Commodity recommendation method and system for overseas e-commerce platform
Through screenshots and information analysis, the recommendation list is regenerated, which solves the problem of users misleading into product pages that are not interested in, improves user experience and reduces usage costs.
Patent Information
- Application Number
- CN202411427217.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-14
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-10-14
AI Technical Summary
The prior art is difficult to find that users are misled by push messages and enter uninterested product display pages, resulting in a decline in user experience and an increase in usage costs.
By performing screenshot processing when the push message is triggered, the push message screenshot is obtained, and jump to the product display page to obtain type information, brand information and style information. Combining the information of some pages that have been read, the basic information recommendation coefficient and content recommendation are calculated. Coefficient, regenerate the recommendation list and reduce the push frequency of uninterested products.
Effectively reduce the frequency of push messages for products that users are not interested in, reduce network fees and loading time, improve user experience, increase the probability of pushing interested products, and reduce user usage costs.
Smart Images

Figure CN118941368A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of product recommendation, and in particular to a product recommendation method and system for overseas e-commerce platforms. Background Art
[0002] CN113327151B relates to a method, device, equipment, and medium for recommending commodity objects. The method includes: obtaining a list of commodities to be recommended, and classifying the commodity objects to be recommended into their respective corresponding target categories according to a preset classification catalog; for any commodity object to be recommended, obtaining the basic recommendation degree and trend status corresponding to its target category during a specific time period; calculating the adjusted recommendation degree under the corresponding trend based on the upward or downward trend represented by the trend status of its target category; adding or subtracting the adjusted recommendation degree under the corresponding trend based on the trend status of its target category to the basic recommendation degree of its target category to obtain the commodity recommendation score of the commodity object; and selecting some commodity objects from the recommended commodity list based on the commodity recommendation score and determining them as a commodity push list. This solution can more accurately and effectively select more trendy commodity objects to be recommended for e-commerce platforms.
[0003] CN116258541A relates to a push method, a computer-readable storage medium, and an electronic device. The method aims to explore the purchasing power of potential users and provide targeted product recommendations to potential users that match their purchasing power. The method comprises: clustering all users based on their respective user attribute information to obtain multiple clustering results, where each clustering result represents a user group formed by users with similar attributes; determining the user group to which a new user among all users who has not performed a preset action on a target product belongs; determining at least one target user from the user group to which the new user belongs who has performed a preset action on the target product; determining the purchasing power value of the new user based on the purchasing power value of the at least one target user; and pushing product information to the user's terminal device based on the user's purchasing power value.
[0004] Therefore, in the related art, push notifications can be sent based on the characteristics of the user or product. However, the display area of the push message is usually small and cannot fully display all the characteristics of the product. It can only display partial product information, such as type, brand, etc. The user may click on the push message based on the partial content displayed in the push message and jump to the product display page to read the full information of the product. The full information of the product is richer than the push message, that is, it contains more information. The user can further learn more about the product, such as the product style, product introduction, etc. However, in the process of further understanding, the user may find that the product is not the product they are interested in and exit the product display page. In this case, the user is misled by the push message and enters the product display page of the product they are not interested in. The related art also has difficulty in detecting the situation where the user is misled by the push message and enters the product display page of the product they are not interested in, and it is also unable to reduce the number of push messages for products that the user is not interested in, resulting in inaccurate push notifications and a poor user experience. Especially in the case of push notifications on overseas e-commerce platforms, each jump requires a long loading time and high network charges. If the push notifications are inaccurate, it will increase the user's usage cost and reduce the user experience.
[0005] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the Invention
[0006] The present invention provides a product recommendation method and system for overseas e-commerce platforms, which can solve the technical problem that related technologies have difficulty in detecting situations where users are misled by push messages and enter product display pages that they are not interested in.
[0007] According to a first aspect of the present invention, there is provided a product recommendation method for an overseas e-commerce platform, comprising: When a push message is triggered, a screenshot of the push message displayed on the screen is taken to obtain a push message screenshot; Jumping to a product display page corresponding to the push message, wherein the push message includes summary information of the content on the product display page, and the push message screenshot is an image obtained by taking a screenshot of the push message displayed on a display; Obtaining the type, brand, and style information of the product on the product display page; When exiting the product display page, determining a portion of pages that have been read in the product display page, wherein the portion of pages that have been read includes a portion of pages of the product display page that have been displayed on the display during a period between jumping to the product display page and exiting the product display page; Obtaining information about the read content in the read portion of the page; Determining a basic information recommendation coefficient based on the push message screenshot, the type information, the brand information, the style information, and the read content information; Determining a content recommendation coefficient based on the push message screenshot and the read content information; Determining recommendation coefficients for a plurality of products based on the basic information recommendation coefficient and the content recommendation coefficient; A recommendation list of the plurality of commodities is obtained according to the recommendation coefficients of the plurality of commodities, and a new push message is generated according to the recommendation list.
[0008] According to a second aspect of the present invention, there is provided a product recommendation system for an overseas e-commerce platform, comprising: The screenshot module is used to take a screenshot of the push message displayed on the screen when the push message is triggered, and obtain a screenshot of the push message; a jump module, configured to jump to a product display page corresponding to the push message, wherein the push message includes summary information of the content on the product display page, and the push message screenshot is an image obtained by taking a screenshot of the push message displayed on a display; The basic information module is used to obtain the type information, brand information and style information of the products on the product display page; a read module, configured to determine a portion of pages that have been read on the product display page when exiting the product display page, wherein the portion of pages that have been read includes a portion of pages that have been displayed on the display during a period between jumping to the product display page and exiting the product display page; A read content information module is used to obtain the read content information of the read part of the page; A basic information recommendation coefficient module, configured to determine a basic information recommendation coefficient based on the push message screenshot, the type information, the brand information, the style information, and the read content information; A content recommendation coefficient module, configured to determine a content recommendation coefficient based on the push message screenshot and the read content information; A recommendation coefficient module, configured to determine recommendation coefficients for a plurality of products based on the basic information recommendation coefficient and the content recommendation coefficient; The push message module is used to obtain a recommendation list of multiple commodities according to the recommendation coefficients of the multiple commodities, and generate a new push message according to the recommendation list.
