WO2022151923A1 - 一种商品文案处理方法、装置、电子设备、介质和程序 - Google Patents

一种商品文案处理方法、装置、电子设备、介质和程序 Download PDF

Info

Publication number
WO2022151923A1
WO2022151923A1 PCT/CN2021/139717 CN2021139717W WO2022151923A1 WO 2022151923 A1 WO2022151923 A1 WO 2022151923A1 CN 2021139717 W CN2021139717 W CN 2021139717W WO 2022151923 A1 WO2022151923 A1 WO 2022151923A1
Authority
WO
WIPO (PCT)
Prior art keywords
commodity
information
user
copy
candidate
Prior art date
Application number
PCT/CN2021/139717
Other languages
English (en)
French (fr)
Inventor
张秀军
Original Assignee
北京沃东天骏信息技术有限公司
北京京东世纪贸易有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京沃东天骏信息技术有限公司, 北京京东世纪贸易有限公司 filed Critical 北京沃东天骏信息技术有限公司
Publication of WO2022151923A1 publication Critical patent/WO2022151923A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0623—Item investigation
    • G06Q30/0625—Directed, with specific intent or strategy
    • G06Q30/0627—Directed, with specific intent or strategy using item specifications
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/903—Querying
    • G06F16/90335—Query processing
    • G06F16/90344—Query processing by using string matching techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0207—Discounts or incentives, e.g. coupons or rebates
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Definitions

