibits recommending the clothing corresponding to the clothing feature to the preset user; otherwise, no determination is made;

[0017] The time collection module is used to collect the time when the preset user browses the corresponding clothes under different clothing characteristics, and preset a minimum browsing time based on the start and end time of the browsing. When the browsing time of the preset user on the clothes under the corresponding clothing characteristics is less than the minimum browsing time, the system determines that the preset user has low interest in the clothes and prohibits recommendation. Otherwise, no determination is made;

[0018] An image color comparison module is used to obtain the color data of the product when it is displayed on the platform, and compare the color of the physical clothing based on the color data, and perform color difference analysis, and at the same time obtain the clothing complementary color and dimming data corresponding to the color data according to the color difference analysis results, and classify the color difference analysis results;

[0019] A result data training module, which responds to the color difference analysis result of the image color comparison module and is used to train and adjust the clothing color complement and dimming data corresponding to the color difference analysis result to obtain characteristic changes in the color difference analysis result.

[0020] As a preferred solution of the present invention, in the commodity consumption data collection module, the preset user consumption data characteristics include the consumption data of the preset user in different time periods, the different time periods include the summer period, the spring period and the winter period, and the characteristics of clothing consumed in different time periods, the clothing characteristics include material quality, sewing technology, version, style, color and fabric characteristics.

[0021] As a preferred solution of the present invention, in the browsing data analysis module, the browsed clothing features are divided into ▽ 1 ,▽ 2 , ..., ▽ n , where n represents the nth clothing feature, and the final consumption records of the preset user under different clothing features are obtained. When the preset user browses clothing on the e-commerce platform in the future, the clothing corresponding to the consumption record is recommended to the preset user, and the clothing not corresponding to the consumption record is not recommended.

[0022] As a preferred solution of the present invention, in the image color comparison module, the color difference analysis result is graded into 1 to 5 levels, wherein 1 level indicates the maximum color difference, 5 level indicates the minimum color difference, and this level is regarded as no color difference. The specific grade standards are as follows:

[0023] Level 1: Color difference requirement is 13.6, tolerance requirement is ±1.0;

[0024] Level 2: Color difference requirement is 9.6, tolerance requirement is ±0.7;

[0025] Level 3: Color difference requirement is 6.8, tolerance requirement is ±0.6;

[0026] Level 4: Color difference requirement is 4.8, tolerance requirement is ±0.5;

[0027] Level 5: The color difference requirement is 0, and the tolerance requirement is ±0.22;

[0028] When the color difference analysis result level is level 3, the system determines that the color difference is large, and prohibits recommending products with clothing color complementation and dimming data corresponding to the result to the preset user in the future period. Otherwise, no determination is made;

[0029] The color data of the clothing includes color mode, hue, tone and saturation, wherein the color mode includes RGB and CMYK;

[0030] The dimming data includes illuminance, color temperature and lighting distribution.

[0031] As a preferred embodiment of the present invention, in which: in the result data training module, the adjustment training method includes reducing the complementary color and dimming data values ​​of the clothing corresponding to the color difference analysis results. Specifically, the color difference analysis results are collected each time a complementary color and dimming value is reduced until the collected color difference analysis results are consistent with the color of the actual clothing, and the total reduction value of the complementary color and dimming values ​​under the original values ​​is counted, and at the same time, based on the total reduction value, the platform history is collected to recommend products with complementary color and dimming values ​​greater than the total reduction value to the preset user, and the product is determined to be a product with a large color difference from the actual clothing, and the clothing characteristics of the product are analyzed, and it is prohibited to recommend products with the same clothing characteristics as the product to the preset user in the future time period.

[0032] As a preferred solution of the present invention, the total reduction value is divided into at least 3 groups of reduction intervals in which the complementary color and dimming values ​​differ by two values, and the browsing records and consumption records of the preset user are collected within different reduction intervals, and the complementary color and dimming data of the physical clothing purchased by the preset user are obtained based on the consumption records, and the reduction intervals corresponding to the complementary color and dimming data are obtained, and the clothing products that have not been returned by the preset user are collected based on the reduction intervals, the clothing characteristics of the products are analyzed, and products with the same clothing characteristics are recommended to the preset user in the future time period.