[0009] Technical effect: According to the present invention, it is possible to determine whether the user has been misled by the push message and entered a product display page in which the user is not interested based on the part of the page that the user has read when exiting the product display page, and to determine the information that the user is interested in and not interested in based on the comparison between the information contained in the push message and the information in the part of the page that the user has read in the product display page, thereby determining the recommendation coefficient of each product, thereby regenerating the recommendation list and push messages, reducing the push frequency of push messages for products that the user is not interested in, reducing network charges and loading time, reducing the user's usage cost, and improving the user experience. When determining the basic information recommendation score, the basic information recommendation coefficient can be determined based on whether the three basic information are fully displayed in the push message and whether the user is interested in the product. When one or more basic information is not fully displayed and the user is not interested in the product, a basic information recommendation score that can attenuate the preference score in the recommendation list can be obtained, thereby reducing the probability of pushing products that the user is not interested in and increasing the probability of pushing products that the user is interested in, improving the user experience, and reducing the user's usage cost. When determining the content recommendation coefficient, the probability of each second keyword causing the user to be uninterested in the product can be determined by word frequency, and the possibility of each second keyword causing the user to be uninterested in the product can be determined based on the location information of the second keyword. This can obtain a content recommendation coefficient that attenuates the preference score in the recommendation list, thereby reducing the probability of pushing products that the user is not interested in, increasing the probability of pushing products that the user is interested in, improving user experience, and reducing user usage costs. When determining the recommendation coefficient of other products, the basic information recommendation coefficient of other products can be determined based on whether the basic information of other products other than the product corresponding to the push message belongs to the set of basic information that the user is not interested in. This can obtain a basic information recommendation score that attenuates the preference score of other products, and can also obtain the basic information recommendation score of other products even if no push message for other products is received. Therefore, when the user is not interested in a certain product, the preference score of similar products can be reduced to reduce the probability of pushing products that the user is not interested in. In addition, the content recommendation coefficient can be determined by word frequency and the occurrence position of the second target keyword with the highest semantic similarity to the third keyword, so as to accurately indicate the possibility that the user is not interested in other products. In addition, the content recommendation score of other products can be obtained even without receiving push messages for other products. Therefore, when the user is not interested in a certain product, the preference score of similar products can be lowered to reduce the probability of pushing products that the user is not interested in, thereby improving the user experience and reducing the user's usage cost.
[0010] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and not limiting of the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can derive other embodiments based on these drawings without inventive efforts. Figure 1 A flowchart of a product recommendation method for an overseas e-commerce platform according to an embodiment of the present invention is exemplarily shown; Figure 2 A schematic diagram exemplarily illustrating a push message displayed on a display of a mobile device according to an embodiment of the present invention; Figure 3 A schematic diagram exemplarily showing a push message displayed on a display of a personal computer according to an embodiment of the present invention; Figure 4 A block diagram of a product recommendation system for an overseas e-commerce platform according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION
[0012] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.
[0013] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.
[0014] Figure 1 A flowchart of a product recommendation method for an overseas e-commerce platform according to an embodiment of the present invention is exemplarily shown. The method includes: Step S101, when a push message is triggered, taking a screenshot of the push message displayed on the screen to obtain a push message screenshot; Step S102: Jump to the product display page corresponding to the push message, wherein the push message includes summary information of the content on the product display page, and the push message screenshot is an image obtained by taking a screenshot of the push message displayed on the display; Step S103, obtaining the type information, brand information and style information of the product on the product display page; Step S104, when exiting the product display page, determining a portion of pages that have been read in the product display page, wherein the portion of pages that have been read includes a portion of pages of the product display page that have been displayed on the display during a period between jumping to the product display page and exiting the product display page; Step S105, obtaining the read content information of the read part of the page; Step S106, determining a basic information recommendation coefficient based on the push message screenshot, the type information, the brand information, the style information, and the read content information; Step S107, determining a content recommendation coefficient based on the push message screenshot and the read content information; Step S108, determining recommendation coefficients for multiple products based on the basic information recommendation coefficient and the content recommendation coefficient; Step S109: obtaining a recommendation list of multiple commodities according to the recommendation coefficients of the multiple commodities, and generating a new push message according to the recommendation list.
[0015] According to an embodiment of the present invention, a product recommendation method for an overseas e-commerce platform can determine whether the user is misled by a push message and enters a product display page in which the user is not interested, based on the read portion of the page when the user exits the product display page, and can determine the information that the user is interested in and not interested in based on the comparison between the information contained in the push message and the information in the read portion of the page in the product display page, thereby determining the recommendation coefficient of each product, and regenerating the recommendation list and push messages, reducing the push frequency of push messages for products that the user is not interested in, reducing network charges and loading time, reducing the user's usage cost, and improving the user experience.
[0016] According to one embodiment of the present invention, a user may install an application of an overseas e-commerce platform in a mobile device (e.g., a mobile phone) or a personal computer. The application may push introductory information about some products to the user, i.e., push messages. For example, the application may record the user's browsing history for products and estimate the user's preference for each type of product through the browsing history. For example, the higher the user's preference for a certain type of product, the more times the user browses this type of product. Based on the ratio of the number of views of this type of product to the total number of views of all products, the preference score of this type of product may be determined, and then a recommendation list of each type of product may be determined based on the preference score of each type of product. Furthermore, you can select multiple types of products that are ranked higher in the recommendation list, and generate push messages based on the type information, brand information, and style information of these products. When generating push messages, you can determine the probability of each type of product generating push messages according to the ranking of each type of product. For example, the product ranked first in the recommendation list has the highest probability of generating a push message, and the product ranked second in the recommendation list has a slightly lower probability of generating a push message than the product ranked first, etc. After the push message is generated, it can be pushed to the message bar of the user's mobile device or personal computer, so that the user can see the push message in the message bar of the mobile device or personal computer.