  • the present application relates to the field of Internet technology, and relates to, but is not limited to, a method, apparatus, electronic device, computer storage medium and computer program product for processing commodity text.
  • platform operation will reach users through push message (PUSH) to promote product promotion activities and promote new activities; however, in related technologies, the content of messages pushed to users is the content of the operator through the content management system (Content Management System). , CMS) manually configured in the background, and the reached user groups are full users; in this way, not only the generation efficiency of message content is low, but also the message content reaching users is too single, reducing the accuracy of message reaching.
  • PUSH push message
  • CMS Content Management System
  • the present application provides a commodity copy processing method, device, electronic device, computer storage medium and computer program product, which can solve the problem of too single copy information reaching users, that is, triggering different copy information for different users.
  • the embodiment of the present application provides a method for processing commodity text, the method comprising:
  • the historical behavior data includes data related to commodities
  • a candidate product set is determined among the multiple products; based on the copy information of the multiple products and the candidate product set, the copy information of each product in the candidate product set is generated. ;
  • the method further includes:
  • generating the copywriting information of each commodity in the candidate commodity set based on the copywriting information of the multiple commodities and the candidate commodity set includes:
  • the purchase benefit information and the initial copy information are spliced to generate copy information of each commodity in the candidate commodity set.
  • the purchase preference information includes at least one of the following: commodity category preference, commodity manufacturer identification preference, commodity repurchase information.
  • the commodity repurchase information includes the repurchased commodity and the commodity repurchase cycle, and the method further includes:
  • the push time of the copywriting information of the repurchased commodity is determined; the at least one commodity includes the repurchased commodity.
  • the pushing the copywriting information of at least one commodity in the candidate commodity set to the terminal of the first user includes:
  • the at least one commodity is selected from the candidate commodity set, and the copy information of the at least one commodity is pushed to the terminal of the first user.
  • the estimation model is obtained by training the following steps:
  • sample data includes: the copy feature of the copy information and the click probability label value
  • the prediction model is trained by using the sample data, and the trained prediction model is obtained.
  • the method further includes:
  • the attribute information includes at least one of the following: a hot-selling score, a hot-selling score, and a purchase score;
  • the hot-selling score represents the commodity related to the region where the second user is located.
  • the popularity score represents the popularity of the product related to the region where the second user is located;
  • the purchase score represents the purchase demand of the product related to the second user; the second user is related to the first user different users of users;
  • the click score value represents the click probability of the copy information of each commodity
  • Sort a plurality of commodities in the area where the second user is located according to the click score value, and obtain a second sorting result; based on the second sorting result, select a target commodity set from the candidate commodity set, and send it to the The terminal of the second user pushes the copy information of the target commodity set, where the target commodity set represents one or more commodities in the candidate commodity set.
  • the method further includes:
  • the first index information of each commodity is processed to obtain the hot-selling score of each commodity.
  • the method further includes:
  • the second index information of each commodity is processed to obtain the popularity score of each commodity.
  • the method further includes:
  • the characteristic information includes information of at least one dimension related to the users
  • the attribute information includes: a hot sale score, a popularity score and a purchase score; accordingly, the attribute information of each commodity in the candidate commodity set is processed to obtain the clicked score of each commodity.
  • Points including:
  • a weighted summation is performed on the hot sale score, popularity score and purchase score of each commodity in the candidate commodity set to obtain the click score value of each commodity.
  • the embodiment of the present application also proposes a commodity copy processing device, the device includes an acquisition module, a determination module, a generation module and a push module, wherein,
  • an acquisition device configured to acquire copywriting information of multiple commodities
  • a determination module configured to determine the purchase preference information of the first user according to the historical behavior data of the first user; the historical behavior data includes data related to commodities;
  • the generating module is configured to determine a candidate commodity set among the plurality of commodities according to the purchase preference information of the first user; The textual information of each product;
  • a push module configured to push the copy information of at least one commodity in the candidate commodity set to the terminal of the first user.
  • An embodiment of the present application provides an electronic device, the device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements one or more of the foregoing techniques when executing the program The product copy processing method provided by the program.
  • An embodiment of the present application provides a computer storage medium, where a computer program is stored in the computer storage medium; after the computer program is executed, the commodity copy processing method provided by one or more of the foregoing technical solutions can be implemented.
  • Embodiments of the present application further provide a computer program product, including computer-readable codes, when the computer-readable codes are executed in an electronic device, the processor in the electronic device executes the code for implementing one or more of the foregoing The commodity copy processing method provided by the technical solution.
  • the embodiments of the present application propose a method, device, electronic device, computer storage medium, and computer program product for processing commodity texts.
  • the method includes: acquiring textual information of multiple commodities; determining the first user according to historical behavior data of the first user Purchase preference information of the user; the historical behavior data includes data related to commodities; according to the purchase preference information of the first user, a candidate commodity set is determined among the plurality of commodities; based on the copywriting information of the plurality of commodities and the candidate product set to generate copy information for each product in the candidate product set; push the copy information of at least one product in the candidate product set to the terminal of the first user; in this way, there is no need to manually configure the copy text of the product through the operator Instead, the copy information of the product is generated according to the user's purchase preference information, which can improve the generation efficiency of the copy information; The preference information generates part of the copy information and pushes it, which can solve the problem that the copy information reaching the user is too simple, and improve the accuracy of the message reaching.
  • Fig. 1 is a schematic diagram of a result of influencing the push of a message to reach a user in the related art
  • FIG. 2 is a schematic structural diagram of a product copy push framework in an embodiment of the present application.
  • 3a is a schematic diagram 1 of a strong touch scenario in an embodiment of the present application.
  • 3b is a second schematic diagram of a strong touch scenario in an embodiment of the present application.
  • FIG. 3c is a schematic diagram 3 of a strong touch scenario in an embodiment of the present application.
  • 4a is a schematic diagram of two kinds of message contents pushed by taking coupons as an example in an embodiment of the application;
  • Fig. 4b is a schematic diagram of two kinds of message contents to be pushed by taking a promotional activity as an example in an embodiment of the application;
  • FIG. 5 is a schematic flowchart of a method for processing product copy in an embodiment of the present application.
  • FIG. 6 is a schematic diagram of the results of obtaining four different types of copywriting information in the embodiment of the application.
  • FIG. 7 is a schematic diagram illustrating a repurchase cycle of repurchased commodities in an embodiment of the application.
  • FIG. 8 is a schematic diagram of a result of generating copywriting information of a commodity in an embodiment of the application.
  • FIG. 9 is a schematic diagram of the result of the user historical behavior sequence log obtained in the embodiment of the application.
  • FIG. 10 is a schematic structural diagram of screening the copywriting information of a commodity by a click-through rate (Click-Through-Rat, CTR) estimation model in an embodiment of the application;
  • FIG. 11 is a schematic diagram of the composition and structure of a commodity copywriting processing device according to an embodiment of the application.
  • FIG. 12 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
  • a method or device including a series of elements not only includes the explicitly stated elements, but also other elements not expressly listed or inherent to the practice of the method or apparatus.
  • an element defined by the phrase “comprises a" does not preclude the presence of additional related elements (eg, steps in a method or a device) in which the element is included.
  • a unit in an apparatus for example, a unit may be part of a circuit, part of a processor, part of a program or software, etc.).
  • the commodity text processing method provided by the embodiment of the present application includes a series of steps, but the commodity text processing method provided by the embodiment of the present application is not limited to the described steps.
  • the commodity text processing device provided by the embodiment of the present application A series of modules are included, but the commodity copy processing device provided by the embodiment of the present application is not limited to including the modules explicitly described, and may also include modules that need to be set for obtaining relevant data or processing based on relevant data.
  • the embodiments of the present application can be applied to a computer system composed of a terminal device and a server, and can operate with many other general-purpose or special-purpose computing system environments or configurations.
  • the terminal devices may be thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, minicomputer systems, etc.
  • the server may be a server Computer Systems Small computer systems, large computer systems, and distributed cloud computing technology environments including any of the above, etc.
  • Electronic devices such as terminal devices, servers, etc., may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system.
  • program modules may include routines, programs, object programs, components, logic, data structures, etc. that perform particular tasks or implement particular abstract data types.
  • Computer systems/servers may be implemented in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network.
  • program modules may be located on local or remote computing system storage media including storage devices.
  • Push refers to the active push of messages to users' mobile devices by operators through their own products or third-party tools.
  • the user can see the PUSH message notification on the lock screen and notification bar of the mobile device. Clicking on the notification bar can evoke the mobile phone software (Application, APP) and go to the corresponding page.
  • PUSH mobile phone software
  • FIG. 1 is a schematic diagram of the results of related technologies that affect the reach of users by push messages.
  • User factors include three: delivery rate, display rate and click rate; among them, the factors affecting the delivery rate include: channel quality, APP online rate, message validity period and token validity; here, channel quality means The pros and cons of the communication channel used to transmit the message content; the validity period of the message indicates the validity period of the message content before it is received by the terminal; the validity period of the Token is generated when the user requests authentication from the server, and is used to indicate the validity of the request. The time, it can be determined through the software development kit (Software Development Kit, SDK) log report.
  • Software Development Kit, SDK Software Development Kit
  • the factor affecting the display rate is the user's disabling situation; the user's disabling situation means that the user has not opened the PUSH channel for receiving the message content.
  • Factors that affect click-through rate include: title, summary, whether the image is engaging and whether it hits the user's interest.
  • FIG. 2 is a schematic structural diagram of a product copy push framework in an embodiment of the present application.
  • the basic service building part includes a message push channel;
  • the push channel can receive messages reported by devices.
  • the PUSH channel includes: Android and IOS; Table 1 shows the delivery rate and open rate of PUSH messages of different device types in the related art; it can be seen that the delivery rate of PUSH messages on the Android channel is only 25.15%, therefore, it is necessary to The Android channel is opened to improve the delivery rate of PUSH messages.
  • the recall part at the functional level includes two parts: click to send log data and send service; among them, the function of click to send log data is to identify the user's intention by reporting the log and perform data mining, which can achieve the accuracy of the next personalized touch. sex.
  • the open rate of Android channel statistics is 0. After data analysis, it can be determined that the SDK version is too low. Therefore, the above problem can be solved by upgrading the PUSH SDK.
  • the delivery service is a strategy for detecting and guiding the opening of PUSH channel closures established in the scenario of strong user access.
  • User opening the PUSH channel is a precondition for subsequent personalized access.
  • 3a is a schematic diagram 1 of a strong touch scenario in an embodiment of the application
  • FIG. 3b is a schematic diagram 2 of a strong touch scenario in an embodiment of the application
  • FIG. 3c is a schematic diagram 3 of a strong touch scenario in an embodiment of the application, as shown in FIG.
  • strong reach scenarios can include: live broadcast reservation, order completion, and start of seckill activities.
  • Through the delivery service it can automatically detect whether the user has closed the PUSH channel when different strong touch scenarios are completed. If it is closed, a pop-up window will be given to guide the user to open the PUSH channel.
  • FIG. 4a is a schematic diagram of two kinds of message contents pushed by taking coupons as an example in an embodiment of the present application.
  • the message contents pushed to the user terminal may be "You have a batch of coupons, please check! is converted into a personalized message content such as "Giving you 130-5 coupons, go buy Lay's Potato Chips Cucumber Flavor 40g".
  • FIG. 4b is a schematic diagram of two kinds of message contents to be pushed by taking the promotion activity as an example in the embodiment of the application. As shown in FIG.
  • the message content “new rights and interests” pushed to the user terminal can be When it arrives, please pay attention to check it! is converted into "Congratulations on getting the 99 free shipping privilege, the original Lay's Potato Chips 40g you often buy is also on special price, go and buy it"
  • This personalized message content. 4a and 4b it can be seen from the comparison of the two message contents that the message content of personalized touch includes the user's purchase preference and specific purchase benefits, which can improve the user's click-through rate on the message content.
  • manual operation can be replaced by system recommendation, and when the message push reaches the user, the user's perception can be enhanced, more benefit points can be revealed, and the user can be guided to click on consumption.
  • the commodity text processing method may be implemented by using a processor in the commodity text processing device, and the above-mentioned processor may be an application specific integrated circuit (ASIC), a digital signal processor (Digital Signal Processor) Processor, DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (Programmable Logic Device, PLD), Field Programmable Gate Array (Field Programmable Gate Array, FPGA), Central Processing Unit (Central At least one of Processing Unit, CPU), controller, microcontroller, microprocessor.
  • ASIC application specific integrated circuit
  • DSP digital signal processor
  • DSPD Digital Signal Processing Device
  • PLD Programmable Logic Device
  • Field Programmable Gate Array Field Programmable Gate Array
  • FPGA Field Programmable Gate Array
  • CPU Central Processing Unit
  • FIG. 5 is a schematic flowchart of a method for processing product copy in an embodiment of the present application. As shown in FIG. 5 , the method includes the following steps:
  • Step 100 Acquire copywriting information of multiple commodities.
  • the commodity may represent any type of item that the seller on the e-commerce platform trades through the Internet; for example, it may be a clothing item, a food item, etc., or a virtual item, etc.; limit.
  • the original title copy of each product on the e-commerce platform can be obtained, and the original title copy of each product can be extracted to obtain key information.
  • the analysis combination of you can get the copy information of each product.
  • the original title copy of the product will list multiple words to cover a large amount of information.
  • the channel is pushed to the user, making it difficult for the user to identify the key content in a short time; in order to solve the above problem, the original title copy of each product can be extracted, and according to the extracted key information, differentiated copy information can be generated to make The copy information can grasp the main body of the product and increase the user's recognition.
  • the extracted key information can be determined according to the actual application scenario, and can be attributes of the product itself, such as product word, color, size, or weight.
  • different types of copy information of the product can be obtained, and the types of copy information may include: short title copy, enhanced copy, heart copy or highlight copy; wherein, under different marketing scenarios , you can select different types of information copy to push.
  • Fig. 6 is a schematic diagram of the results of obtaining four different types of copywriting information in the embodiment of the application.
  • the original title of the men's backpack can be The text "SWISSGEAR Fashion Backpack 14.6-inch Computer Bag Men's Business Backpack Apple Notebook Bag Leisure Travel Bag Student Schoolbag SA-9911 Black” can get four different types of copy information, which are the short title copy "SWISSGEAR Student Casual Fashion Computer Bag", Enhanced copy "What kind of bags are durable and beautiful? These computer bags are right!, the heart copy "Men's casual backpack makes travel easier" and the highlight copy "Water repellent fabric, no fear of rain”.
  • the short title copy can be combined according to the manufacturer's logo of the product, the product word and the extracted key information of the product; the enhanced copy is more slogan, the description of the heart copy is more subjective, and the description of the highlight copy is more objective.