[0033] As a preferred solution of the present invention, the product click data in the browsing data of the preset user in different time periods is collected, and the start and end times of the preset user's previous browsing in the corresponding time period are counted. Within the start and end time, the clicked products and the number of clicks of the preset user are collected, the clothing characteristics of the clicked products are analyzed, a click rate prediction model is constructed, and the interest preference of the preset user to the products is calculated by the collaborative filtering algorithm, which is calculated according to the following formula:

[0034] Among them, R ui represents the i-th clicked clothing product during the u-th browsing of the preset user in the corresponding time period;

[0035] Where, T represents the browsing time of the preset user for the i-th clothing product clicked, M represents the selling price of the browsed clothing product, and C i Represents the consumption record of the i-th clicked clothing item.

[0036] As a preferred solution of the present invention, in which: during the start and end times of the preset users' previous browsing in different time periods, the number of clothes of the products recommended by the platform within the start and end times is collected, and the proportion of the number of clothes of the products clicked by the preset user in the number of clothes of the products recommended by the platform is calculated. When the proportion exceeds half of the number of clothes of the products recommended by the platform, the system determines that the preset user has a high degree of interest in the clothing features of the recommended products, otherwise, no determination is made.

[0037] As a preferred solution of the present invention, the consumption times of the preset user in the high-interest clothing items are collected, and a recommendation index is preset based on the consumption times. When the consumption times are less than two pieces, the system lowers the recommendation index of the corresponding clothing items and reduces the recommendation times in the future.

[0038] Beneficial effects:

[0039] The present invention utilizes deep learning technology, collects consumption records of preset users on the platform at different time periods, and obtains product clothing features corresponding to the consumption records, so as to more accurately recommend desired products and clothing to preset users in future time periods. Moreover, according to the product and clothing browsing records at different time periods in the past, by collecting the click rate of preset users on each product and clothing, the desired products of the preset users can be better locked, and the product and clothing that are not clicked by the user will be prohibited and / or the recommendation frequency will be reduced in future time periods, thereby improving the efficiency of product and clothing recommendation for preset users, so that the preset users can browse the desired products and clothing faster, improving the selection efficiency of the preset users, and being beneficial to facilitating consumption transactions. At the same time, by calculating the complementary color and dimming data of the product and clothing when displayed on the platform, the color difference between the product and clothing displayed on the platform webpage and the actual clothing can be better predicted, and then the product and clothing with large color difference can be prohibited from being recommended in future time periods, which is beneficial to further improving the accurate recommendation of product and clothing. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0041] Figure 1 This is a schematic diagram of the modular structure of an e-commerce platform product recommendation system based on deep learning according to an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of the process structure of an embodiment of the present invention;

[0043] Numbers in the figure: 110 - commodity consumption data collection module; 120 - browsing data analysis module; 130 - fusion data analysis unit; 1301 - calculation module; 1302 - determination module; 1303 - time collection module; 140 - image color comparison module; 150 - result data training module. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0045] Since the recommendation method of existing e-commerce platforms is still to intelligently recommend products based on users' historical behaviors and preferences, there is sometimes still the problem of inaccurate recommendation results, which causes users to feel troubled. In addition, over-reliance on recommendation systems will also cause users to be sold unnecessary products and even disrupt their shopping decisions.

[0046] Based on this, the present invention proposes an e-commerce platform product recommendation system based on deep learning, which utilizes deep learning technology to collect consumption records of preset users on the platform at different time periods, and obtain product and clothing features corresponding to the consumption records, so as to more accurately recommend desired products and clothing to preset users in future time periods. Moreover, based on the product and clothing browsing records at different time periods in the past, by collecting the click rate of each product and clothing by preset users, the desired products of preset users can be better locked, which is beneficial to facilitating consumer transactions.