[0017] Figure 2 A schematic diagram exemplarily shows a push message displayed on a display of a mobile device according to an embodiment of the present invention. Figure 3 A schematic diagram exemplarily shows a push message displayed on a display of a personal computer according to an embodiment of the present invention.
[0018] According to one embodiment of the present invention, Figure 2 On the display of the mobile device in the Figure 3On a personal computer monitor, a push message may be displayed in the lower right corner of the screen. The area where the push message is displayed is relatively small and cannot display all product information. It may only display partial product information, such as product type and brand information. A user may see information of interest in the push message. For example, if the user is interested in a certain brand and the push message includes information about the brand, the user may click on the push message to view detailed information about the brand's products. However, since a push message cannot display all product information, the product may not be the one the user is truly interested in. For example, if the user is interested in the brand of the product, the user may click on the push message and enter the product display page to read the product details. However, if the user discovers that the product style is not of interest while browsing the details, they may exit the product display page. They may even exit immediately upon encountering a keyword that does not interest them, without completing the entire product display page. Therefore, a preset percentage can be set for each product display page. Of the entire content on the product display page, the content before the preset percentage fully describes the product's characteristics, while the content after the preset percentage provides a detailed explanation or description of these characteristics. For example, the first 40% of a product page's content fully describes the product's key features, while the remaining 60% provides a more detailed explanation of these features. For example, if the product is clothing, the first 40% of the page might describe the brand, type (e.g., jacket, down jacket, sweater, or T-shirt), and style (e.g., long or short sleeves, stand-up or crew neck, color, and fabric). The remaining 60% provides a detailed introduction to the brand or style (e.g., brand history, fabric selection, and production methods). Therefore, the first 40% of the content primarily describes the clothing's features and is more valuable for users when selecting clothing. Therefore, 40% can be used as the default percentage. If a user reads more than or equal to the default percentage on a product page, it indicates that the user has continued reading after understanding the product's key features, indicating interest in the product. If the content read on the user's product display page is less than the preset percentage, it means that the user has chosen to exit before fully understanding the main features of the product. This may mean that the user may have discovered features that are not of interest to them while understanding the main features of the product, and therefore exited the product display page.
[0019] According to one embodiment of the present invention, in step S101, as described above, the push message may include part of the product information. When the user receives the push message, he or she may be interested in part of the information displayed in the push message, and thus clicks on the push message. When the push message is triggered, the push message may be firstly screenshotted, for example, Figure 2 Take a screenshot of the push message in the upper middle, or Figure 3 The push message in the lower right corner is captured to obtain a push message screenshot, in which part of the content displayed by the push message can be identified. Further, in step S102, the page is jumped to the product display page corresponding to the push message, that is, all information about the product is displayed.
[0020] According to an embodiment of the present invention, in step S103, basic information such as type information, brand information, and style information of the product in the product display page may be obtained for reference.
[0021] According to one embodiment of the present invention, in step S104, when the product display page is exited, the read portion of the product display page can be determined, that is, the portion of the product display page displayed on the display during the time period between jumping to the product display page and exiting the product display page. The area ratio of the read portion of the page to the product display page can also be used as the read percentage, or the ratio of the number of words in the text content of the read portion of the page to all the text content of the product display page can be used as the read percentage.
[0022] According to an embodiment of the present invention, in step S105 , information about the read content in the read portion of the page may be obtained, for example, text content in the read portion of the page.
[0023] According to one embodiment of the present invention, in step S106, the basic information recommendation coefficient may be determined based on a comparison between the basic information of the product in the product display page (eg, type information, brand information, and style information) and part of the basic information in the push message.
[0024] According to one embodiment of the present invention, a basic information recommendation coefficient is determined based on the push message screenshot, the type information, the brand information, the style information and the read content information, including: recognizing and processing the text in the push message screenshot through a text recognition model to obtain the text content in the push message screenshot; recognizing the text content to obtain the first basic information that can be fully displayed in the push message screenshot and the second basic information that cannot be fully displayed in the type information, brand information and style information of the product; determining the read ratio of the read content information in all the content on the product display page based on the read content information; determining the basic information recommendation coefficient based on the first basic information, the second basic information, the type information, the brand information, the style information and the read ratio.
[0025] According to one embodiment of the present invention, the text recognition model is a deep learning neural network model, for example, a convolutional neural network model. The present invention does not limit the type of the text recognition model. The push message screenshot can be processed to extract the text in the push message screenshot and obtain the text content in the push message screenshot. These text contents are the information that can be displayed in the push message and are also the information that can be read by the user. Furthermore, since the user clicks on the push message after reading this information, this information is also information that the user is interested in.
[0026] According to one embodiment of the present invention, text content can be identified and it can be determined whether the text content can fully display the type information, brand information and style information in the product display page. If it can be fully displayed, the information is classified as the first basic information. If it cannot be fully displayed, the information is classified as the second basic information. For example, if the brand information of the product is fully displayed in the text content, the brand information is classified as the first basic information. For another example, if the style information of the product cannot be fully displayed in the text content, the style information can be classified as the second basic information.
[0027] According to one embodiment of the present invention, the read ratio corresponding to the read content information can be obtained. The method for determining the read ratio is as described above and will not be repeated here.