  • the copy information of each commodity on the e-commerce platform may be acquired, and the manner of acquiring the above four types of copy information is not limited in the embodiment of this application.
  • the relevant copy generation algorithm may be used.
  • the short title text is used as an example to describe the text information of the commodity.
  • Step 101 Determine the purchase preference information of the first user according to the historical behavior data of the first user; the historical behavior data includes data related to commodities.
  • the first user refers to any user who has historical behavior data on the e-commerce platform.
  • the historical behavior data may include historical search data, historical browsing data, historical shopping cart data, and historical ordering data, etc. .
  • the historical behavior data of the first user may be data mined through preset calculation rules to determine the purchase preference information of the first user; wherein the purchase preference information may include at least one of the following: commodity category Preferences, product manufacturer identification preferences, product repurchase information.
  • the historical behavior data of the first user may be the historical behavior data of the first user on the e-commerce platform in the past one year, or the historical behavior data in the past six months;
  • the time period is not limited.
  • the calculation rule for determining the commodity category preference of the first user may be: first, according to the historical behavior data of the first user, obtain an order placed by the first user within a set time period and the order status is Data of all completed orders; according to the data of all orders, the following four user dimension data are determined, which are: sku width of first-level category orders (daily deduplication), sku width of all orders, and total transaction volume of first-level category orders (Gross Merchandise Volume, GMV), the total GMV of all orders. Then, according to formulas (1) and (2), respectively calculate the scores of the two data indicators, Score_widths_rate and Score_amount_rate, of the first user in all first-level categories:
  • Score_widths_rate sku width of first-level category order/sku width of all orders (1)
  • Score_amount_rate cumulative GMV of the first-level category/total GMV of all orders (2)
  • Score_widths_rate represents the sku width ratio of first-level category orders
  • Score_amount_rate represents the proportion of first-level category GMV.
  • the data logic of the width of the first-level category sku is shown in Table 2, including two fields, namely the user ID and the first-level category information; wherein, the first-level category information includes: the first-level category name, the first-level category information Category ID and first-level category order sku width.
  • the output label of the first-level category is determined.
  • all first-level categories and corresponding label scores can be determined from all the order data placed by the first user in the past six months and the order status is completed; Counting the duration; further, you can sort the tag scores in descending order to get the sorting result, and then use the first-level category corresponding to the tag score ranked first in the sorting result as the commodity category of the first user preference.
  • an update period may be set for the calculation of the commodity category preference of the first user, so that the commodity category preference of the first user is updated every preset time; Preferences may change, and re-determining the corresponding commodity category preferences according to the order data continuously generated by the user can ensure the accuracy of the user's commodity category preferences.
  • the update period may be one month, half a year, etc., which may be set according to the actual situation, which is not limited in this embodiment of the present application.
  • the following four categories can be determined according to the historical behavior data of the first user, namely: clothing, accessories, beauty makeup and snacks, and their corresponding label scores are: 0.5, 0.2 , 0.1, 0.8; then snacks can be used as the commodity category preference of the first user, and snacks and clothing can also be used as commodity category preferences of the first user.
  • the above-mentioned first category may also be a second-level category or a third-level category, which may be specifically set according to actual application scenarios.
  • the calculation rule for determining the preference of the first user's product manufacturer identifier is the same as the above-mentioned calculation rule for determining the preference of the product category, and it is only necessary to replace the first category with the manufacturer's identifier, which is not repeated here. burden.
  • the commodity repurchase information includes: the repurchased commodity and the commodity repurchase cycle; the repurchased commodity represents the commodity for which the first user has repurchased behavior; the calculation rule for the commodity repurchase cycle of the first user may be: : According to the historical behavior data of the first user, obtain all the order data that the first user placed an order within the set time period and the order status is completed. (4) Calculate the repurchase cycle for the first user to purchase the unit repurchased goods:
  • t is an integer greater than zero
  • P t represents the repurchase cycle of the first user’s t-th repurchased product
  • T t+1 represents the time when the first user purchased the repurchased product for the t+1th time
  • T t represents the time when the first user purchased the repurchased product for the tth time
  • Nt represents the number of repurchased products purchased by the first user at the tth time
  • the time for purchasing repurchased products is in days. If the first user purchases multiple orders on the same day, the number of repurchased products is accumulated.
  • the repurchase cycle of the first user may be different; that is, the first user may repurchase some repurchased commodities more than once within a set period of time;
  • FIG. 7 shows the embodiment of the present application.
  • the time at which the product was purchased that is, the historical order data of the first user; combined with the above formula (4), the first user’s purchase unit in the time intervals T1-T2, T2-T3, T3-T4, and T4-T5 are calculated respectively.
  • Repurchase cycle P t of purchased commodities sort these repurchase cycles in ascending order, and take the median as the repurchase cycle of repurchased commodities by the first user purchasing unit.
  • the repurchase cycles of the first user's purchase unit repurchasing commodity A are determined according to formula (4) as: 10 days, 15 days and 18 days, respectively, then 15 days are used as the first user's purchase unit repurchase cycle.
  • the repurchase cycle of the purchased product A is assumed that the repurchase cycles of the first user's purchase unit repurchasing commodity A are determined according to formula (4) as: 10 days, 15 days and 18 days, respectively, then 15 days are used as the first user's purchase unit repurchase cycle.
  • the repurchase cycle of the purchased product A is assumed that the repurchase cycles of the first user's purchase unit repurchasing commodity A are determined according to formula (4) as: 10 days, 15 days and 18 days, respectively, then 15 days are used as the first user's purchase unit repurchase cycle.
  • the repurchase cycle of the purchased product A is assumed that the repurchase cycles of the first user's purchase unit repurchasing commodity A are determined according to formula (4) as: 10 days, 15 days and 18 days, respectively
  • the update cycle can be set to every preset time.
  • the repurchase cycle of the repurchased commodity purchased by the first user is updated.
  • the update cycle can be determined according to the type of repurchased commodities. For example, for repurchased commodities such as refrigerators or computers, the update cycle can be one year or two years, etc.; for repurchased commodities such as snacks, the update cycle can be half a year. month or month etc.
  • the above method may further include: determining the purchase benefit information of the first user according to the historical behavior data of the first user.
  • the purchase benefit information can be determined according to different marketing scenarios.
  • the purchase benefit information can be coupons, etc.
  • the purchase benefit information in the scenario of acquiring new users, can be shipping coupons, etc.;
  • the purchase benefit information is described as an example.
  • the coupon includes a coupon threshold value and a rebate value; wherein, the coupon threshold is defined according to the unit price of the product purchased by the first user, because if the coupon threshold value is defined too high, then The daily consumption of the first user does not reach the threshold, and further, it is difficult to recall the user for consumption through the coupon; if the threshold value of the coupon is defined too low, it is not conducive to maximizing the interests of the e-commerce platform. Due to the different unit prices of products purchased by different users, the threshold values of coupons corresponding to different users are different.
  • the calculation rule for the purchase benefit information of the first user may be: first, according to the historical behavior data of the first user, obtain the transaction amount of each order data within the set time period of the first user (order Status has been paid, cash on delivery has been completed) and take the median; then, use formula (5) to calculate the coupon threshold value corresponding to the first user:
  • Coupon threshold median transaction amount * 1.1 (5)
  • the coupon threshold value is a decimal
  • the decimal places are removed and then rounded up, and the non-integer rounds are rounded up.
  • the profit margin value After obtaining the coupon threshold value corresponding to the first user, determine the profit margin value; exemplarily, by analyzing the historical behavior data of the first user and the gross profit data of purchases and sales, the deduction or exemption value of the profit margin obtained is that the coupon threshold value is lower than
  • the discount value of 300 yuan is 5 yuan
  • the coupon threshold value is between 300 and 400 yuan
  • the discount value is 10 yuan
  • the coupon threshold value is between 500 and 600 yuan
  • the discount value is 20 yuan
  • the coupon threshold value is 20 yuan.
  • the value of the profit is 25 yuan
  • the value of the coupon is 30 yuan if the threshold value of the coupon is higher than 1,000 yuan.
  • the coupon is set as the coupon threshold value - profit, that is, the coupon of 350-10.
  • the coupons include three types: platform coupons, which can be used to purchase all commodities on the e-commerce platform; brand coupons, which can be used to purchase designated brand products; category coupons, which can be used to purchase designated categories of products.
  • platform coupons which can be used to purchase all commodities on the e-commerce platform
  • brand coupons which can be used to purchase designated brand products
  • category coupons which can be used to purchase designated categories of products.
  • the cost of platform coupons is borne by the e-commerce platform
  • the cost of brand coupons and category coupons is borne by brand owners.
  • brand coupons and category coupons can be issued preferentially, so the first user's preference for commodity categories and commodity manufacturer identification preferences will be combined when the message is reached.
  • a coupon is issued under the condition that the purchase preference information of the first user is determined and the platform benefits are maximized. If there is no centralized trend in the preference of the first user's commodity category and commodity manufacturer identification, a platform coupon will be issued.
  • Step 102 According to the purchase preference information of the first user, a candidate commodity set is determined among the plurality of commodities; based on the copy information of the multiple commodities and the candidate commodity set, the copy information of each commodity in the candidate commodity set is generated.
  • the purchase preference information may include at least one of the following: commodity category preference, commodity manufacturer identification preference, and commodity repurchase information; therefore, all commodities involved in the commodity category preference of the first user may be included In the candidate commodity set, all commodities involved in the first user's commodity manufacturer identification preference may also be placed in the candidate commodity set, and the repurchased commodities of the first user may also be placed in the candidate commodity set.
  • the following judgment operation needs to be performed: it is necessary to judge the next purchase of the repurchased product according to the product of the quantity of the repurchased product purchased last time and the repurchase cycle Whether the time interval from the current purchase is less than the set value, if the time interval is less than the set value, put the repurchased product into the candidate product set; if the time interval is greater than or equal to the set value, no longer Put the repurchased products into the candidate product set.
  • the set value may be set according to the actual situation, which is not limited in this embodiment of the present application; for example, it may be one day, three days, or five days, and so on.
  • the candidate commodity set includes some commodities on the e-commerce platform; before the candidate commodity set is determined, the copy information corresponding to all commodities on the e-commerce platform has been generated; therefore, after the candidate commodity set is determined , the copy information corresponding to each item in the candidate item set can be found according to the correspondence between the item and the copy information.
  • generating the copy information of each item in the candidate item set based on the copy information of the multiple items and the candidate item set may include: generating the candidate item set based on the copy information of the multiple items and the candidate item set The initial copy information of each product in the candidate product set; splicing the purchase benefit information and the initial copy information to generate the copy information of each product in the candidate product set.
  • the copy information corresponding to each item in the candidate item set found according to the above-mentioned correspondence may be used as the initial copy information, for example, the short title copy of each item may be used as the initial copy information;
  • the purchase benefit information is spliced with the initial copy information of each product, for example, the coupon corresponding to the first user is spliced with the short title copy of each product to generate copy information for each product.
  • the first user's purchase benefit information is a coupon of 130-5 determined in step 101
  • the first user's preference for commodity category is: potato chips
  • the preference for commodity manufacturer identification is: XX
  • the The purchased product is: "XX potato chips 40g office leisure puffed zero food snack cucumber flavor production date August”; through step 100, the original title of the product "XX potato chips 40g office leisure puffed zero food snack cucumber flavor production date August” is carried out.
  • the copy template may be pre-configured in the e-commerce platform, and the type of the copy template may be self-defined according to the actual scene, which is not limited in this embodiment of the present application.
  • the coupon copy template is: give you XXX-X coupons, go and buy XXX short title copy.
  • a 5 yuan coupon without threshold will be triggered, and the corresponding copy template is: give you a 5 yuan coupon without threshold, go buy XXX short title copy.
  • Step 103 Push the copy information of at least one commodity in the candidate commodity set to the terminal of the first user.
  • the corresponding copy information can be determined for each commodity in the candidate commodity set; however, when pushing these copy messages to the terminal of the first user, if all the copy information of these products is Pushing will disturb the first user, and at the same time, it is likely to cause the first user to close the message push channel. In this way, even if the copy information of the product generated subsequently is pushed to the terminal of the first user, it will not be read by the first user. User clicks.
  • pushing the copy information of at least one item in the candidate item set to the terminal of the first user may include: extracting copy features of the copy information of each item in the candidate item set; using a trained estimation model Process the copy features of the copy information of each product to obtain the click probability of the copy information of each product; sort the copy information of each product according to the click probability to obtain the first sorting result; At least one commodity is selected from the set, and the copy information of the at least one commodity is pushed to the terminal of the first user.
  • the click probability can be sorted in descending order to obtain a sorting result; and the click probability in the sorting result is arranged first.
  • the copy information of at least one commodity is pushed to the terminal of the first user.
  • the copy information of the candidate products of the first user are A1, A2 and A3 respectively, and the click probabilities corresponding to the three copy information are 0.8, 0.2 and 0.5 in sequence, and the sorting result after sorting is A1, A3 and A2, the copy information A1 may be pushed to the terminal of the first user, and the copy information A1 and A3 may also be pushed to the terminal of the first user.
  • the above method further includes: when at least one commodity includes a repurchased commodity, determining the push time of the copywriting information of the repurchased commodity according to the commodity repurchase information.
  • the time for the next purchase of the repurchased product may be determined according to the number of the repurchased product last purchased by the first user from the current moment; and then , the time when the repurchased product is purchased next time can be used as the push time of the copywriting information of the repurchased product, so that the click rate of the user can be further improved.
  • the repurchased product of the first user is a toothbrush
  • the repurchase cycle of a single toothbrush obtained according to step 101 is 30 days, and it is determined that the number of toothbrushes purchased by the first user last time is 2, Then it can be determined that the next time the first user buys the toothbrush is 60 days after the last purchase, and further, the copy information of the toothbrush can be pushed to the first user 60 days after the last purchase.
  • the above-mentioned estimation model is obtained by training through the following steps: obtaining sample data; the sample data includes: the copy feature of the copy information and the click probability label value; the estimation model is trained through the sample data, and the training is obtained.
  • the completed estimation model is obtained by training through the following steps: obtaining sample data; the sample data includes: the copy feature of the copy information and the click probability label value; the estimation model is trained through the sample data, and the training is obtained. The completed estimation model.
  • the training of the prediction model is supervised learning, that is, there is an actual value Y corresponding to the input X; here, the input X represents the copy feature of each copy information, and the actual value Y represents the value of each copy information.
  • the loss function between the input X of the prediction model and the actual value Y is the network backpropagation, and the training process of the entire neural network is the process of continuously reducing the value of the loss function.