[0047] The present invention is further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0048] Reference Figure 1 to Figure 2 , is an embodiment of the present invention, which provides an e-commerce platform product recommendation system based on deep learning, including:

[0049] The commodity consumption data collection module 110 is used to collect the historical clothing consumption data of the preset user on the e-commerce platform and analyze the characteristics of the clothing consumption data of the preset user;

[0050] It should be noted that in this embodiment, the preset user consumption data features include the preset user's consumption data in different time periods, the different time periods include the summer period, the spring period and the winter period, and the clothing features consumed in different time periods, the clothing features include material quality, sewing technology, version, style, color and fabric characteristics;

[0051] On the basis of the above, this embodiment collects the commodity click data from the preset user's browsing data in different time periods, and counts the start and end time of the preset user's previous browsing in the corresponding time period. Within the start and end time, the preset user's clicked commodities and click times are collected, the clothing characteristics of the clicked commodities are analyzed, a click rate prediction model is constructed, and the preset user's interest preference for the commodity is calculated through a collaborative filtering algorithm, which is calculated according to the following formula:

[0052] Among them, R ui Indicates the clothing product that is clicked by the preset user during the u-th browsing time in the corresponding time period;

[0053] In the formula, T represents the browsing time of the preset user for the i-th clothing product clicked, M represents the price of the browsed clothing product, and C i represents the consumption record of the i-th clicked clothing item;

[0054] A browsing data analysis module 120, which responds to the historical clothing consumption data collected by the commodity consumption data collection module, is used to obtain the clothing browsed by the preset user before the corresponding clothing consumption, and analyze its clothing characteristics;

[0055] It should be emphasized in this embodiment that the browsed clothing features are divided into Where n represents the nth clothing feature, and the final consumption records of the preset user under different clothing features are obtained. When the preset user browses clothing on the e-commerce platform in the future, the clothing corresponding to the consumption record is recommended to the preset user, and the clothing not corresponding to the consumption record is not recommended;

[0056] The fusion data analysis unit 130 is used to analyze the characteristic differences of corresponding clothes according to the return data and the commodity evaluation information corresponding to the preset user's historical clothing consumption data, wherein the characteristic differences include clothing pattern cutting differences, clothing style differences and clothing color differences. The fusion data analysis unit 130 includes a calculation module 1301, a determination module 1302 and a time collection module 1303;

[0057] The calculation module 1301 calculates the return rate based on the return data corresponding to the preset user's historical clothing consumption data and the product evaluation feedback information, and calculates the preset user's historical consumption times corresponding to different clothing characteristics based on the return rate;

[0058] The determination module 1302 is used to preset a critical threshold value based on the return rate accounting for 20% of the corresponding historical consumption times based on the preset user historical consumption times calculated by the calculation module 1301. When the return rate corresponding to a certain clothing feature exceeds the critical threshold value, the system determines that the return rate of the clothing corresponding to the clothing feature is high, and prohibits recommending the clothing corresponding to the clothing feature to the preset user; otherwise, no determination is made;

[0059] The time collection module 1303 is used to collect the time when the preset user browses the corresponding clothes under different clothing characteristics, and preset a minimum browsing time based on the browsing start and end time. When the browsing time of the preset user on the clothes under the corresponding clothing characteristics is less than the minimum browsing time, the system determines that the preset user has low interest in the clothes and prohibits recommendation. Otherwise, no determination is made;

[0060] The image color comparison module 140 is used to obtain the color data of the product when it is displayed on the platform, and compare the color of the physical clothing based on the color data, and perform color difference analysis, and at the same time obtain the clothing complementary color and dimming data corresponding to the color data according to the color difference analysis results, and classify the color difference analysis results;

[0061] Specifically, in this embodiment, the color difference analysis result is graded into 1 to 5 levels, where 1 represents the largest color difference, 5 represents the smallest color difference, and this level is regarded as no color difference. The specific grade standards are as follows:

[0062] Level 1: Color difference requirement is 13.6, tolerance requirement is ±1.0;

[0063] Level 2: Color difference requirement is 9.6, tolerance requirement is ±0.7;

[0064] Level 3: Color difference requirement is 6.8, tolerance requirement is ±0.6;

[0065] Level 4: Color difference requirement is 4.8, tolerance requirement is ±0.5;

[0066] Level 5: The color difference requirement is 0, and the tolerance requirement is ±0.22;

[0067] When the color difference analysis result is level 3, the system determines that the color difference is large, and prohibits recommending products with clothing color matching and dimming data corresponding to the result to the preset user in the future period. Otherwise, no determination is made;

[0068] The color data of the clothing includes color mode, hue, tone and saturation, wherein the color mode includes RGB and CMYK;

[0069] Dimming data includes illuminance, color temperature and lighting distribution;

[0070] A result data training module 150, which responds to the color difference analysis result of the image color comparison module and is used to train and adjust the clothing color complementation and dimming data corresponding to the color difference analysis result to obtain characteristic changes of the color difference analysis result;

[0071] It should be noted that in this embodiment, the adjustment training method includes reducing the complementary color and dimming data values ​​of the clothing corresponding to the color difference analysis results. Specifically, the color difference analysis results are collected by reducing each complementary color and dimming value until the collected color difference analysis results are consistent with the color of the actual clothing, and the total reduction value of the complementary color and dimming values ​​under the original values ​​is counted, and at the same time, based on the total reduction value, the platform history is collected to recommend products with complementary color and dimming values ​​greater than the total reduction value to the preset user, and the product is determined to be a product with a large color difference from the actual clothing, and the clothing characteristics of the product are analyzed, and it is forbidden to recommend products with the same clothing characteristics as the product to the preset user in the future time period;

[0072] Further explanation, according to the experience of the e-commerce industry, e-commerce platforms usually perform color complementation and dimming on clothes before displaying them, the purpose of which is to better display the color, texture and details of the clothes, so that the clothes look more beautiful and attractive in the pictures; however, in this process, the original color of the clothes will be changed to a certain extent, and then there will be color errors. In real life, color errors are one of the reasons why many users return goods. Therefore, the present invention collects the color complementation and dimming data given by the e-commerce platform during the product shooting process, and compares and analyzes the color difference generated by such data with the actual clothes, so that the color difference between the two can be more effectively judged. If such color difference is unacceptable to the preset user and the product is returned, the system will prohibit the recommendation of products with this color difference to the preset user in the future period, thereby improving the effectiveness of accurately recommending products to users;

[0073] On the basis of the above, this embodiment further divides the total reduction value into at least 3 groups of reduction intervals in which the complementary color and dimming values ​​differ by two values, and collects the browsing records and consumption records of the preset user in different reduction intervals, and at the same time obtains the complementary color and dimming data of the physical clothing purchased by the preset user based on the consumption record, obtains the reduction interval corresponding to the complementary color and dimming data, and collects the clothing products that the preset user has not returned based on the reduction interval, analyzes the characteristics of the product clothing, and recommends products with the same clothing characteristics to the preset user in the future time period;

[0074] Among them, this embodiment further collects the number of clothes of the products recommended by the platform during the start and end times of the previous browsing of the preset user in different time periods, and calculates the proportion of the number of clothes of the products clicked by the preset user in the number of clothes of the products recommended by the platform. When the proportion exceeds half of the number of clothes of the products recommended by the platform, the system determines that the preset user has a high degree of interest in the clothes features of the recommended products, otherwise, no determination is made;

[0075] This embodiment also includes collecting the consumption times of preset users among the high-interest clothing products, and presetting the recommendation index based on the consumption times. When the consumption times are less than two pieces, the system lowers the recommendation index of the corresponding clothing product and reduces the recommendation times in the future.