[0028] According to an embodiment of the present invention, determining the basic information recommendation coefficient based on the first basic information, the second basic information, the type information, the brand information, the style information and the read ratio includes: determining the basic information recommendation coefficient according to formula (1) , (1) in, For the type information, For the brand information, For the style information, is a set of first basic information, is the reading percentage, is the preset ratio, is the conditional function.
[0029] According to one embodiment of the present invention, Is a conditional function, indicating that under the condition If the condition is met, the conditional function value is 1, otherwise If the type information belongs to the set of the first basic information, that is, the type information is fully displayed in the text content of the push message screenshot, then the user is interested in the type information of the product, and the condition function value is 1. Otherwise, the condition function value is the inner condition function. , the inner condition function is expressed in In the case of , the inner condition function value is 1, otherwise , which means that if the read ratio is greater than or equal to the preset ratio, it means that the user did not exit after understanding the main features of the product, but continued to read, which means that the user is interested in the product, and the inner condition function value is 1. Otherwise, it means that the user is not interested in the product, and a ratio less than 1 can be used. , the preference score of the product in the product recommendation list is attenuated, so that the ranking of the product in the recommendation list is reduced, thereby reducing the probability of generating a push message for the product. If the condition function value is 1, the preference score of the product can be maintained. The above condition function can be used to represent the recommendation coefficient corresponding to type information. Similarly, the recommendation coefficient corresponding to brand information can also be obtained. Recommendation coefficient corresponding to style information The recommendation coefficients corresponding to the three types of basic information are averaged to obtain the basic information recommendation coefficient. Therefore, when one or more basic information items are not fully displayed and the user is not interested in the product, a basic information recommendation coefficient is obtained that can attenuate the preference score in the recommendation list, thereby reducing the probability of recommending products that the user is not interested in.
[0030] In this way, the basic information recommendation coefficient can be determined based on whether the three basic information are fully displayed in the push message and whether the user is interested in the product. When one or more basic information is not fully displayed and the user is not interested in the product, a basic information recommendation score can be obtained that can attenuate the preference score in the recommendation list, thereby reducing the probability of pushing products that the user is not interested in, increasing the probability of pushing products that the user is interested in, improving user experience, and reducing user usage costs.
[0031] According to one embodiment of the present invention, in step S107, it is also possible to determine whether the user is not interested in certain information about the product based on the content that the user has read on the product display page and the content in the push message screenshot, thereby determining the content recommendation coefficient to reduce the probability of pushing products that the user is not interested in.
[0032] According to one embodiment of the present invention, a content recommendation coefficient is determined based on the push message screenshot and the read content information, including: identifying and processing the text in the push message screenshot through a text recognition model to obtain the text content in the push message screenshot; performing word segmentation and deduplication processing on the text content to obtain multiple first keywords of the text content; performing feature extraction on the first keyword through the text recognition model to obtain first semantic feature information of each first keyword; determining the read ratio of the read content information in all the content on the product display page based on the read content information; performing word segmentation and deduplication processing on the read content information to obtain multiple second keywords of the read content information; determining the appearance position information of the second keyword in all the content on the product display page for the first time; performing feature extraction on the second keyword through the text recognition model to obtain second semantic feature information of each second keyword; determining the content recommendation coefficient based on the first semantic feature information, the second semantic feature information, the appearance position information and the read ratio.
[0033] According to one embodiment of the present invention, the method of obtaining the text content in the push message screenshot and obtaining the read percentage is as described above and will not be repeated here. The text content can be pre-processed by word segmentation and word segmentation, and multiple first keywords of the text content can be obtained. Each first keyword can appear once or multiple times in the text content. Further, the encoding layer of the text recognition model can be used to obtain the first semantic feature information of each first keyword. The first semantic feature information can be information in the form of a vector. Similarly, the read content information can be similarly processed to obtain multiple second keywords of the read content information and the second semantic feature information of each second keyword. Each second keyword can also appear once or multiple times in the read content information. The first appearance position information of the second keyword in all the content on the product display page can be recorded. The appearance position information can be ratio information. For example, the text on the product display page has a total of 1,000 words, and the first word of a second keyword appears in the 200th word on the product display page. Then the appearance position information of the first appearance of the second keyword is 200 / 1,000=20%.
[0034] According to an embodiment of the present invention, determining a content recommendation coefficient based on the first semantic feature information, the second semantic feature information and the read ratio includes: determining the content recommendation coefficient according to formula (2) , (2) in, is the second semantic feature information of the i-th second keyword, is the first semantic feature information of the jth first keyword, for The transposed vector of for and The cosine similarity of is the preset similarity threshold, max is the maximum value function, m is the number of first keywords, is the conditional function, is the reading percentage, is the preset ratio, is the first occurrence position information of the i-th second keyword in all the content on the product display page, is the frequency of the second keyword i in the read content information, is the average frequency of multiple second keywords, n is the number of first keywords, j≤m, i≤n, and i, j, m and n are all positive integers.
[0035] According to one embodiment of the present invention, in formula (2), for and The cosine similarity can also be used as the semantic similarity between the i-th second keyword and the j-th first keyword. It represents the maximum value of the semantic similarity between multiple first keywords and the i-th second keyword. If the maximum value is greater than or equal to the preset similarity threshold, it means that among the multiple first keywords, there is a first keyword with a higher semantic similarity with the i-th second keyword (for example, the same as the i-th second keyword, or a synonym of the i-th second keyword). Furthermore, after receiving the push message, the user enters the product display page, which means that the user is interested in the information in the push message. In other words, the user is interested in the first keyword. Therefore, if the i-th second keyword is a synonym of a first keyword, the user is interested in the i-th second keyword, and the conditional function value is set to 1. Otherwise, the i-th second keyword is a new keyword that appears in the product display page, that is, a keyword that is semantically different from all keywords in the text content of the push message, and the conditional function value is the inner conditional function .