  • the copy features of the copy information may include: the user's product category preference, the user's product manufacturer identification preference, the customer unit price, the historical behavior sequence log, the product product word, the product category, the product manufacturer identification, and the product price.
  • FIG. 9 is a schematic diagram of the result of the user historical behavior sequence log obtained in the embodiment of the application.
  • the historical behavior sequence log is sorted in chronological order from top to bottom. It can be seen from the user historical behavior sequence log that the user There is repeated browsing behavior for each product. After multiple browsing, there will be a behavior of adding a car, and eventually there will be an behavior of placing an order. That is to say, each user's behavior flow has certain characteristics.
  • Using the user's historical behavior sequence log as the copy feature of the copy information is beneficial to the prediction model to estimate the click probability of the copy information push. Finally, the highest click probability is selected. The copy information of the product is pushed.
  • the sample data may further include a weight; the weight represents a value associated with the copy feature of the input copy information, and is used to indicate the importance of the copy feature of the copy information in the predicted output value; the weight can Indicates how important each copy feature is in predicting the output value; i.e., a copy feature with a lower weight is less important in the prediction process than a copy feature with a higher weight.
  • the type of the estimation model is not limited, and the CTR estimation model may be used, and other types of estimation models may also be used.
  • Fig. 10 is a schematic structural diagram of screening the copywriting information of a commodity through a CTR estimation model in an embodiment of the application, as shown in Fig. 10, the input of the CTR estimation model is the historical behavior data of the first user and the copy feature database, wherein the copy feature library includes the copy feature of the copy information of each product in the candidate product set, and the copy feature of each copy information can be obtained according to the historical behavior data of the first user and weight; the output of the CTR prediction model is the click probability corresponding to the copy feature of each copy information; sort according to the click probability from large to small, and the copy information of the product with the click probability first is downloaded through the PUSH channel send.
  • the following describes the hot sale score, popularity score and purchase score in the sorting algorithm layer.
  • the embodiment of the present application proposes a method, device, electronic device, computer storage medium and computer program product for processing commodity text.
  • the method includes: acquiring text information of multiple commodities; determining the first user according to historical behavior data of the first user. Purchase preference information of the user; historical behavior data includes data related to commodities; according to the purchase preference information of the first user, a candidate commodity set is determined among multiple commodities; based on the copy information of the multiple commodities and the candidate commodity set, a candidate commodity is generated Copy information of each item in the collection; push copy information of at least one item in the candidate item set to the terminal of the first user; in this way, the copy information of the item does not need to be manually configured by the operator, but is generated according to the user's purchase preference information.
  • the copywriting information of the product can improve the generation efficiency of copywriting information; in addition, since there are certain differences between the purchase preference information of different users, some copywriting information will be generated according to the user's own purchase preference information and pushed, which can solve the problem of conflict. The problem that the copy information of the user is too single, and the accuracy of the message is improved.
  • the above method may further include: determining attribute information of each commodity in the candidate commodity set; the attribute information includes at least one of the following: a hot sale score, a popularity score and a purchase score; The sales volume of commodities related to the user's region; the popularity score represents the popularity of commodities related to the second user's region; the purchase score represents the purchase needs of commodities related to the second user; the second user is a different user from the first user;
  • the attribute information of each commodity in the candidate commodity set is processed to obtain the click score value of each commodity; the click score value represents the click probability of the copy information of each commodity; according to the click score value, the location of the second user is determined Sort a plurality of commodities in the middle to obtain a second sorting result; based on the second sorting result, select a target commodity set from the candidate commodity set, and push the copy information of the target commodity set to the terminal of the second user, where the target commodity set represents the candidate commodity set one or more of the items in .
  • the second user may represent a user who has no historical behavior data on the e-commerce platform, that is, belongs to a newly registered user; it may also represent a user who has historical behavior data, but cannot be determined because the historical behavior data is too small or the historical behavior data is not concentrated. Users who purchase preference information.
  • the second user jointly determines the click score value of each product according to the hot sale score, popularity score and purchase score of each product in the candidate set, and sends the second user to the second user according to the sorting result of the click score value.
  • the user's terminal pushes the copy information of the products in the front row; since the products in the candidate product set are determined according to the purchase preference information of each first user, and the copy information of each product in the candidate product set is determined, therefore , the copy information of the commodity pushed to the terminal of the second user is also determined.
  • the above method may further include: acquiring historical sales information of each commodity in the region where the second user is located; acquiring first indicator information of each commodity from the historical sales information, where the first indicator information represents a relationship with the commodity order Relevant information; process the first index information of each commodity to obtain the hot-selling score of each commodity.
  • the first indicator information may include three indicator values: the number of users placing orders, the number of orders, and the order amount, which are called first indicator values; here, the first indicator information of each commodity is processed,
  • the implementation manner of obtaining the hot-selling score of each commodity may be as follows: first, taking the logarithm to remove the dimension for each indicator value of the first indicator information, and performing an operation of adding 1 to each indicator value, that is, ln(the first indicator value +1); then, normalize ln (first index value + 1) and add 1 operation, that is (current value-minimum value)/(maximum value-minimum value+1), to obtain the first index information
  • the corresponding index values; the above operation of adding 1 is to avoid the occurrence of 0 values; finally, formula (6) is used to perform weighted summation of the index values corresponding to the first index information to calculate the hot-selling score of each product Score2:
  • Score2 Number of users placing orders*0.5+Number of orders*0.4+Order amount*0.1 (6)
  • the above method may further include: acquiring historical sales information of each commodity in the area where the second user is located; acquiring second indicator information of each commodity from the historical sales information, where the second indicator information represents a relationship with the commodity order and information related to the user's purchasing behavior; the second index information of each commodity is processed to obtain the popularity score of each commodity.
  • the second indicator information may include six indicator values: the number of users placing orders, the number of orders, the order amount, the number of users who add cars, the number of users who click on searches, and the number of users who enter the business details page, which are referred to as the second index value.
  • Index value here, the second index information of each commodity is processed to obtain the popularity score of each commodity.
  • the operation of adding 1 is performed on each index value, that is, ln (the second index value+1); then, ln (the second index value+1) is normalized and the operation of adding 1 is performed, that is, (current value-minimum value) /(Maximum value-minimum value+1), to obtain the index values corresponding to the second index information; the above operation of adding 1 is to avoid the occurrence of 0 values; finally, formula (7) is used to calculate the corresponding index values of the second index information. Weighted summation of item index values to calculate the popularity score Score3 of each product:
  • Score3 Number of users placing orders*0.4+Number of orders*0.25+Amount of order*0.05+Number of users who add cars*0.15+Number of users who click on search*0.1+Number of users who enter the business details page*0.05 (7)
  • the above method may further include: acquiring feature information of a plurality of first users; the feature information includes information of at least one dimension related to the users; determining each first user and the first user among the plurality of first users and The similarity value between the feature information of the two users; determine at least one first user whose similarity value is greater than the set threshold, assign the candidate product set of one of the at least one first user to the second user, and assign one of the user and the The similarity value of the second user is used as the purchase score of each commodity in the candidate commodity set of one of the users.
  • the feature information can be determined according to some basic information entered by the user when registering the e-commerce platform; for example, the user's age, user interest, user location, etc., can be specifically set according to the actual application scenario, the embodiment of the present application No restrictions apply.
  • the similarity value between the characteristic information of each first user and the second user is calculated by the similarity algorithm, and the judgment is made. Whether the similarity value of the two is greater than the set threshold, in the case of determining that the similarity value of the two is greater than the set threshold, assign the first user's candidate product set to the second user, and use the similarity value of the two as the candidate The purchase score for each item in the item collection.
  • the setting of the set threshold may be set according to the actual situation, for example, 0.8, 0.9, etc., which are not limited in this embodiment of the present application.
  • the attribute information includes: hot sales score, popularity score and purchase score; processing the attribute information of each commodity in the candidate commodity set to obtain the click score value of each commodity, which may include: The hot-selling score, popularity score and purchase score of each product in the product set are weighted and summed to obtain the click score value of each product.
  • the similarity between users with similar interests and preferences is calculated by analyzing the feature information of users when they register, and the commodity preferences of similar users are assigned to users without historical behavior data according to the similarity, and those users who have no historical behavior data are matched to the users. They may “like” products, and push the copy information of these products to users, which can enhance user perception and guide users to click on consumption.
  • the weighted sum of the hot sale score, hotness score and purchase score of each commodity can be calculated by using formula (8).
  • Score4 Hot sales score*0.3+Hotness score*0.2+Purchase score*0.5 (8)
  • the copywriting message of the product pushed to the user is comprehensively determined from three angles, which can effectively improve the accuracy of message delivery.
  • FIG. 11 is a schematic diagram of the composition and structure of the commodity copywriting processing device according to the embodiment of the application. As shown in FIG. 11 , the device includes: an acquisition module 200, a determination module 201, a generation module 202, and a push module 203, wherein:
  • the obtaining device 200 is configured to obtain copy information of a plurality of commodities
  • the determining module 201 is configured to determine the purchase preference information of the first user according to the historical behavior data of the first user; the historical behavior data includes data related to commodities;
  • the generating module 202 is configured to determine a candidate commodity set among a plurality of commodities according to the purchase preference information of the first user; based on the copywriting information of the multiple commodities and the candidate commodity set, generate the copywriting information of each commodity in the candidate commodity set;
  • the push module 203 is configured to push the copy information of at least one commodity in the candidate commodity set to the terminal of the first user.
  • the determining module 201 is further configured to:
  • the generating module 202 is further configured to generate, based on the copywriting information of the multiple products and the candidate product set, the copywriting information of each product in the candidate product set, including:
  • the initial copy information of each product in the candidate product set is generated
  • the purchase benefit information and the initial copy information are spliced together to generate copy information for each product in the candidate product set.
  • the purchase preference information includes at least one of the following: commodity category preference, commodity manufacturer identification preference, commodity repurchase information.
  • the product repurchase information includes the repurchased product and the product repurchase cycle
  • the push module 203 is further configured to:
  • the push time of the copy information of the repurchased commodity is determined; at least one commodity includes the repurchased commodity.
  • the push module 203 is configured to push the copy information of at least one commodity in the candidate commodity set to the terminal of the first user, including:
  • At least one commodity is selected from the candidate commodity set, and the copy information of the at least one commodity is pushed to the terminal of the first user.
  • the prediction model is trained by the following steps:
  • sample data includes: the copy feature of the copy information and the click probability label value
  • the prediction model is trained through the sample data, and the trained prediction model is obtained.
  • the push module 203 is further configured to:
  • the attribute information includes at least one of the following: a hot-selling score, a hot-selling score and a purchase score;
  • the hot-selling score represents the sales volume of the commodity related to the region of the second user;
  • the purchase score represents the purchase needs of commodities related to the second user;
  • the second user is a different user from the first user;
  • the attribute information of each commodity in the candidate commodity set is processed to obtain the click score value of each commodity;
  • the click score value represents the click probability of the copy information of each commodity;
  • the push module 203 is further configured to:
  • the first index information of each commodity is processed to obtain the hot-selling score of each commodity.
  • the push module 203 is further configured to:
  • the second index information of each commodity is processed to obtain the popularity score of each commodity.
  • the push module 203 is further configured to:
  • the characteristic information includes information of at least one dimension related to the users
  • the attribute information includes: hot sales score, popularity score and purchase score; correspondingly, the push module 203 is further configured to process the attribute information of each commodity in the candidate commodity set to obtain the click of each commodity Score values, including:
  • Weighted summation is performed on the hot-selling score, hotness score and purchase score of each commodity in the candidate commodity set, and the click score value of each commodity is obtained.
  • the above acquisition module 200, determination module 201, generation module 202 and push module 203 can all be implemented by a processor located in an electronic device, and the processor can be an ASIC, DSP, DSPD, PLD, FPGA, CPU, At least one of a controller, a microcontroller, and a microprocessor.
  • each functional module in this embodiment may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware, or can be implemented in the form of software function modules.
  • the integrated unit is implemented in the form of software function modules and is not sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of this embodiment is essentially or correct. Part of the contribution made by the related art or all or part of the technical solution can be embodied in the form of a software product, the computer software product is stored in a storage medium, and includes several instructions to make a computer device (which can be a personal computer). , server, or network device, etc.) or processor (processor) executes all or part of the steps of the method in this embodiment.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read only memory (Read Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program codes.
  • the computer program instructions corresponding to a commodity copy processing method in this embodiment may be stored on a storage medium such as an optical disc, a hard disk, a USB flash drive, etc.
  • a storage medium such as an optical disc, a hard disk, a USB flash drive, etc.
  • FIG. 12 shows the electronic device 300 provided by the present application, which may include: a memory 301 and a processor 302; wherein,
  • memory 301 configured to store computer programs and data
  • the processor 302 is configured to execute the computer program stored in the memory, so as to implement any one of the commodity copy processing methods in the foregoing embodiments.
  • the above-mentioned memory 301 can be a volatile memory (volatile memory), such as RAM; or a non-volatile memory (non-volatile memory), such as ROM, flash memory (flash memory), hard disk (Hard Disk) Drive, HDD) or solid-state drive (Solid-State Drive, SSD); or a combination of the above types of memory, and provide instructions and data to the processor 302.
  • volatile memory such as RAM
  • non-volatile memory such as ROM, flash memory (flash memory), hard disk (Hard Disk) Drive, HDD) or solid-state drive (Solid-State Drive, SSD); or a combination of the above types of memory, and provide instructions and data to the processor 302.
  • the above-mentioned processor 302 may be at least one of ASIC, DSP, DSPD, PLD, FPGA, CPU, controller, microcontroller, and microprocessor. It can be understood that, for different commodity document processing devices, the electronic device used to implement the above processor function may also be other, which is not specifically limited in the embodiment of the present application.
  • the functions or modules included in the apparatuses provided in the embodiments of the present application may be used to execute the methods described in the above method embodiments.
  • the functions or modules included in the apparatuses provided in the embodiments of the present application may be used to execute the methods described in the above method embodiments.
  • the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the application may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, optical storage, and the like.