[0076] In summary, the present invention utilizes deep learning technology to collect consumption records of preset users on the platform at different time periods, and by obtaining the characteristics of goods and clothing corresponding to the consumption records, it can more accurately recommend desired goods and clothing to preset users in future time periods. Moreover, based on the browsing records of goods and clothing in different time periods in the past, by collecting the click rate of each goods and clothing by the preset users, the desired goods of the preset users can be better locked. In order to obtain the goods and clothing clicked by the users, the recommendation frequency is prohibited and / or reduced in future time periods, thereby improving the efficiency of recommending goods and clothing to the preset users, enabling the preset users to browse the desired goods and clothing faster, thereby improving the selection efficiency of the preset users, and being beneficial to facilitating consumer transactions.

[0077] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. The e-commerce platform product recommendation system based on deep learning is characterized by: include: The commodity consumption data collection module is used to collect the historical clothing consumption data of the preset user on the e-commerce platform and analyze the characteristics of the clothing consumption data of the preset user; A browsing data analysis module, which responds to the historical clothing consumption data collected by the commodity consumption data collection module, and is used to obtain the clothing browsed by the preset user before the corresponding clothing consumption, and analyze its clothing characteristics; A fusion data analysis unit, configured to analyze characteristic differences of corresponding clothing according to the return data and product evaluation information corresponding to the preset user's historical clothing consumption data, wherein the characteristic differences include clothing pattern cutting differences, clothing style differences, and clothing color differences, and the fusion data analysis unit includes a calculation module, a determination module, and a time acquisition module; The calculation module calculates a return rate based on the return data corresponding to the preset user's historical clothing consumption data and the product evaluation feedback information, and calculates the preset user's historical consumption times corresponding to different clothing characteristics based on the return rate; The determination module is based on the preset user historical consumption times corresponding to different clothing features calculated by the calculation module, and is used to preset a critical threshold value based on the return rate accounting for 20% of the corresponding historical consumption times. When the return rate corresponding to a certain clothing feature exceeds the critical threshold value, the system determines that the return rate of the clothing corresponding to the clothing feature is high, and prohibits recommending the clothing corresponding to the clothing feature to the preset user; otherwise, no determination is made; The time collection module is used to collect the time when the preset user browses the corresponding clothes under different clothing characteristics, and preset a minimum browsing time based on the start and end time of the browsing. When the browsing time of the preset user on the clothes under the corresponding clothing characteristics is less than the minimum browsing time, the system determines that the preset user has low interest in the clothes and prohibits recommendation. Otherwise, no determination is made; An image color comparison module is used to obtain the color data of the product when it is displayed on the platform, and compare the color of the physical clothing based on the color data, and perform color difference analysis, and at the same time obtain the clothing complementary color and dimming data corresponding to the color data according to the color difference analysis results, and classify the color difference analysis results; A result data training module, which responds to the color difference analysis result of the image color comparison module and is used to train and adjust the clothing color complement and dimming data corresponding to the color difference analysis result to obtain characteristic changes in the color difference analysis result.

2. The e-commerce platform product recommendation system based on deep learning as claimed in claim 1, characterized in that: In the commodity consumption data collection module, the preset user consumption data characteristics include the consumption data of the preset user in different time periods, the different time periods include the summer period, the spring period and the winter period, and the characteristics of clothing consumed in different time periods, the clothing characteristics include material quality, sewing technology, version, style, color and fabric characteristics.

3. The e-commerce platform product recommendation system based on deep learning as claimed in claim 2, characterized in that: In the browsing data analysis module, the browsed clothing features are divided into ▽1, ▽2, ..., ▽ n , where n represents the nth clothing feature, and the final consumption records of the preset user under different clothing features are obtained. When the preset user browses clothing on the e-commerce platform in the future, the clothing corresponding to the consumption record is recommended to the preset user, and the clothing not corresponding to the consumption record is not recommended.