[0036] According to one embodiment of the present invention, in the inner condition function, if the read ratio is greater than or equal to the preset ratio, it means that the user is interested in the product, that is, after reading the i-th second keyword, the user is still interested in the product, and the inner condition function value is 1; otherwise, it means that the user is not interested in the product, and the inner condition function value is ,in, is the ratio of the word frequency of the ith second keyword to the average word frequency of all second keywords. The larger the ratio, the more times the second keyword appears, and the more likely the user is to be uninterested in the product after reading the second keyword. is the attenuation ratio of the preference score, which can be and The product of is the attenuation coefficient of the preference score of the second keyword i, Then represents the remaining proportion of the preference score after attenuation. It represents the relative gap between the read ratio and the appearance position information of the i-th second keyword. The smaller the relative gap between the two, the more likely it is that the user will feel uninterested in the product after reading the i-th second keyword and then exit the product display page. Specifically, if the i-th second keyword first appears at an earlier position, the relative gap between the read ratio and the appearance position information of the i-th second keyword will be larger, that is, the user did not immediately exit the product display page after reading the i-th second keyword. In other words, it is less likely that the user will feel uninterested in the product due to the i-th second keyword. On the contrary, if the i-th second keyword first appears at a later position, close to the read ratio, it means that the user will immediately exit the product display page after reading the i-th second keyword. In other words, it is more likely that the user will feel uninterested in the product due to the i-th second keyword. Therefore, the relative gap can be used to continue to attenuate the above-mentioned remaining ratio, thereby accurately describing the user's preference for the characteristics of the product described by the i-th second keyword.
[0037] According to one embodiment of the present invention, after the above analysis, the conditional function As the recommendation coefficient corresponding to the i-th second keyword. Further, the recommendation coefficients corresponding to all second keywords can be obtained and averaged to obtain the content recommendation coefficient. Therefore, if the second keyword is not a keyword that the user is interested in and the user is not interested in the product, the content recommendation coefficient that attenuates the preference score in the recommendation list based on the word frequency and the position information of the second keyword is obtained, thereby reducing the probability of pushing products that the user is not interested in.
[0038] In this way, the probability that each second keyword will cause the user to be uninterested in the product can be determined by word frequency, and the possibility that each second keyword will cause the user to be uninterested in the product can be determined based on the appearance position information of the second keyword, so as to obtain a content recommendation coefficient that attenuates the preference score in the recommendation list, thereby reducing the probability of pushing products that the user is not interested in, increasing the probability of pushing products that the user is interested in, improving user experience, and reducing user usage costs.
[0039] According to one embodiment of the present invention, in step S108, the recommendation coefficients of multiple products can be determined, that is, the recommendation coefficients of the products corresponding to the above-mentioned push messages, as well as the recommendation coefficients of other products, and the recommendation coefficients of each product can be multiplied by the preference score of each product to obtain a new preference score, and then a new product recommendation list can be obtained.
[0040] According to one embodiment of the present invention, the recommendation coefficients of multiple commodities are determined based on the basic information recommendation coefficient and the content recommendation coefficient, including: determining the recommendation coefficient of the commodity corresponding to the push message based on the basic information recommendation coefficient and the content recommendation coefficient; if the read ratio is less than the preset ratio, obtaining the recommendation coefficients of the commodities other than the commodity corresponding to the push message based on the type information, brand information, style information and content in the commodity display page of the commodities other than the commodity corresponding to the push message, as well as the second basic information and the second keyword; if the read ratio is greater than or equal to the preset ratio, setting the recommendation coefficients of the commodities other than the commodity corresponding to the push message to 1.
[0041] According to one embodiment of the present invention, determining the recommendation coefficient of the product corresponding to the push message based on the basic information recommendation coefficient and the content recommendation coefficient includes: determining the recommendation coefficient of the product corresponding to the push message according to formula (3) , (3) in, is the content recommendation coefficient, is the basic information recommendation coefficient.
[0042] According to one embodiment of the present invention, the content recommendation coefficient and the basic information recommendation coefficient calculated above can be multiplied together to obtain the recommendation coefficient of the product corresponding to the push message. If the user is not interested in the product, the preference score of the product can be attenuated based on the recommendation coefficient, and then the ranking in the recommendation list can be lowered, thereby reducing the probability of pushing the product and improving the user experience.
[0043] According to one embodiment of the present invention, the recommendation coefficients of other products can also be calculated with reference to the calculation method of the recommendation coefficients of the above-mentioned products, so as to determine the recommendation coefficients of each product, and then obtain new preference scores for each product and obtain a new recommendation list.
[0044] According to one embodiment of the present invention, if the read ratio is less than a preset ratio, then based on the type information, brand information, style information and content on the product display page of the products other than the product corresponding to the push message, as well as the second basic information and the second keyword, obtaining the recommendation coefficient of the products other than the product corresponding to the push message includes: obtaining the judgment condition J according to formula (4), (4) in, is the second semantic feature information of the i-th second keyword, is the first semantic feature information of the jth first keyword, for The transposed vector of for and The cosine similarity of is the preset similarity threshold, max is the maximum value function, and m is the number of first keywords; if the second semantic feature information of the i-th second keyword meets the discrimination condition J, the i-th second keyword is determined as the second target keyword; based on the content of the product display page of other products except the product corresponding to the push message, word segmentation and deduplication processing are performed to obtain multiple third keywords; through the text recognition model, the third keywords are feature extracted to obtain the third semantic feature information of each third keyword; according to formula (5), the recommendation coefficient of the k-th product except the product corresponding to the push message is obtained , (5) in, is a set of the second basic information, is the reading percentage, is the preset ratio, is the conditional function, is the type information of the kth product, is the brand information of the kth product, is the style information of the kth product, is the second semantic feature information of the t-th second target keyword, is the number of the second target keywords, is the third semantic feature information of the sth third keyword, for The transposed vector of for and The cosine similarity of is the preset similarity threshold, max is the maximum value function, is the frequency of the sth third keyword in the content of the product display page of the kth product, is the average frequency of multiple third keywords in the content of the product display page of the kth product, is the number of third keywords in the content of the product display page of the kth product, To make The first occurrence position of the second target keyword with the maximum value in all content on the product display page, s≤ , t≤ , and s, , t, k and All are positive integers.