Landscapes

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

Abstract

提出了一种商品文案处理方法、装置、电子设备、计算机存储介质和计算机程序产品，方法包括：获取多个商品的文案信息（100）；根据第一用户的历史行为数据，确定第一用户的购买偏好信息；历史行为数据包括与商品相关的数据（101）；根据第一用户的购买偏好信息，在多个商品中确定候选商品集合；基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的文案信息（102）；向第一用户的终端推送候选商品集合中至少一个商品的文案信息（103）。

Description

一种商品文案处理方法、装置、电子设备、介质和程序
相关申请的交叉引用
本申请基于申请号为202110057323.4、申请日为2021年01月15日的中国专利申请提出，申请人为：北京沃东天骏信息技术有限公司，北京京东世纪贸易有限公司，申请名称为“一种商品文案处理方法、装置、电子设备和存储介质”的技术方案，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本申请涉及互联网技术领域，涉及但不限于一种商品文案处理方法、装置、电子设备、计算机存储介质和计算机程序产品。
背景技术
目前，平台运营会通过消息推送(PUSH)触达用户，以推广商品促销活动及拉新促活等；然而，相关技术中，向用户推送的消息内容是运营人员通过内容管理系统(Content Management System，CMS)后台手动配置的，且触达的用户群体为全量用户；这样，不仅消息内容的生成效率低，且触达用户的消息内容过于单一，降低消息触达的精准性。
发明内容
本申请提供一种商品文案处理方法、装置、电子设备、计算机存储介质和计算机程序产品，可以解决触达用户的文案信息过于单一的问题，即，针对不同的用户触发不同的文案信息。
本申请的技术方案是这样实现的：
本申请实施例提供了一种商品文案处理方法，所述方法包括：
获取多个商品的文案信息；
根据第一用户的历史行为数据，确定第一用户的购买偏好信息；所述历史行为数据包括与商品相关的数据；
根据所述第一用户的购买偏好信息，在所述多个商品中确定候选商品集合；基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的文案信息；
向所述第一用户的终端推送候选商品集合中至少一个商品的文案信息。
在一些实施例中，所述方法还包括：
根据第一用户的历史行为数据，确定所述第一用户的购买利益信息；
相应地，所述基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的文案信息，包括：
基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的初始文案信息；
将所述购买利益信息和所述初始文案信息进行拼接，生成候选商品集合中每个商品的文案信息。
在一些实施例中，所述购买偏好信息包括以下至少之一：商品类目偏好、商品厂商标识偏好、商品复购信息。
在一些实施例中，所述商品复购信息包括复购商品和商品复购周期，所述方法还包括：
根据所述商品复购信息，确定所述复购商品的文案信息的推送时间；所述至少一个商品包括复购商品。
在一些实施例中，所述向所述第一用户的终端推送候选商品集合中至少一个商品的文案信息，包括：
提取候选商品集合中每个商品的文案信息的文案特征；
使用训练完成的预估模型对所述每个商品的文案信息的文案特征进行处理，得到所述每个商品的文案信息的点击概率；
按照所述点击概率对所述每个商品的文案信息进行排序，得到第一排序结果；
根据所述排序结果，在候选商品集合中选取出所述至少一个商品，向所述第一用户的终端推送所述至少一个商品的文案信息。
在一些实施例中，所述预估模型是通过以下步骤训练得到的：
获取样本数据；所述样本数据包括：文案信息的文案特征和点击概率标签值；
通过所述样本数据对所述预估模型进行训练，得到训练完成的预估模型。
在一些实施例中，所述方法还包括：
确定所述候选商品集合中每个商品的属性信息；所述属性信息包括以下至少之一：热销分数、热度分数和购买分数；所述热销分数表示与第二用户所在地域相关的商品的销量；所述热度分数表示与所述第二用户所在地域相关的商品的热度；所述购买分数表示与所述第二用户相关的商品的购买需要；所述第二用户为与所述第一用户不同的用户；
对所述候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值；所述点击得分值表征每个商品的文案信息的点击概率；
按照所述点击得分值对所述第二用户所在地域中多个商品进行排序，得到第二排序结果；基于所述第二排序结果，在所述候选商品集合中选取目标商品集合，向所述第二用户的终端推送目标商品集合的文案信息，所述目标商品集合表示所述候选商品集合中的一个或多个商品。
在一些实施例中，所述方法还包括：
获取第二用户所在地域中每个商品的历史销售信息；从所述历史销售信息获取每个商品的第一指标信息，所述第一指标信息表示与商品订单相关的信息；
对所述每个商品的第一指标信息进行处理，得到每个商品的热销分数。
在一些实施例中，所述方法还包括：
获取第二用户所在地域中每个商品的历史销售信息；从所述历史销售信息获取每个商品的第二指标信息，所述第二指标信息表示与商品订单以及用户购买行为相关的信息；
对所述每个商品的第二指标信息进行处理，得到每个商品的热度分数。
在一些实施例中，所述方法还包括：
获取多个第一用户的特征信息；所述特征信息包括与用户相关的至少一个维度的信息；
确定多个第一用户中每个第一用户和第二用户的特征信息之间的相似值；
确定相似值大于设定阈值的至少一个第一用户，将所述至少一个第一用户中的其中一个用户的候选商品集合赋予第二用户，将所述其中一个用户和第二用户的相似值作为所述其中一个用户的候选商品集合中每个商品的购买分数。
在一些实施例中，所述属性信息包括：热销分数、热度分数和购买分数；相应地，所述对所述候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值，包括：
对所述候选商品集合中每个商品的热销分数、热度分数和购买分数进行加权求和，得到所述每个商品的点击得分值。
本申请实施例还提出了一种商品文案处理装置，所述装置包括获取模块、确定模块、生成模块和推送模块，其中，
获取装置，配置为获取多个商品的文案信息；
确定模块，配置为根据第一用户的历史行为数据，确定第一用户的购买偏好信息； 所述历史行为数据包括与商品相关的数据；
生成模块，配置为根据所述第一用户的购买偏好信息，在所述多个商品中确定候选商品集合；基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的文案信息；
推送模块，配置为向所述第一用户的终端推送候选商品集合中至少一个商品的文案信息。
本申请实施例提供一种电子设备，所述设备包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现前述一个或多个技术方案提供的商品文案处理方法。
本申请实施例提供一种计算机存储介质，所述计算机存储介质存储有计算机程序；所述计算机程序被执行后能够实现前述一个或多个技术方案提供的商品文案处理方法。
本申请实施例还提供了一种计算机程序产品，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行用于实现前述一个或多个技术方案提供的商品文案处理方法。
本申请实施例提出了一种商品文案处理方法、装置、电子设备、计算机存储介质和计算机程序产品，该方法包括：获取多个商品的文案信息；根据第一用户的历史行为数据，确定第一用户的购买偏好信息；所述历史行为数据包括与商品相关的数据；根据所述第一用户的购买偏好信息，在所述多个商品中确定候选商品集合；基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的文案信息；向所述第一用户的终端推送候选商品集合中至少一个商品的文案信息；如此，无需通过运营人员手动配置商品的文案信息，而是根据用户的购买偏好信息对应生成商品的文案信息，可以提高文案信息的生成效率；另外，由于不同用户的购买偏好信息之间存在一定的差异性，因而，将根据用户自身的购买偏好信息生成部分文案信息进行推送，可以解决触达用户的文案信息过于单一的问题，提高消息触达的精准性。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本申请的实施例，并与说明书一起用于说明本申请的技术方案。
图1为相关技术中影响消息推送触达用户的结果示意图；
图2是本申请实施例中的一种商品文案推送框架的结构示意图；
图3a为本申请实施例中强触达场景的示意图一；
图3b为本申请实施例中强触达场景的示意图二；
图3c为本申请实施例中强触达场景的示意图三；
图4a为本申请实施例中两种以优惠券为例进行推送的消息内容的示意图；
图4b为本申请实施例中两种以促销活动为例进行推送的消息内容的示意图；
图5是本申请实施例中的一种商品文案处理方法的流程示意图；
图6为本申请实施例中得到四种不同类型文案信息的结果示意图；
图7为本申请实施例中对复购商品的复购周期进行说明的示意图；
图8为本申请实施例中生成商品的文案信息的结果示意图；
图9为本申请实施例中获取的用户历史行为序列日志的结果示意图；
图10为本申请实施例中通过点击通过率(Click-Through-Rat，CTR)预估模型对商品的文案信息进行筛选的结构示意图；
图11为本申请实施例的商品文案处理装置的组成结构示意图；
图12为本申请实施例的电子设备的结构示意图。
具体实施方式
以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所提供的实施例仅仅用以解释本申请，并不用于限定本申请。另外，以下所提供的实施例是用于实施本申请的部分实施例，而非提供实施本申请的全部实施例，在不冲突的情况下，本申请实施例记载的技术方案可以任意组合的方式实施。
需要说明的是，在本申请实施例中，术语“包括”、“包含”或者其任何其它变体意在涵盖非排他性的包含，从而使得包括一系列要素的方法或者装置不仅包括所明确记载的要素，而且还包括没有明确列出的其它要素，或者是还包括为实施方法或者装置所固有的要素。在没有更多限制的情况下，由语句“包括一个......”限定的要素，并不排除在包括该要素的方法或者装置中还存在另外的相关要素(例如方法中的步骤或者装置中的单元，例如的单元可以是部分电路、部分处理器、部分程序或软件等等)。
本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，I和/或J，可以表示：单独存在I，同时存在I和J，单独存在J这三种情况。另外，本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合，例如，包括I、J、R中的至少一种，可以表示包括从I、J和R构成的集合中选择的任意一 个或多个元素。
例如，本申请实施例提供的商品文案处理方法包含了一系列的步骤，但是本申请实施例提供的商品文案处理方法不限于所记载的步骤，同样地，本申请实施例提供的商品文案处理装置包括了一系列模块，但是本申请实施例提供的商品文案处理装置不限于包括所明确记载的模块，还可以包括为获取相关数据、或基于相关数据进行处理时所需要设置的模块。
本申请实施例可以应用于终端设备和服务器组成的计算机系统中，并可以与众多其它通用或专用计算系统环境或配置一起操作。这里，终端设备可以是瘦客户机、厚客户机、手持或膝上设备、基于微处理器的系统、机顶盒、可编程消费电子产品、网络个人电脑、小型计算机系统，等等，服务器可以是服务器计算机系统小型计算机系统﹑大型计算机系统和包括上述任何系统的分布式云计算技术环境，等等。
终端设备、服务器等电子设备可以在由计算机系统执行的计算机系统可执行指令(诸如程序模块)的一般语境下描述。通常，程序模块可以包括例程、程序、目标程序、组件、逻辑、数据结构等等，它们执行特定的任务或者实现特定的抽象数据类型。计算机系统/服务器可以在分布式云计算环境中实施，分布式云计算环境中，任务是由通过通信网络链接的远程处理设备执行的。在分布式云计算环境中，程序模块可以位于包括存储设备的本地或远程计算系统存储介质上。
推送(PUSH)指运营人员通过自己的产品或第三方工具对用户移动设备进行的主动消息推送。用户可以在移动设备锁定屏幕和通知栏看到PUSH消息通知，通知栏点击可唤起手机软件(Application，APP)并去往相应页面。
在进行消息推送时，千篇一律的消息内容很容易对用户形成打扰，导致APP被用户卸载；图1为相关技术中影响消息推送触达用户的结果示意图，如图1所示，影响消息推送触达用户的因素包括三个：送达率、展现率和点击率；其中，影响送达率的因素包括：通道质量、APP在线率、消息有效期和令牌(Token)有效性；这里，通道质量表示用于传输消息内容的通信通道的优劣程度；消息有效期表示消息内容在未被终端接收前的有效时间；Token有效期是用户向服务端进行请求认证时产生的，用于表明请求合法性的有效时间，它可以通过软件开发工具包(Software Development Kit，SDK)日志上报进行确定。影响展现率的因素为用户禁用情况；用户禁用情况表示用户未开启接收消息内容的PUSH通道。影响点击率的因素包括：标题、摘要、图片是否吸引用户以及是否命中用户兴趣。
可以看出，虽然推送消息的送达率在50.7％，但此部分数据仅是送达的用户，由于消息通道及APP卸载等原因，导致近一段时间未登录过APP的用户大概占三分之一，此部分用户已无法通过消息推送进行触达，也就是说仅有33.8％(2/3*50.7％)用户可接收到推送消息，而这批用户消息内容的展示率0.33％，点击率0.17％。上述点击率低的根本原因在于发送的消息内容对用户来说缺少吸引力，即缺少对用户利益点信息的露出，缺乏精细化能力。
这里，送达率可以通过基础服务搭建部分进行解决，展现率需要通过功能层面唤回部分解决，而点击率可通过个性化触达部分解决，即，向不同的用户展现不同的消息内容。图2是本申请实施例中的一种商品文案推送框架的结构示意图，如图2所示，基础服务搭建部分包括消息推送通道；消息推送(PUSH)通道是设备消息上行的通道，用户基于消息推送通道可以收到设备上报的消息。
这里，PUSH通道包括：Android和IOS；表1为相关技术中不同设备类型PUSH消息的送达率和打开率；可以看出，Android通道PUSH消息的送达率仅为25.15％，因而，需要对Android通道打通以提高PUSH消息的送达率。
功能层面唤回部分包括点击下发日志数据和下发服务两部分；其中，点击下发日志数据的作用是通过上报日志来识别用户意图进行数据挖掘后可实现下一步的个性化触达的精准性。从表1中可以看出，Android通道统计的打开率是0，经过数据分析可以确定是SDK版本太低所致，因而，可以通过升级PUSH SDK以解决上述问题。
Figure PCTCN2021139717-appb-000001
表1
下发服务是在各用户强触达场景下建立的PUSH通道关闭检测及引导打开的策略，用户打开PUSH通道是后续个性化触达的前置条件。图3a为本申请实施例中强触达场景的示意图一，图3b为本申请实施例中强触达场景的示意图二，图3c为本申请实施例中强触达场景的示意图三，如图3a-3c所示，强触达场景可以包括：直播预约、订单完成和秒杀活动开启。通过下发服务，可以在不同强触达场景完成时自动检测用户是否关闭了PUSH通道，若关闭则给与弹窗触达，以引导用户打开PUSH通道。
上述内容是对基础服务搭建部分和功能层面唤回部分进行的简单说明，其中，线框 部分对应的商品文案的个性化触达部分，这部分内容将在下面的实施例中进行具体说明。
对不同的用户特征精细化，以推送不同的消息，需要满足下面几个方面：精准人群投放、消息推送方式满足用户习惯、文案信息戳中用户痛点及利益点和触达品类符合用户需求；也就是说，消息内容的标题、摘要、图片要吸引人，且命中用户兴趣。考虑到流量及性能问题，目前PUSH通道暂未增加图片功能，本申请实施例中主要以摘要内容为例进行说明。
图4a为本申请实施例中两种以优惠券为例进行推送的消息内容的示意图，如图4a所示，以优惠券为例，本申请实施例中，可以将向用户终端推送的消息内容“你有一批优惠券到账，请查收！”转换成“赠送您130-5优惠券，快去购买乐事薯片黄瓜味40g”这种个性化触达的消息内容。同样地，图4b为本申请实施例中两种以促销活动为例进行推送的消息内容的示意图，如图4b所示，以促销活动为例，可以将向用户终端推送的消息内容“新权益到账，请注意查收！”转换成“恭喜你获得99免运费特权，你常购的乐事薯片原味40g也在特价，快去购买吧”这种个性化触达的消息内容。通过图4a和图4b，对两个消息内容的对比可知，个性化触达的消息内容包含用户的购买偏好以及具体的购买利益，可以提高用户对消息内容的点击率。
本申请实施例中，可以由系统推荐代替人工运营，在消息推送到达用户时，能够增强用户感知，漏出更多的利益点，引导用户点击消费。