4. The e-commerce platform product recommendation system based on deep learning as claimed in claim 1, characterized in that: In the image color comparison module, the color difference analysis result is graded into 1 to 5 levels, where 1 represents the largest color difference, 5 represents the smallest color difference, and this level is regarded as no color difference. The specific grade standards are as follows: Level 1: Color difference requirement is 13.6, tolerance requirement is ±1.0; Level 2: Color difference requirement is 9.6, tolerance requirement is ±0.7; Level 3: Color difference requirement is 6.8, tolerance requirement is ±0.6; Level 4: Color difference requirement is 4.8, tolerance requirement is ±0.5; Level 5: The color difference requirement is 0, and the tolerance requirement is ±0.22; When the color difference analysis result level is level 3, the system determines that the color difference is large, and prohibits recommending products with clothing color complementation and dimming data corresponding to the result to the preset user in the future period. Otherwise, no determination is made; The color data of the clothing includes color mode, hue, tone and saturation, wherein the color mode includes RGB and CMYK; The dimming data includes illuminance, color temperature and lighting distribution.

5. The e-commerce platform product recommendation system based on deep learning as claimed in claim 1, characterized in that: In the result data training module, the adjustment training method includes reducing the complementary color and dimming data values ​​of the clothing corresponding to the color difference analysis results. Specifically, the color difference analysis results are collected each time a complementary color and dimming value is reduced until the collected color difference analysis results are consistent with the color of the actual clothing, and the total reduction value of the complementary color and dimming values ​​under the original values ​​is counted. At the same time, based on the total reduction value, the platform history is collected to recommend products with complementary color and dimming values ​​greater than the total reduction value to the preset user, and the product is determined to be a product with a large color difference from the actual clothing. At the same time, the clothing characteristics of the product are analyzed, and it is prohibited to recommend products with the same clothing characteristics as the product to the preset user in the future time period.

6. The e-commerce platform product recommendation system based on deep learning as claimed in claim 5, characterized in that: The total reduction value is split into at least 3 groups of reduction intervals whose complementary color and dimming values ​​differ by two values, and the browsing history and consumption history of the preset user are collected within different reduction intervals, and the complementary color and dimming data of the physical clothing purchased by the preset user are obtained based on the consumption records, and the reduction intervals corresponding to the complementary color and dimming data are obtained, and clothing products that have not been returned by the preset user are collected based on the reduction intervals, the clothing characteristics of the products are analyzed, and products with the same clothing characteristics are recommended to the preset user in a future time period.

7. The e-commerce platform product recommendation system based on deep learning as claimed in claim 2, characterized in that: Collect the product click data from the preset user's browsing data in different time periods, and count the start and end times of the preset user's previous browsing in the corresponding time period. Collect the clicked products and click times of the preset user within the start and end time, analyze the clothing features of the clicked products, build a click rate prediction model, and calculate the preset user's interest preference for the products through a collaborative filtering algorithm, which is calculated according to the following formula: Among them, R ui represents the i-th clicked clothing product during the u-th browsing of the preset user in the corresponding time period; Where, T represents the browsing time of the preset user for the i-th clothing product clicked, M represents the selling price of the browsed clothing product, and C i Represents the consumption record of the i-th clicked clothing item.

8. The e-commerce platform product recommendation system based on deep learning as claimed in claim 7, characterized in that: At the start and end times of the preset users' previous browsing in different time periods, the number of clothes recommended by the platform within the start and end times is collected, and the proportion of the number of clothes clicked by the preset users in the number of clothes recommended by the platform is calculated. When the proportion exceeds half of the number of clothes recommended by the platform, the system determines that the preset user has a high level of interest in the clothing features of the recommended products, otherwise, no determination is made.

9. The e-commerce platform product recommendation system based on deep learning as claimed in claim 8, characterized in that: The consumption times of the preset user in the high-interest clothing items are collected, and a recommendation index is preset based on the consumption times. When the consumption times are less than two pieces, the system lowers the recommendation index of the corresponding clothing items and reduces the recommendation times in the future.

Images