[0045] According to one embodiment of the present invention, if the percentage of times a user has read the product display page for the product is less than a preset percentage, it can be considered that the user is not interested in the product. In this case, the second target keyword can be selected based on formula (4). The judgment condition J described in formula (4) indicates that among the multiple first keywords, there is no keyword with semantic similarity to the i-th second keyword. In other words, the i-th second keyword is a newly appeared keyword on the product display page. Since the user is interested in the first keyword, if the i-th second keyword is not semantically similar to the first keyword, the i-th second keyword may be a keyword that causes the user to have no interest in the product, that is, a candidate keyword that the user is not interested in, and also the second target keyword.
[0046] According to one embodiment of the present invention, the third keyword and its third semantic feature information in the product display page of other products can be obtained in a similar manner to that of the second keyword and the second semantic feature information, which will not be repeated here.
[0047] According to one embodiment of the present invention, the recommendation coefficient of the kth commodity other than the commodity corresponding to the push message may be determined based on formula (5), where: Indicates that if the type information of the kth product belongs to the set composed of the second basic information, the conditional function value is , otherwise it is 1. At this time, if the user is not interested in the product corresponding to the push message, the user may not be interested in the second basic information. Therefore, if the type information of the kth product is the second basic information, the user may also be not interested in the kth product with this type information. The preference score of the kth product is decayed. Otherwise, there is no need to decay the preference score of the kth product. The above conditional function represents the recommendation coefficient corresponding to the type information of the kth product. Similarly, the recommendation coefficient corresponding to the brand information of the kth product can also be obtained. The recommendation coefficient corresponding to the style information of the k-th product , averaging these three recommendation coefficients, we can get the basic information recommendation coefficient of the kth product That is, when the push message of the kth product is not received, the basic information recommendation coefficient of the kth product other than the product corresponding to the push message can be obtained based on the above push message.
[0048] According to one embodiment of the present invention, Indicates that among the second target keywords, there are keywords with semantic similarity to the sth third keyword, that is, the third keyword is semantically similar to a candidate keyword that a user is not interested in. In this case, the conditional function value is , otherwise it is 1. In the case that the third keyword has semantic similarity with a candidate keyword that a user is not interested in, the higher the frequency of the third keyword, The larger the value is, the more likely the user is to be uninterested in the kth product based on the third keyword. and attenuation ratio Multiply them and subtract the product from 1 to obtain the remaining proportion of the preference score after attenuation, It can represent the possibility that the user will feel no interest in the kth product after reading the sth third keyword. Specifically, this item can represent the possibility that the second target keyword with the highest semantic similarity to the sth third keyword will make the user feel no interest. It is used to represent the possibility that the user will feel no interest in the kth product after reading the sth third keyword. This item can be used to continue to attenuate the remaining proportion of the attenuated preference score, so as to accurately describe the user's preference for the characteristics of the product described by the sth third keyword. Furthermore, the user's preference for the characteristics of the products described by each third keyword can be averaged to obtain the content recommendation coefficient of the kth product excluding the product corresponding to the push message. That is, if the push message of the kth product is not received, the content recommendation coefficient of the kth product excluding the product corresponding to the push message can be obtained based on the above push message.
[0049] According to one embodiment of the present invention, the content recommendation coefficient and the basic information recommendation coefficient of the k-th product may be multiplied together to obtain the recommendation coefficient of the k-th product.
[0050] According to one embodiment of the present invention, if the read ratio is greater than or equal to the preset ratio, that is, the user is interested in the product corresponding to the push message, then no other calculations are required and the recommendation coefficients of other products are set to 1, that is, the preference scores of other products are not attenuated.
[0051] In this way, the basic information recommendation coefficients of other products, excluding the product corresponding to the push message, can be determined based on whether the basic information of other products belongs to a set of basic information that the user is not interested in. This allows for obtaining a basic information recommendation score that attenuates the preference score of other products. Furthermore, even without receiving push messages for other products, the basic information recommendation scores of other products can be obtained. Therefore, when a user is not interested in a certain product, the preference scores of similar products can be lowered to reduce the probability of pushing products that the user is not interested in. Furthermore, the content recommendation coefficient can be determined based on word frequency and the occurrence position of the second target keyword with the highest semantic similarity to the third keyword to accurately indicate the likelihood that the user is not interested in other products. Furthermore, even without receiving push messages for other products, the content recommendation scores of other products can be obtained. Therefore, when a user is not interested in a certain product, the preference scores of similar products can be lowered to reduce the probability of pushing products that the user is not interested in, thereby improving the user experience and reducing user usage costs.
[0052] According to one embodiment of the present invention, in step S109, the recommendation coefficient of each product can be multiplied by the preference score of each product to obtain a new preference score, and then the new preference scores are arranged to obtain a new recommendation list. In the new recommendation list, the ranking of products that the user is not interested in can be lowered, and the probability of generating push messages for products with higher rankings in the new recommendation list can be increased, thereby reducing push messages for products that the user is not interested in and improving the user experience.