针对上述应用场景以及技术问题，提出以下各实施例。
在本申请的一些实施例中，商品文案处理方法可以利用商品文案处理装置中的处理器实现，上述处理器可以为特定用途集成电路(Application Specific Integrated Circuit，ASIC)、数字信号处理器(Digital Signal Processor，DSP)、数字信号处理装置(Digital Signal Processing Device，DSPD)、可编程逻辑装置(Programmable Logic Device，PLD)、现场可编程逻辑门阵列(Field Programmable Gate Array，FPGA)、中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器中的至少一种。
图5是本申请实施例中的一种商品文案处理方法的流程示意图，如图5所示，该方法包括如下步骤：
步骤100：获取多个商品的文案信息。
这里，商品可以表示电商平台上的卖家通过互联网进行交易的任意类型的物品；例如，可以是服饰类物品、食品类物品等，还可以是虚拟物品等；本申请实施例对物品的类型不作限制。
在一种实施方式中，在获取多个商品的文案信息之前，可以获取电商平台上每个商品的原标题文案，对每个商品的原标题文案进行提取，得到关键信息，通过对关键信息的分析组合，可以得到每个商品的文案信息。
示例性地，为了提升商品文案信息的可读性，使商品更容易被搜索引擎检索到，商品原标题文案会罗列多个词覆盖大量的信息，然而，将纯文字内容的原标题文案通过PUSH通道推送给用户，使得用户很难在短时间内识别到关键内容；为了解决上述问题，可以对每个商品的原标题文案进行提取，并根据提取的关键信息，生成差异化的文案信息，使得文案信息能够抓住商品主体，增加用户的识别度。
这里，提取的关键信息可以根据实际应用场景进行确定，可以是商品自身的属性，例如，产品词、颜色、尺寸或重量等。
在一种实施方式中，根据关键信息，可以得到商品的不同类型的文案信息，文案信息的类型可以包括：短标题文案、增强文案、走心文案或亮点文案；其中，在不同的营销场景下，可选择不同类型的信息文案进行推送。
这里，可以通过图6对上述情况进行说明，图6为本申请实施例中得到四种不同类型文案信息的结果示意图，如图6所示，对于商品男士背包，可以根据该男士背包的原标题文案“SWISSGEAR时尚双肩包14.6英寸电脑包男商务背包苹果笔记本包休闲旅行包学生书包SA-9911黑色”可以得到四种不同类型的文案信息，分别为短标题文案“SWISSGEAR学生休闲时尚电脑包”、增强文案“怎样的箱包耐用又美丽？这些电脑包准没错！”、走心文案“男士休闲双肩包，让出行更轻松”以及亮点文案“防泼水面料，无惧雨水侵袭”。可以看出，短标题文案可以根据商品的厂家标识、产品词以及抽取的商品关键信息进行组合而成；增强文案偏广告宣传语，走心文案的描述偏主观，亮点文案的描述偏客观。
在一种实施方式中，可以对电商平台上每个商品的文案信息进行获取，而对于获取上述四种文案信息的方式，本申请实施例中不作限定，例如，可以通过相关的文案生成算法获取商品的文案信息，也可以通过其他方式进行获取。本申请实施例中以短标题文案为例对商品的文案信息进行说明。
步骤101：根据第一用户的历史行为数据，确定第一用户的购买偏好信息；历史行为数据包括与商品相关的数据。
本申请实施例中，第一用户表示在电商平台上有历史行为数据的任意一个用户，这里，历史行为数据可以包括历史搜索数据、历史浏览数据、历史加购物车数据以及历史 下单数据等。
在一种实施方式中，可以通过预设的计算规则对第一用户的历史行为数据进行数据挖掘，确定第一用户的购买偏好信息；其中，购买偏好信息可以包括以下至少之一：商品类目偏好、商品厂商标识偏好、商品复购信息。
示例性地，对于第一用户的历史行为数据，可以是第一用户在电商平台上近一年的历史行为数据，也可以是近六个月的历史行为数据；这里，对历史行为数据的时间段不作限制。
在一种实施方式中，对于确定第一用户的商品类目偏好的计算规则可以为：首先，根据第一用户的历史行为数据，获取第一用户在设定时间段内下单且订单状态为已完成的全部订单数据；根据该全部订单数据，确定以下四个用户维度数据，分别为：一级类目订单sku宽度(每日去重)、全部订单sku宽度、一级类目订单成交总额(Gross Merchandise Volume，GMV)，全部订单总GMV。然后，根据公式(1)和(2)分别计算第一用户在所有一级类目的Score_widths_rate和Score_amount_rate这两项数据指标得分：
Score_widths_rate＝一级类目订单sku宽度/全部订单sku宽度  (1)
Score_amount_rate＝一级类目累计GMV/全部订单总GMV     (2)
上述公式中，Score_widths_rate表示一级类目订单sku宽度占比，Score_amount_rate表示一级类目GMV占比。这里，一级类目sku宽度的数据逻辑如表2所示，包括两个字段，分别为用户ID和一级类目信息；其中，一级类目信息包括：一级类目名称、一级类目ID和一级类目订单sku宽度。
Figure PCTCN2021139717-appb-000002
表2
最后，根据上述两项数据指标得分，确定一级类目的输出标签，该过程包括以下步骤：首先，从公式(1)的计算结果中选取Score_widths_rate>＝0.1分的一级类目；然后， 根据公式(3)计算第一用户在该一级类目下对应的标签分值Score1：
Score1＝Score_widths_rate*0.7+Score_amount_rate*0.3    (3)
可以看出，根据上述计算规则，可以从第一用户近半年下单且订单状态为已完成的全部订单数据中确定所有一级类目以及对应的标签分值；这里，近半年表示订单数据的统计时长；进一步地，可以按照标签分值从大到小的顺序进行排序，得到排序结果，再将排序结果中排序在前的标签分值对应的一级类目作为第一用户的商品类目偏好。
示例性地，对于第一用户的商品类目偏好的计算可以设置更新周期，这样，每隔预设时间便对第一用户的商品类目偏好进行更新；由于用户在消费过程中，商品类目偏好可能会发生变化，而根据用户不断产生的订单数据重新确定对应的商品类目偏好，能够确保用户商品类目偏好的准确性。这里，更新周期可以为一个月、半年等，可以根据实际情况进行设置，本申请实施例对此不作限制。
在一种实施例中，假设根据第一用户的历史行为数据可以确定出以下四个类目，分别为：服装、配饰、美妆和零食，且它们对应的标签分值依次为：0.5、0.2、0.1、0.8；则可以将零食作为第一用户的商品类目偏好，也可以将零食和服装作为第一用户的商品类目偏好。
示例性地，上述第一类目也可以为二级类目或三级类目，具体可以根据实际应用场景进行对应设置。
在一种实施方式中，对于确定第一用户的商品厂商标识偏好的计算规则和上述确定商品类目偏好的计算规则相同，仅需将第一类目替换成厂商标识即可，此处不再累赘。
在一种实施方式中，商品复购信息包括：复购商品和商品复购周期；复购商品表示第一用户存在复购行为的商品；对于第一用户的商品复购周期的计算规则可以为：根据第一用户的历史行为数据，获取第一用户在设定时间段内下单且订单状态为已完成的全部订单数据，在根据该全部订单数据确定存在复购商品的情况下，通过公式(4)计算第一用户购买单位复购商品的复购周期：
Figure PCTCN2021139717-appb-000003
其中，t为大于零的整数，P t表示第一用户第t次购买单位复购商品的复购周期；T t+1表示第一用户第t+1次购买该复购商品的时间；T t表示第一用户第t次购买该复购商品的时间；N t表示第一用户第t次购买的复购商品数量；V t表示第一用户第t次购买的复 购商品单位(例如，多少件、多少瓶或多少毫升等，无＝1)。这里，购买复购商品的时间以天为单位，如果第一用户在同一天购买多单，则对复购商品数量进行累加。
对于不同的复购商品，第一用户的复购周期可能不同；即，第一用户在设定时间段内对一些复购商品可能存在不止一次的复购行为；图7为本申请实施例中对复购商品的复购周期进行说明的示意图，如图7所示，T5表示当前时间，T1、T2、T3和T4分别表示第一次、第二次、第三次和第四次购买复购商品的时间，即，第一用户的历史下单数据；结合上述公式(4)分别计算在T1～T2、T2～T3、T3～T4和T4～T5这些时间间隔内第一用户购买单位复购商品的复购周期P t；对这些复购周期按照从小到大的顺序进行排序，并取中位数作为第一用户购买单位复购商品的复购周期。
在一种实施方式中，假设根据公式(4)确定第一用户购买单位复购商品A的复购周期分别为：10天、15天和18天，则将15天作为第一用户购买单位复购商品A的复购周期。
示例性地，随着第一用户使用电商平台时间的不断增加，会不断产生对某些复购商品的复购行为；为了确保复购周期的准确性，可以设置更新周期每隔预设时间按照公式(4)对第一用户购买单位复购商品的复购周期进行更新。这里，更新周期可以根据复购商品类型进行确定，例如，对于冰箱或电脑这类复购商品，更新周期可以为一年或者两年等；对于零食这类复购商品，更新周期可以为半个月或者一个月等。
在一种实施方式中，上述方法还可以包括：根据第一用户的历史行为数据，确定第一用户的购买利益信息。
这里，购买利益信息可以根据不同的营销场景进行确定，例如，在促销场景下，购买利益信息可以为优惠券等；在获取新用户的场景下，购买利益信息可以运费券等；下面以优惠券为例对购买利益信息进行说明。
在一种实施方式中，优惠券包括优惠券门槛值和让利值；其中，优惠券门槛是依据第一用户购买商品的客单价进行定义的，这是因为若优惠券门槛值定义过高，则第一用户的日常消费达不到门槛，进而，很难通过该优惠券唤回用户进行消费；若优惠券门槛值定义太低，则不利于电商平台的利益最大化。由于不同用户购买商品的客单价不同，因而，不同用户对应的优惠券门槛值不相同。
在一种实施方式中，对于第一用户的购买利益信息的计算规则可以为：首先，根据第一用户的历史行为数据，获取第一用户设定时间段内每笔订单数据的成交金额(订单状态已支付，货到付款已完成)并取中位数；然后，使用公式(5)计算第一用户对应 的优惠券门槛值：
优惠券门槛值＝成交金额中位数*1.1            (5)
这里，在优惠券门槛值为小数的情况下，去掉小数位后取整，非整数进位取整。
在得到第一用户对应的优惠券门槛值后，确定让利值；示例性地，通过分析第一用户的历史行为数据以及采销毛利数据，得出的减免让利值分别为优惠券门槛值低于300元的让利值均为5元，优惠券门槛值处于300～400元之间的让利值为10元，优惠券门槛值处于500～600元之间的让利值为20元，优惠券门槛值处于700～900之间让利值为25元，优惠券门槛值高于1000元的让利值为30元。
在一种实施方式中，假设成交金额的中位数是319元，则优惠券门槛值定义为319*1.1＝350.9，则取350为优惠券门槛值，根据上述优惠券门槛值与让利值的对应关系，将优惠券设定为优惠券门槛值-让利，即350-10的优惠券。
在一种实施方式中，假设成交金额的中位数是319元，则优惠券门槛值定义为112*1.1＝123.2，则取130为优惠券门槛值，即，该优惠券设定为130-5的优惠券。
在一种实施方式中，优惠券包括三种类型：平台券，可购买电商平台上的全部商品；品牌券，可购买指定品牌商品；品类券，可购买指定品类商品。其中，平台券成本由电商平台承担，品牌券和品类券成本则由品牌商承担。
示例性地，在对第一用户进行优惠券下发的时候可以优先下发品牌券和品类券，故进行消息触达时会结合第一用户的商品类目偏好、商品厂商标识偏好，即优先确定第一用户的购买偏好信息且平台利益最大化情况下下发优惠券。若第一用户的商品类目和商品厂商标识偏好无集中趋势，则下发平台券。
步骤102：根据第一用户的购买偏好信息，在多个商品中确定候选商品集合；基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的文案信息。
在一种实施方式中，由于购买偏好信息可以包括以下至少之一：商品类目偏好、商品厂商标识偏好和商品复购信息；因而，可以将第一用户的商品类目偏好涉及到的所有商品放入候选商品集合中，也可以将第一用户的商品厂商标识偏好涉及到的所有商品放入候选商品集合中，还可以将第一用户的复购商品放入候选商品集合中。
这里，在将第一用户的复购商品放入候选商品集合之前，需要执行以下判断操作：需要根据上次购买该复购商品的数量与复购周期的乘积，判断下次购买该复购商品距离本次购买的时间间隔是否小于设定值，在时间间隔小于设定值的情况下，将复购商品放入候选商品集合中；在时间间隔大于或等于设定值的情况下，不再将复购商品放入候选 商品集合中。
这里，设定值可以根据实际情况进行设置，本申请实施例不作限制；例如，可以为一天、三天或五天等。
在一种实施方式中，候选商品集合中包括电商平台上的部分商品；由于在确定候选商品集合之前，已经生成了电商平台上所有商品对应的文案信息；因而，在确定候选商品集合后，可以根据商品与文案信息之间的对应关系，查找到候选商品集合中每个商品对应的文案信息。
在一种实施方式中，基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的文案信息，可以包括：基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的初始文案信息；将购买利益信息和初始文案信息进行拼接，生成候选商品集合中每个商品的文案信息。
示例性地，可以将根据上述对应关系查找到的候选商品集合中每个商品对应的文案信息作为初始文案信息，例如，将每个商品的短标题文案作为初始文案信息；再将第一用户的购买利益信息与每个商品的初始文案信息进行拼接，例如，将第一用户对应的优惠券与每个商品的短标题文案进行拼接，生成每个商品的文案信息。
在一种实施方式中，假设通过步骤101确定出第一用户的购买利益信息为130-5的优惠券，第一用户的商品类目偏好为：薯片，商品厂商标识偏好为：XX，复购商品为：“XX薯片40g办公室休闲膨化零食品小吃黄瓜味生产日期8月”；通过步骤100对商品的原标题“XX薯片40g办公室休闲膨化零食品小吃黄瓜味生产日期8月”进行处理，得到该商品的短标题文案“XX薯片黄瓜味40g”；将第一用户的购买利益信息与该商品的初始文案信息以及文案模板进行拼接，可以得到该商品的文案信息“赠送你130-5优惠券，快去购买XX薯片黄瓜味40g”。即，最终生成的商品的文案信息如图8所示。
这里，文案模板可以在电商平台中进行预先配置，对于文案模板的类型可以根据实际场景自行定义，本申请实施例不作限制。例如，优惠券文案模板为：赠送你XXX-X优惠券，快去购买XXX短标题文案。
在一种实施方式中，对于无历史订单行为的用户，会触发无门槛5元优惠券，对应的文案模板为：赠送你5元无门槛优惠券，快去购买XXX短标题文案。
可见，对于相同的初始文案信息，使用不同的文案模板，可对应生成不同的文案消息，这些文案消息可以应用于不同的营销场景，以提高文案消息的适用性。
步骤103：向第一用户的终端推送候选商品集合中至少一个商品的文案信息。
在一种实施方式中，通过上述步骤可以对候选商品集合中的每个商品确定出对应的文案信息；然而，在向第一用户的终端推送这些文案消息时，若将这些商品的文案信息全部进行推送，会给第一用户带来打扰的同时，很可能导致第一用户关闭消息推送通道，这样，对于后续生成的商品的文案信息，即使推送到第一用户的终端上，也将无法被用户点击。
可见，在向第一用户的终端推送文案消息时，需要对候选商品集合中每个商品的文案信息进行筛选，将最有可能被用户点击商品的文案消息进行推送，将不太可能被用户点击商品的文案消息进行忽略。
在一种实施方式中，向第一用户的终端推送候选商品集合中至少一个商品的文案信息，可以包括：提取候选商品集合中每个商品的文案信息的文案特征；使用训练完成的预估模型对每个商品的文案信息的文案特征进行处理，得到每个商品的文案信息的点击概率；按照点击概率对每个商品的文案信息进行排序，得到第一排序结果；根据排序结果，在候选商品集合中选取出至少一个商品，向第一用户的终端推送至少一个商品的文案信息。
在一种实施方式中，对于候选商品集合中每个商品的文案信息对应的点击概率，可按照点击概率从大到小的顺序进行排序，得到排序结果；并将排序结果中点击概率排列在前的至少一个商品的文案信息向第一用户的终端进行推送。
在一种实施方式中，假设第一用户的候选商品的文案信息分别为A1、A2和A3，且这三个文案信息对应的点击概率依次为0.8、0.2、0.5，经过排序后的排序结果为A1、A3和A2，则可以将文案信息A1向第一用户的终端进行推送，也可以将文案信息A1和A3向第一用户的终端进行推送。
在一种实施方式中，上述方法还包括：在至少一个商品包括复购商品的情况下，可以根据商品复购信息，确定复购商品的文案信息的推送时间。
示例性地，在根据步骤101得到单位复购商品的复购周期后，可以根据第一用户距离当前时刻的上次购买复购商品的数量，确定出下次购买该复购商品的时间；进而，可以将下次购买该复购商品的时间作为该复购商品的文案信息的推送时间，这样，可以进一步提高用户的点击率。
在一种实施方式中，在第一用户的复购商品为牙刷的情况下，假设根据步骤101得到单个牙刷的复购周期为30天，且确定第一用户上次购买牙刷的数量为2，则可以确定出第一用户下次购买牙刷的时间为距离上次购买后的60天，进一步地，可以在距离上 次购买后的60天，向第一用户推送该牙刷的文案信息。
在一种实施方式中，上述预估模型是通过以下步骤训练得到的：获取样本数据；样本数据包括：文案信息的文案特征和点击概率标签值；通过样本数据对预估模型进行训练，得到训练完成的预估模型。
这里，对预估模型的训练是有监督的学习，即，对于输入X有着与之对应的实际值Y；这里，输入X表示每个文案信息的文案特征，实际值Y表示每个文案信息的文案特征对应的点击概率标签值。而预估模型的输入X与实际值Y之间的损失函数就是网络反向传播，整个神经网络的训练过程就是不断缩小损失函数的值的过程。
这里，文案信息的文案特征可以包括：用户的商品类目偏好、用户的商品厂商标识偏好、客单价、历史行为序列日志、商品产品词、商品类目、商品厂商标识和商品价格等。
图9为本申请实施例中获取的用户历史行为序列日志的结果示意图，如图9所示，该历史行为序列日志从上到下是按时间顺序进行排序，从用户历史行为序列日志可见，用户对每个商品都存在重复浏览行为，多次浏览后会有加车行为，最终会有下单行为。也就是说，每个用户的行为流都存在某种特征，将用户历史行为序列日志作为文案信息的文案特征，有利于预估模型预估文案信息推送的点击概率，最终，选出点击概率最高的商品的文案信息进行推送。
在一种实施方式中，样本数据还可以包括权重；权重表示与输入的文案信息的文案特征相关联的值，用于表明该文案信息的文案特征在预测输出值中的重要程度；该权重能够表明每个文案特征在预测输出值中的重要程度；即，与权重较大的文案特征相比，权重较小的文案特征在预测过程中的重要性较低。
本申请实施例中，对预估模型的类型不作限制，可以使用CTR预估模型，也可以为其他类型的预估模型等。
结合图2，这里对个性化触达部分进行说明；图10为本申请实施例中通过CTR预估模型对商品的文案信息进行筛选的结构示意图，如图10所示，CTR预估模型的输入为第一用户的历史行为数据以及文案特征库，其中，文案特征库包括候选商品集合中每个商品的文案信息的文案特征，根据第一用户的历史行为数据可以得到每个文案信息的文案特征以及权重；CTR预估模型的输出为每个文案信息的文案特征对应的点击概率；按照点击概率从大到小的顺序进行排序，将点击概率排序在前的商品的文案信息通过PUSH通道进行下发。下面对排序算法层中的热销分数、热度分数和购买分数进行说明。
本申请实施例提出了一种商品文案处理方法、装置、电子设备、计算机存储介质和计算机程序产品，该方法包括：获取多个商品的文案信息；根据第一用户的历史行为数据，确定第一用户的购买偏好信息；历史行为数据包括与商品相关的数据；根据第一用户的购买偏好信息，在多个商品中确定候选商品集合；基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的文案信息；向第一用户的终端推送候选商品集合中至少一个商品的文案信息；如此，无需通过运营人员手动配置商品的文案信息，而是根据用户的购买偏好信息对应生成商品的文案信息，可以提高文案信息的生成效率；另外，由于不同用户的购买偏好信息之间存在一定的差异性，因而，将根据用户自身的购买偏好信息生成部分文案信息进行推送，可以解决触达用户的文案信息过于单一的问题，提高消息触达的精准性。
在一种实施方式中，上述方法还可以包括：确定候选商品集合中每个商品的属性信息；属性信息包括以下至少之一：热销分数、热度分数和购买分数；热销分数表示与第二用户所在地域相关的商品的销量；热度分数表示与第二用户所在地域相关的商品的热度；购买分数表示与第二用户相关的商品的购买需要；第二用户为与第一用户不同的用户；对候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值；点击得分值表征每个商品的文案信息的点击概率；按照点击得分值对第二用户所在地域中多个商品进行排序，得到第二排序结果；基于第二排序结果，在候选商品集合中选取目标商品集合，向第二用户的终端推送目标商品集合的文案信息，目标商品集合表示候选商品集合中的一个或多个商品。
这里，第二用户可以表示在电商平台上无历史行为数据的用户，即，属于新注册用户；也可以表示有历史行为数据，但因历史行为数据过少或历史行为数据不集中而无法确定购买偏好信息的用户。
在一种实施方式中，第二用户根据候选集合中每个商品的热销分数、热度分数和购买分数共同确定每个商品的点击得分值，并根据点击得分值的排序结果向第二用户的终端推送排列在前的商品的文案信息；由于候选商品集合中的商品是根据每个第一用户的购买偏好信息确定的，且候选商品集合中每个商品的文案信息是确定的，因而，向第二用户的终端推送的商品的文案信息也是确定的。
在一种实施方式中，上述方法还可以包括：获取第二用户所在地域中每个商品的历史销售信息；从历史销售信息获取每个商品的第一指标信息，第一指标信息表示与商品订单相关的信息；对每个商品的第一指标信息进行处理，得到每个商品的热销分数。
在一些实施例中，第一指标信息可以包括：下单用户数、订单数和订单金额这三个指标值，称为第一指标值；这里，对每个商品的第一指标信息进行处理，得到每个商品的热销分数的实现方式可以为：首先，对第一指标信息的各个指标值取对数去除量纲，并对各个指标值执行加1操作，即，ln(第一指标值+1)；然后，对ln(第一指标值+1)进行归一化并执行加1操作，即(当前值-最小值)/(最大值-最小值+1)，得到第一指标信息对应的各项指标值；上述加1操作是为了避免出现0值情况；最后，使用公式(6)对第一指标信息对应的各项指标值进行加权求和，计算每个商品的热销分数Score2：
Score2＝下单用户数*0.5+订单数*0.4+订单金额*0.1         (6)
在一种实施方式中，上述方法还可以包括：获取第二用户所在地域中每个商品的历史销售信息；从历史销售信息获取每个商品的第二指标信息，第二指标信息表示与商品订单以及用户购买行为相关的信息；对每个商品的第二指标信息进行处理，得到每个商品的热度分数。
在一些实施例中，第二指标信息可以包括：下单用户数、订单数、订单金额、加车用户数、搜索点击用户数、进入商详页用户数这六个指标值，称为第二指标值；这里，对每个商品的第二指标信息进行处理，得到每个商品的热度分数的实现方式可以为：首先，对第二指标信息的各个指标值取对数去除量纲，并对各个指标值执行加1操作，即，ln(第二指标值+1)；然后，对ln(第二指标值+1)进行归一化并执行加1操作，即(当前值-最小值)/(最大值-最小值+1)，得到第二指标信息对应的各项指标值；上述加1操作是为了避免出现0值情况；最后，使用公式(7)对第二指标信息对应的各项指标值进行加权求和，计算每个商品的热度分数Score3：
Score3＝下单用户数*0.4+订单数*0.25+订单金额*0.05+加车用户数*0.15+搜索点击用户数*0.1+进入商详页用户数*0.05         (7)
在一种实施方式中，上述方法还可以包括：获取多个第一用户的特征信息；特征信息包括与用户相关的至少一个维度的信息；确定多个第一用户中每个第一用户和第二用户的特征信息之间的相似值；确定相似值大于设定阈值的至少一个第一用户，将至少一个第一用户中的其中一个用户的候选商品集合赋予第二用户，将其中一个用户和第二用户的相似值作为其中一个用户的候选商品集合中每个商品的购买分数。
在一种实施方式中，特征信息可以根据用户注册电商平台时录入的一些基本信息进行确定；例如，用户年龄，用户兴趣、用户所在地等，具体可以根据实际应用场景进行设置，本申请实施例不作限制。
本申请实施例中，分别获取电商平台上每个第一用户与第二用户的特征信息后，通过相似度算法计算每个第一用户与第二用户的特征信息之间的相似值，判断两者的相似值是否大于设定阈值，在确定两者的相似值大于设定阈值的情况下，将该第一用户的候选商品集合赋予第二用户，并将两者的相似值作为该候选商品集合中每个商品的购买分数。
这里，设定阈值的设置可以根据实际情况进行设置，例如，0.8、0.9等，本申请实施例中不作限制。
在一种实施方式中，属性信息包括：热销分数、热度分数和购买分数；对候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值，可以包括：对候选商品集合中每个商品的热销分数、热度分数和购买分数进行加权求和，得到每个商品的点击得分值。
本发明实施中，通过分析用户注册时的特征信息，计算具有相似兴趣偏好的用户之间的相似度，根据相似度的高低将相似用户的商品偏好赋予无历史行为数据的用户，给用户匹配那些他们可能“喜欢”的商品，将这些商品的文案信息推送给用户，可增强用户感知，引导用户点击消费。
在一种实施方式中，在根据上述步骤得到候选集合中每个商品的属性信息后，可以通过使用公式(8)对每个商品的热销分数、热度分数和购买分数进行加权求和，计算每个商品的点击得分值Score4：
Score4＝热销分数*0.3+热度分数*0.2+购买分数*0.5        (8)
在一种实施方式中，对于无历史行为的用户，通过三个角度综合确定向用户推送的商品的文案消息，可以有效提高消息触达的精准性。
图11为本申请实施例的商品文案处理装置的组成结构示意图，如图11所示，装置包括：获取模块200、确定模块201、生成模块202和推送模块203，其中：
获取装置200，配置为获取多个商品的文案信息；
确定模块201，配置为根据第一用户的历史行为数据，确定第一用户的购买偏好信息；历史行为数据包括与商品相关的数据；
生成模块202，配置为根据第一用户的购买偏好信息，在多个商品中确定候选商品集合；基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的文案信息；
推送模块203，配置为向第一用户的终端推送候选商品集合中至少一个商品的文案 信息。
在一些实施例中，确定模块201，还配置为：
根据第一用户的历史行为数据，确定第一用户的购买利益信息；
相应地，生成模块202，还配置为基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的文案信息，包括：
基于多个商品的文案信息和候选商品集合，生成候选商品集合中每个商品的初始文案信息；
将购买利益信息和初始文案信息进行拼接，生成候选商品集合中每个商品的文案信息。
在一些实施例中，购买偏好信息包括以下至少之一：商品类目偏好、商品厂商标识偏好、商品复购信息。
在一些实施例中，商品复购信息包括复购商品和商品复购周期，推送模块203，还配置为：
根据商品复购信息，确定复购商品的文案信息的推送时间；至少一个商品包括复购商品。
在一些实施例中，推送模块203，配置为向第一用户的终端推送候选商品集合中至少一个商品的文案信息，包括：
提取候选商品集合中每个商品的文案信息的文案特征；
使用训练完成的预估模型对每个商品的文案信息的文案特征进行处理，得到每个商品的文案信息的点击概率；
按照点击概率对每个商品的文案信息进行排序，得到第一排序结果；
根据排序结果，在候选商品集合中选取出至少一个商品，向第一用户的终端推送至少一个商品的文案信息。
在一些实施例中，预估模型是通过以下步骤训练得到的：
获取样本数据；样本数据包括：文案信息的文案特征和点击概率标签值；
通过样本数据对预估模型进行训练，得到训练完成的预估模型。
在一些实施例中，推送模块203，还配置为：
确定候选商品集合中每个商品的属性信息；属性信息包括以下至少之一：热销分数、热度分数和购买分数；热销分数表示与第二用户所在地域相关的商品的销量；热度分数表示与第二用户所在地域相关的商品的热度；购买分数表示与第二用户相关的商品 的购买需要；第二用户为与第一用户不同的用户；
对候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值；点击得分值表征每个商品的文案信息的点击概率；
按照点击得分值对第二用户所在地域中多个商品进行排序，得到第二排序结果；基于第二排序结果，在所述候选商品集合中选取目标商品集合，向第二用户的终端推送目标商品集合的文案信息，所述目标商品集合表示所述候选商品集合中的一个或多个商品。
在一些实施例中，推送模块203，还配置为：
获取第二用户所在地域中每个商品的历史销售信息；从历史销售信息获取每个商品的第一指标信息，第一指标信息表示与商品订单相关的信息；
对每个商品的第一指标信息进行处理，得到每个商品的热销分数。
在一些实施例中，推送模块203，还配置为：
获取第二用户所在地域中每个商品的历史销售信息；从历史销售信息获取每个商品的第二指标信息，第二指标信息表示与商品订单以及用户购买行为相关的信息；
对每个商品的第二指标信息进行处理，得到每个商品的热度分数。
在一些实施例中，推送模块203，还配置为：
获取多个第一用户的特征信息；特征信息包括与用户相关的至少一个维度的信息；
确定多个第一用户中每个第一用户和第二用户的特征信息之间的相似值；
确定相似值大于设定阈值的至少一个第一用户，将至少一个第一用户中的其中一个用户的候选商品集合赋予第二用户，将其中一个用户和第二用户的相似值作为其中一个用户的候选商品集合中每个商品的购买分数。
在一些实施例中，属性信息包括：热销分数、热度分数和购买分数；相应地，推送模块203，还配置为对候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值，包括：
对候选商品集合中每个商品的热销分数、热度分数和购买分数进行加权求和，得到每个商品的点击得分值。
在实际应用中，上述获取模块200、确定模块201、生成模块202和推送模块203均可以由位于电子设备中的处理器实现，该处理器可以为ASIC、DSP、DSPD、PLD、FPGA、CPU、控制器、微控制器、微处理器中的至少一种。
另外，在本实施例中的各功能模块可以集成在一个处理单元中，也可以是各个单 元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能模块的形式实现。
集成的单元如果以软件功能模块的形式实现并非作为独立的产品进行销售或使用时，可以存储在一个计算机可读取存储介质中，基于这样的理解，本实施例的技术方案本质上或者说对相关技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机、服务器、或者网络设备等)或processor(处理器)执行本实施例方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
具体来讲，本实施例中的一种商品文案处理方法对应的计算机程序指令可以被存储在光盘、硬盘、U盘等存储介质上，当存储介质中的与一种商品文案处理方法对应的计算机程序指令被一电子设备读取或被执行时，实现前述实施例的任意一种商品文案处理方法。
基于前述实施例相同的技术构思，参见图12，其示出了本申请提供的电子设备300，可以包括：存储器301和处理器302；其中，
存储器301，配置为存储计算机程序和数据；
处理器302，配置为执行存储器中存储的计算机程序，以实现前述实施例的任意一种商品文案处理方法。
在实际应用中，上述存储器301可以是易失性存储器(volatile memory)，例如RAM；或者非易失性存储器(non-volatile memory)，例如ROM、快闪存储器(flash memory)、硬盘(Hard Disk Drive，HDD)或固态硬盘(Solid-State Drive，SSD)；或者上述种类的存储器的组合，并向处理器302提供指令和数据。
上述处理器302可以为ASIC、DSP、DSPD、PLD、FPGA、CPU、控制器、微控制器、微处理器中的至少一种。可以理解地，对于不同的商品文案处理设备，用于实现上述处理器功能的电子器件还可以为其它，本申请实施例不作具体限定。
在一些实施例中，本申请实施例提供的装置具有的功能或包含的模块可以用于执行上文方法实施例描述的方法，其具体实现可以参照上文方法实施例的描述，为了简洁，这里不再赘述。
上文对各个实施例的描述倾向于强调各个实施例之间的不同之处，其相同或相似之 处可以互相参考，为了简洁，本文不再赘述。
本申请所提供的各方法实施例中所揭露的方法，在不冲突的情况下可以任意组合，得到新的方法实施例。
本申请所提供的各产品实施例中所揭露的特征，在不冲突的情况下可以任意组合，得到新的产品实施例。
本申请所提供的各方法或设备实施例中所揭露的特征，在不冲突的情况下可以任意组合，得到新的方法实施例或设备实施例。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用硬件实施例、软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器和光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
以上，仅为本申请的较佳实施例而已，并非用于限定本申请的保护范围。