[0053] According to an embodiment of the present invention, a product recommendation method for an overseas e-commerce platform can determine whether a user has been misled by a push message into a product display page of no interest based on the portion of the page they have read when exiting the product display page. Furthermore, based on a comparison between the information contained in the push message and the information in the portion of the page they have read, the method can determine which products the user is interested in and which they are not interested in. This method can then determine the recommendation coefficient for each product, thereby regenerating the recommendation list and push messages. This reduces the push frequency of push messages for products of no interest to the user, reduces network charges and loading times, lowers user costs, and improves the user experience. When determining the basic information recommendation score, the basic information recommendation coefficient can be determined based on whether three types of basic information are fully displayed in the push message and whether the user is interested in the product. If one or more basic information items are not fully displayed and the user is not interested in the product, a basic information recommendation score can be obtained that can attenuate the preference score in the recommendation list. This reduces the probability of recommending products of no interest to the user and increases the probability of recommending products of interest to the user, thereby improving the user experience and reducing user costs. When determining the content recommendation coefficient, the probability of each second keyword causing the user to be uninterested in the product can be determined by word frequency, and the possibility of each second keyword causing the user to be uninterested in the product can be determined based on the location information of the second keyword. This can obtain a content recommendation coefficient that attenuates the preference score in the recommendation list, thereby reducing the probability of pushing products that the user is not interested in, increasing the probability of pushing products that the user is interested in, improving user experience, and reducing user usage costs. When determining the recommendation coefficient of other products, the basic information recommendation coefficient of other products can be determined based on whether the basic information of other products other than the product corresponding to the push message belongs to the set of basic information that the user is not interested in. This can obtain a basic information recommendation score that attenuates the preference score of other products, and can also obtain the basic information recommendation score of other products even if no push message for other products is received. Therefore, when the user is not interested in a certain product, the preference score of similar products can be reduced to reduce the probability of pushing products that the user is not interested in. In addition, the content recommendation coefficient can be determined by word frequency and the occurrence position of the second target keyword with the highest semantic similarity to the third keyword, so as to accurately indicate the possibility that the user is not interested in other products. In addition, the content recommendation score of other products can be obtained even without receiving push messages for other products. Therefore, when the user is not interested in a certain product, the preference score of similar products can be lowered to reduce the probability of pushing products that the user is not interested in, thereby improving the user experience and reducing the user's usage cost.
[0054] Figure 4A block diagram of a product recommendation system for an overseas e-commerce platform according to an embodiment of the present invention is exemplarily shown. The system includes: The screenshot module is used to take a screenshot of the push message displayed on the screen when the push message is triggered, and obtain a screenshot of the push message; a jump module, configured to jump to a product display page corresponding to the push message, wherein the push message includes summary information of the content on the product display page, and the push message screenshot is an image obtained by taking a screenshot of the push message displayed on a display; Basic information module, used to obtain the type information, brand information and style information of the products on the product display page; a read module, configured to determine a portion of pages that have been read on the product display page when exiting the product display page, wherein the portion of pages that have been read includes a portion of pages that have been displayed on the display during a period between jumping to the product display page and exiting the product display page; A read content information module is used to obtain the read content information of the read part of the page; A basic information recommendation coefficient module, configured to determine a basic information recommendation coefficient based on the push message screenshot, the type information, the brand information, the style information, and the read content information; A content recommendation coefficient module, configured to determine a content recommendation coefficient based on the push message screenshot and the read content information; A recommendation coefficient module, configured to determine recommendation coefficients for a plurality of products based on the basic information recommendation coefficient and the content recommendation coefficient; The push message module is used to obtain a recommendation list of multiple commodities according to the recommendation coefficients of the multiple commodities, and generate a new push message according to the recommendation list.
[0055] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from the principles described.
[0056] Finally, 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A product recommendation method for overseas e-commerce platforms, It is characterized in that include: When a push message is triggered, a screenshot of the push message displayed on the screen is taken to obtain a push message screenshot; Jump to the product display page corresponding to the push message, wherein the push message includes summary information of the content in the product display page, and the push message screenshot is an image obtained by taking a screenshot of the push message displayed on the display; Obtaining the type information, brand information and style information of the product on the product display page; When exiting the product display page, determining a portion of pages that have been read in the product display page, wherein the portion of pages that have been read includes a portion of pages that have been displayed on the display during a period between jumping to the product display page and exiting the product display page; Obtaining information about the read content in the read part of the page; Determine a basic information recommendation coefficient according to the push message screenshot, the type information, the brand information, the style information and the read content information; Determining a content recommendation coefficient according to the push message screenshot and the read content information; Determining recommendation coefficients of a plurality of commodities according to the basic information recommendation coefficient and the content recommendation coefficient; A recommendation list of the multiple commodities is obtained according to the recommendation coefficients of the multiple commodities, and a new push message is generated according to the recommendation list.
2. The product recommendation method for overseas e-commerce platforms according to claim 1, It is characterized in that Determining a basic information recommendation coefficient according to the push message screenshot, the type information, the brand information, the style information, and the read content information includes: Recognize the text in the push message screenshot using a text recognition model to obtain the text content in the push message screenshot; Recognize the text content to obtain first basic information that can be fully displayed in the push message screenshot and second basic information that cannot be fully displayed among the product type information, brand information, and style information; Determine, according to the read content information, a read ratio of the read content information in all contents on the product display page; A basic information recommendation coefficient is determined according to the first basic information, the second basic information, the type information, the brand information, the style information and the read ratio.
3. The product recommendation method for overseas e-commerce platforms according to claim 2, It is characterized in that Determining a basic information recommendation coefficient according to the first basic information, the second basic information, the type information, the brand information, the style information, and the read percentage includes: According to the formula Determine the basic information recommendation coefficient ,in, For the type information, For the brand information, For the style information, is a set of the first basic information, is the reading percentage, is the preset ratio, is the conditional function.
4. The product recommendation method for overseas e-commerce platforms according to claim 2, It is characterized in that Determining a content recommendation coefficient according to the push message screenshot and the read content information includes: Recognize the text in the push message screenshot using a text recognition model to obtain the text content in the push message screenshot; Performing word segmentation and deduplication processing on the text content to obtain multiple first keywords of the text content; By using the text recognition model, feature extraction is performed on the first keywords to obtain first semantic feature information of each first keyword; Determine, according to the read content information, a read ratio of the read content information in all contents on the product display page; Performing word segmentation and deduplication processing on the read content information to obtain a plurality of second keywords of the read content information; Determine the first occurrence position information of the second keyword in all the content on the product display page; By using the text recognition model, feature extraction is performed on the second keywords to obtain second semantic feature information of each second keyword; A content recommendation coefficient is determined according to the first semantic feature information, the second semantic feature information, the appearance position information, and the read ratio.
5. The product recommendation method for overseas e-commerce platforms according to claim 4, It is characterized in that Determining a content recommendation coefficient according to the first semantic feature information, the second semantic feature information, and the read percentage includes: According to the formula Determine content recommendation coefficient ,in, is the second semantic feature information of the i-th second keyword, is the first semantic feature information of the jth first keyword, for The transposed vector of for and The cosine similarity of is the preset similarity threshold, max is the maximum value function, m is the number of first keywords, is the conditional function, is the reading percentage, is the preset ratio, is the first occurrence position information of the i-th second keyword in all the content on the product display page, is the frequency of the i-th second keyword in the read content information, is the average frequency of multiple second keywords, n is the number of first keywords, j≤m, i≤n, and i, j, m and n are all positive integers.
6. The product recommendation method for overseas e-commerce platforms according to claim 4, It is characterized in that Determining recommendation coefficients of a plurality of commodities according to the basic information recommendation coefficient and the content recommendation coefficient includes: Determining a recommendation coefficient of the product corresponding to the push message according to the basic information recommendation coefficient and the content recommendation coefficient; If the read ratio is less than the preset ratio, obtaining the recommendation coefficient of the other products except the product corresponding to the push message according to the type information, brand information, style information and content in the product display page of the other products except the product corresponding to the push message, as well as the second basic information and the second keyword; If the read ratio is greater than or equal to the preset ratio, the recommendation coefficients of the other commodities except the commodity corresponding to the push message are set to 1.
7. The product recommendation method for overseas e-commerce platforms according to claim 6, It is characterized in that Determining a recommendation coefficient of a product corresponding to the push message according to the basic information recommendation coefficient and the content recommendation coefficient includes: According to the formula Determine the recommendation coefficient of the product corresponding to the push message ,in, is the content recommendation coefficient, is the basic information recommendation coefficient.
8. The product recommendation method for overseas e-commerce platforms according to claim 6, It is characterized in that If the read ratio is less than the preset ratio, obtaining the recommendation coefficient of the other products except the product corresponding to the push message according to the type information, brand information, style information and content in the product display page of the other products except the product corresponding to the push message, as well as the second basic information and the second keyword, including: According to the formula Get the judgment condition J, where is the second semantic feature information of the i-th second keyword, is the first semantic feature information of the jth first keyword, for The transposed vector of for and The cosine similarity of is the preset similarity threshold, max is the maximum value function, and m is the number of first keywords; If the second semantic feature information of the i-th second keyword meets the discrimination condition J, the i-th second keyword is determined as the second target keyword; Perform word segmentation and deduplication processing on the content of product display pages of other products except the product corresponding to the push message to obtain multiple third keywords; By using the text recognition model, feature extraction is performed on the third keywords to obtain third semantic feature information of each third keyword; According to the formula Get the recommendation coefficient of the kth product other than the product corresponding to the push message ,in, is a set of the second basic information, is the reading percentage, is the preset ratio, is the conditional function, is the type information of the kth product, is the brand information of the kth product, is the style information of the kth product, is the second semantic feature information of the t-th second target keyword, is the number of the second target keywords, is the third semantic feature information of the sth third keyword, for The transposed vector of for and The cosine similarity of is the preset similarity threshold, max is the maximum value function, is the frequency of the sth third keyword in the content of the product display page of the kth product, is the average frequency of multiple third keywords in the content of the product display page of the kth product, is the number of third keywords in the content of the product display page of the kth product, To make The location information of the first appearance of the second target keyword with the maximum value in all the content on the product display page, s≤ , t≤ , and s, , t, k, and All are positive integers.
9. A product recommendation system for overseas e-commerce platforms. It is characterized in that include: A screenshot module is used to take a screenshot of the push message displayed on the screen when the push message is triggered, and obtain a screenshot of the push message; A jump module, used to jump to a product display page corresponding to the push message, wherein the push message includes summary information of the content in the product display page, and the push message screenshot is an image obtained by taking a screenshot of the push message displayed on the display; The basic information module is used to obtain the type information, brand information and style information of the products on the product display page; A read module, used for determining a portion of pages that have been read in the product display page when exiting the product display page, wherein the portion of pages that have been read includes a portion of pages that have been displayed on the display during a period between jumping to the product display page and exiting the product display page; A read content information module is used to obtain the read content information in the read part of the page; A basic information recommendation coefficient module, used to determine a basic information recommendation coefficient according to the push message screenshot, the type information, the brand information, the style information and the read content information; A content recommendation coefficient module, used to determine a content recommendation coefficient according to the push message screenshot and the read content information; A recommendation coefficient module, used to determine the recommendation coefficients of multiple commodities according to the basic information recommendation coefficient and the content recommendation coefficient; The push message module is used to obtain a recommendation list of multiple commodities according to the recommendation coefficients of the multiple commodities, and generate a new push message according to the recommendation list.
Citation Information
Patent Citations
Commodity recommendation method and commodity recommendation device, equipment and storage medium
CN108182621A
Commodity recommendation method and device, electronic equipment and readable medium
CN110415063A
Data acquisition and recommendation method based on shopping behaviors
CN113674037A
E-commerce user behavior analysis method and system based on data mining
CN118134608A
Information recommendation method and apparatus, computer device, and storage medium
WO2023169112A1