Claims (15)

  1. 一种商品文案处理方法，所述方法包括：
    获取多个商品的文案信息；
    根据第一用户的历史行为数据，确定第一用户的购买偏好信息；所述历史行为数据包括与商品相关的数据；
    根据所述第一用户的购买偏好信息，在所述多个商品中确定候选商品集合；基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的文案信息；
    向所述第一用户的终端推送候选商品集合中至少一个商品的文案信息。
  2. 根据权利要求1所述的方法，其中，所述方法还包括：
    根据第一用户的历史行为数据，确定所述第一用户的购买利益信息；
    相应地，所述基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的文案信息，包括：
    基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的初始文案信息；
    将所述购买利益信息和所述初始文案信息进行拼接，生成候选商品集合中每个商品的文案信息。
  3. 根据权利要求1所述的方法，其中，所述购买偏好信息包括以下至少之一：商品类目偏好、商品厂商标识偏好、商品复购信息。
  4. 根据权利要求3所述的方法，其中，所述商品复购信息包括复购商品和商品复购周期，所述方法还包括：
    根据所述商品复购信息，确定所述复购商品的文案信息的推送时间；所述至少一个商品包括复购商品。
  5. 根据权利要求1所述的方法，其中，所述向所述第一用户的终端推送候选商品集合中至少一个商品的文案信息，包括：
    提取候选商品集合中每个商品的文案信息的文案特征；
    使用训练完成的预估模型对所述每个商品的文案信息的文案特征进行处理，得到所述每个商品的文案信息的点击概率；
    按照所述点击概率对所述每个商品的文案信息进行排序，得到第一排序结果；
    根据所述排序结果，在候选商品集合中选取出所述至少一个商品，向所述第一用户 的终端推送所述至少一个商品的文案信息。
  6. 根据权利要求5所述的方法，其中，所述预估模型是通过以下步骤训练得到的：
    获取样本数据；所述样本数据包括：文案信息的文案特征和点击概率标签值；
    通过所述样本数据对所述预估模型进行训练，得到训练完成的预估模型。
  7. 根据权利要求1所述的方法，其中，所述方法还包括：
    确定所述候选商品集合中每个商品的属性信息；所述属性信息包括以下至少之一：热销分数、热度分数和购买分数；所述热销分数表示与第二用户所在地域相关的商品的销量；所述热度分数表示与所述第二用户所在地域相关的商品的热度；所述购买分数表示与所述第二用户相关的商品的购买需要；所述第二用户为与所述第一用户不同的用户；
    对所述候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值；所述点击得分值表征每个商品的文案信息的点击概率；
    按照所述点击得分值对所述第二用户所在地域中多个商品进行排序，得到第二排序结果；基于所述第二排序结果，在所述候选商品集合中选取目标商品集合，向所述第二用户的终端推送目标商品集合的文案信息，所述目标商品集合表示所述候选商品集合中的一个或多个商品。
  8. 根据权利要求7所述的方法，其中，所述方法还包括：
    获取第二用户所在地域中每个商品的历史销售信息；从所述历史销售信息获取每个商品的第一指标信息，所述第一指标信息表示与商品订单相关的信息；
    对所述每个商品的第一指标信息进行处理，得到每个商品的热销分数。
  9. 根据权利要求7所述的方法，其中，所述方法还包括：
    获取第二用户所在地域中每个商品的历史销售信息；从所述历史销售信息获取每个商品的第二指标信息，所述第二指标信息表示与商品订单以及用户购买行为相关的信息；
    对所述每个商品的第二指标信息进行处理，得到每个商品的热度分数。
  10. 根据权利要求7所述的方法，其中，所述方法还包括：
    获取多个第一用户的特征信息；所述特征信息包括与用户相关的至少一个维度的信息；
    确定多个第一用户中每个第一用户和第二用户的特征信息之间的相似值；
    确定相似值大于设定阈值的至少一个第一用户，将所述至少一个第一用户中的其中一个用户的候选商品集合赋予第二用户，将所述其中一个用户和第二用户的相似值作为所述其中一个用户的候选商品集合中每个商品的购买分数。
  11. 根据权利要求7所述的方法，其中，所述属性信息包括：热销分数、热度分数和购买分数；
    相应地，所述对所述候选商品集合中每个商品的属性信息进行处理，得到每个商品的点击得分值，包括：
    对所述候选商品集合中每个商品的热销分数、热度分数和购买分数进行加权求和，得到所述每个商品的点击得分值。
  12. 一种商品文案处理装置，所述装置包括：
    获取装置，配置为获取多个商品的文案信息；
    确定模块，配置为根据第一用户的历史行为数据，确定第一用户的购买偏好信息；所述历史行为数据包括与商品相关的数据；
    生成模块，配置为根据所述第一用户的购买偏好信息，在所述多个商品中确定候选商品集合；基于所述多个商品的文案信息和所述候选商品集合，生成候选商品集合中每个商品的文案信息；
    推送模块，配置为向所述第一用户的终端推送候选商品集合中至少一个商品的文案信息。
  13. 一种电子设备，所述设备包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现权利要求1至11任一项所述的方法。
  14. 一种计算机存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现权利要求1至11任一项所述的方法。
  15. 一种计算机程序产品，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行权利要求1至11任一项所述的方法。
PCT/CN2021/139717 2021-01-15 2021-12-20 一种商品文案处理方法、装置、电子设备、介质和程序 WO2022151923A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110057323.4A CN113744020A (zh) 2021-01-15 2021-01-15 一种商品文案处理方法、装置、电子设备和存储介质
CN202110057323.4 2021-01-15

Publications (1)

Publication Number Publication Date
WO2022151923A1 true WO2022151923A1 (zh) 2022-07-21

Family

ID=78728203

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/139717 WO2022151923A1 (zh) 2021-01-15 2021-12-20 一种商品文案处理方法、装置、电子设备、介质和程序

Country Status (2)

Country Link
CN (1) CN113744020A (zh)
WO (1) WO2022151923A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117312533A (zh) * 2023-11-27 2023-12-29 乐刷科技有限公司 基于人工智能模型的文案生成方法、装置、设备及介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113744020A (zh) * 2021-01-15 2021-12-03 北京沃东天骏信息技术有限公司 一种商品文案处理方法、装置、电子设备和存储介质
CN115238198A (zh) * 2022-09-26 2022-10-25 江苏银承网络科技股份有限公司 文案信息推送方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160085816A1 (en) * 2014-09-19 2016-03-24 Kabushiki Kaisha Toshiba Information processing apparatus, information processing system, information processing method, and recording medium
CN109544201A (zh) * 2018-10-11 2019-03-29 北京奇虎科技有限公司 一种广告文案的生成方法、装置及电子设备
CN110163703A (zh) * 2018-02-12 2019-08-23 阿里巴巴集团控股有限公司 一种分类模型建立方法、文案推送方法和服务器
CN110765273A (zh) * 2019-09-17 2020-02-07 北京三快在线科技有限公司 推荐文案生成方法、装置、电子设备及可读存储介质
CN111782784A (zh) * 2020-06-24 2020-10-16 京东数字科技控股有限公司 一种文案生成方法、装置、电子设备及存储介质
CN113744020A (zh) * 2021-01-15 2021-12-03 北京沃东天骏信息技术有限公司 一种商品文案处理方法、装置、电子设备和存储介质

Family Cites Families (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105096152A (zh) * 2014-05-20 2015-11-25 阿里巴巴集团控股有限公司 一种基于商品热度的操作执行方法及装置
CN106469382A (zh) * 2015-08-14 2017-03-01 阿里巴巴集团控股有限公司 闲置商品对象信息处理方法及装置
CN106919543A (zh) * 2015-12-24 2017-07-04 阿里巴巴集团控股有限公司 确定商品对象标题文本的方法及装置
CN107730337A (zh) * 2016-08-12 2018-02-23 北京京东尚科信息技术有限公司 信息推送方法和装置
CN106326375A (zh) * 2016-08-15 2017-01-11 合肥华凌股份有限公司 一种商品推荐方法、推荐装置及冰箱
CN110135915B (zh) * 2016-08-22 2023-05-02 北京京东尚科信息技术有限公司 商品推荐方法和系统
CN107437203A (zh) * 2017-08-07 2017-12-05 北京京东尚科信息技术有限公司 信息推送方法、装置、电子装置及计算机可读介质
CN109214893A (zh) * 2018-08-31 2019-01-15 深圳春沐源控股有限公司 商品推荐方法、推荐系统及计算机装置
CN109670858A (zh) * 2018-11-21 2019-04-23 北京三快在线科技有限公司 一种优惠券发放方法、装置、电子设备及存储介质
CN111242709A (zh) * 2018-11-28 2020-06-05 北京京东尚科信息技术有限公司 一种消息推送方法及其装置、设备、存储介质
CN111523954A (zh) * 2019-02-03 2020-08-11 阿里巴巴集团控股有限公司 信息处理方法、信息展示方法及装置
CN110097433B (zh) * 2019-05-14 2022-05-17 苏州大学 基于注意力机制的推荐方法、装置、设备及存储介质
CN110263161B (zh) * 2019-05-29 2023-09-26 创新先进技术有限公司 一种信息的处理方法、装置及设备
CN112131848A (zh) * 2019-06-25 2020-12-25 北京沃东天骏信息技术有限公司 文案信息生成方法及装置、存储介质、电子设备
CN110532462A (zh) * 2019-07-25 2019-12-03 北京三快在线科技有限公司 一种推荐方法、装置、设备及可读存储介质
CN110674620A (zh) * 2019-09-20 2020-01-10 阿里巴巴（中国）有限公司 目标文案生成方法、装置、介质及电子设备
CN110852793A (zh) * 2019-10-28 2020-02-28 北京深演智能科技股份有限公司 文案推荐方法及装置、电子设备
CN111178972A (zh) * 2019-12-30 2020-05-19 北京每日优鲜电子商务有限公司 消息推送方法、装置、存储介质及设备
CN111311351B (zh) * 2020-01-15 2023-09-01 上海风秩科技有限公司 文案推荐方法、装置、计算机设备和可读存储介质
CN111586169A (zh) * 2020-05-07 2020-08-25 苏宁云计算有限公司 一种商品营销信息推送方法及系统
CN111681086A (zh) * 2020-06-16 2020-09-18 上海风秩科技有限公司 商品推荐方法、装置、计算机设备和可读存储介质
CN112102037B (zh) * 2020-09-16 2021-05-07 广州伊的家网络科技有限公司 一种基于大数据的直播电商平台商品内容智能推送管理系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160085816A1 (en) * 2014-09-19 2016-03-24 Kabushiki Kaisha Toshiba Information processing apparatus, information processing system, information processing method, and recording medium
CN110163703A (zh) * 2018-02-12 2019-08-23 阿里巴巴集团控股有限公司 一种分类模型建立方法、文案推送方法和服务器
CN109544201A (zh) * 2018-10-11 2019-03-29 北京奇虎科技有限公司 一种广告文案的生成方法、装置及电子设备
CN110765273A (zh) * 2019-09-17 2020-02-07 北京三快在线科技有限公司 推荐文案生成方法、装置、电子设备及可读存储介质
CN111782784A (zh) * 2020-06-24 2020-10-16 京东数字科技控股有限公司 一种文案生成方法、装置、电子设备及存储介质
CN113744020A (zh) * 2021-01-15 2021-12-03 北京沃东天骏信息技术有限公司 一种商品文案处理方法、装置、电子设备和存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117312533A (zh) * 2023-11-27 2023-12-29 乐刷科技有限公司 基于人工智能模型的文案生成方法、装置、设备及介质
CN117312533B (zh) * 2023-11-27 2024-02-27 乐刷科技有限公司 基于人工智能模型的文案生成方法、装置、设备及介质

Also Published As

Publication number Publication date
CN113744020A (zh) 2021-12-03

Similar Documents

Publication Publication Date Title
WO2022151923A1 (zh) 一种商品文案处理方法、装置、电子设备、介质和程序
Ngai et al. Machine learning in marketing: A literature review, conceptual framework, and research agenda
Erdmann et al. Search engine optimization: The long-term strategy of keyword choice
WO2022095701A1 (zh) 一种对象推荐方法及装置、设备、存储介质
CN108777701B (zh) 一种确定信息受众的方法及装置
TW201520790A (zh) 個性化資料搜尋方法和裝置
KR102505103B1 (ko) 온라인 커머스 데이터 기반의 큐레이션 서비스를 제공하는 방법 및 장치
Tanaka et al. Inferring latent triggers of purchases with consideration of social effects and media advertisements
US11615455B2 (en) Systems and methods for using keywords extracted from reviews
WO2023070844A1 (zh) 个性化折扣券的智能设计方法、电子装置和存储介质
CN112148973A (zh) 一种信息推送的数据处理方法及装置
US11393013B2 (en) Method, non-transitory computer-readable device, and system for intelligent listing creation
Zheng et al. A scalable purchase intention prediction system using extreme gradient boosting machines with browsing content entropy
CN111429214B (zh) 一种基于交易数据的买卖双方匹配方法及装置
CN113763035A (zh) 广告投放的效果预测方法、装置、计算机设备和存储介质
Nurmi et al. Promotionrank: Ranking and recommending grocery product promotions using personal shopping lists
Srivastava et al. Analyzing online consumer purchase psychology through hybrid machine learning
CN113254780A (zh) 信息处理的方法、装置、电子设备及计算机存储介质
CN109658195B (zh) 一种商品展示决策方法
US11657107B2 (en) Systems and methods for using keywords extracted from reviews
CN112015970A (zh) 产品推荐方法、相关设备及计算机存储介质
KR102429104B1 (ko) 인공지능에 기반한 상품 카탈로그 자동 분류 시스템
CN113781171A (zh) 信息推送方法、装置、设备及存储介质
CN113935529A (zh) 客户生命周期价值的确定方法、装置、设备及介质
CN110020136B (zh) 对象推荐方法及相关设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21919110

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 18/10/2023)