CN115482054A - Commodity recommendation method and device, storage medium and electronic equipment - Google Patents

Commodity recommendation method and device, storage medium and electronic equipment Download PDF

Info

Publication number
CN115482054A
CN115482054A CN202110666837.XA CN202110666837A CN115482054A CN 115482054 A CN115482054 A CN 115482054A CN 202110666837 A CN202110666837 A CN 202110666837A CN 115482054 A CN115482054 A CN 115482054A
Authority
CN
China
Prior art keywords
commodity
data
user
recommendation
interactive
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110666837.XA
Other languages
Chinese (zh)
Inventor
王珅
李天浩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jingdong Technology Holding Co Ltd
Original Assignee
Jingdong Technology Holding Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jingdong Technology Holding Co Ltd filed Critical Jingdong Technology Holding Co Ltd
Priority to CN202110666837.XA priority Critical patent/CN115482054A/en
Publication of CN115482054A publication Critical patent/CN115482054A/en
Pending legal-status Critical Current

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]
    • G06Q30/0631—Item recommendations
    • 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]
    • G06Q30/0641—Shopping interfaces
    • G06Q30/0643—Graphical representation of items or shoppers

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention provides a commodity recommendation method and device, a storage medium and electronic equipment, wherein the method comprises the following steps: when a commodity display instruction sent by a user through E-commerce software is received, acquiring user characteristic data, the above interactive characteristic data and commodity characteristic data; sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a multi-task commodity recommendation model for processing so that the multi-task commodity recommendation model updates a commodity recommendation sequence of each commodity recommendation version block contained in E-commerce software; and sending each commodity recommendation sequence to E-commerce software, so that the E-commerce software renders each commodity in each commodity recommendation sequence to a display page corresponding to a commodity recommendation version corresponding to the commodity, and recommends the commodity to a user. And providing multi-type characteristic data for the multi-task commodity recommendation model, deeply mining the current shopping preference of the user, recommending commodities which best meet the current shopping preference to the user, and improving the click rate and the conversion rate of the commodities.

Description

Commodity recommendation method and device, storage medium and electronic equipment
Technical Field
The present invention relates to the field of computer technologies, and in particular, to a method and an apparatus for recommending a commodity, a storage medium, and an electronic device.
Background
With the development of the e-commerce industry, various e-commerce software layers are in a large quantity, and in order to provide users with a good shopping experience and improve the economic income of merchants, a commodity recommendation system is generally used for recommending commodities meeting the interests and requirements of the users for the users, so that the click rate and the conversion rate of the commodities can be improved.
At present, the commodity recommended to the user is determined only based on historical browsing data of the user, and the data applied in the conventional commodity recommending mode is single, so that the commodity recommended to the user does not meet the latest shopping interest and preference of the user, and the click rate and the conversion rate of the commodity recommended to the user are greatly reduced.
Disclosure of Invention
In view of this, the present invention provides a method and an apparatus for recommending a commodity, a storage medium, and an electronic device, so as to recommend a commodity that better meets the current shopping interest and preference of a user to the user, thereby improving the click rate and the conversion rate of the commodity recommended to the user.
In order to achieve the above object, the embodiments of the present invention provide the following technical solutions:
a method of merchandise recommendation, comprising:
when a commodity display instruction sent by a user through E-commerce software is received, obtaining user characteristic data, above interactive characteristic data and commodity characteristic data of the E-commerce software, wherein the above interactive characteristic data are data acquired by the E-commerce software based on the commodity display instruction and used for interaction between the user and the E-commerce software;
sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multitask commodity recommendation model;
triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation version block contained in the E-commerce software;
and sending each commodity recommendation sequence to the E-commerce software, so that the E-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation version block corresponding to the commodity, and recommends the commodity to the user.
Optionally, the method for acquiring the user characteristic data, the above interaction characteristic data, and the commodity characteristic data of the e-commerce software includes:
acquiring user identification information of the user, acquiring attribute information and historical interactive behavior data of the user based on the user identification information, filtering the historical interactive behavior data to obtain qualified historical interactive behavior data, and forming the attribute information and the qualified historical interactive behavior data into the user characteristic data;
analyzing the commodity display instruction, determining the instruction type of the commodity display instruction, acquiring interactive data corresponding to the instruction type, and taking the interactive data as the above interactive feature data;
determining each commodity recommendation block contained in the E-commerce software, acquiring commodity to-be-displayed information of each commodity recommendation block, and forming the commodity characteristic data based on the commodity to-be-displayed information.
Optionally, in the method, the analyzing the commodity display instruction, determining an instruction type of the commodity display instruction, acquiring interactive data corresponding to the instruction type, and using the interactive data as the interactive feature data includes:
analyzing the commodity display instruction to obtain a type identifier in the commodity display instruction, and judging whether the instruction type of the commodity display instruction is a skip type or not based on the type identifier;
if the instruction type of the commodity display instruction is a skip type, determining interaction data generated when the user executes an interaction behavior on a page currently displayed by the E-commerce software, and taking the interaction data as the above interaction feature data;
and if the instruction type of the commodity display instruction is not the jump type, determining that the instruction type of the commodity display instruction is the starting type, generating data representing that the interaction between the user and the E-commerce software is zero, and taking the data as the above interactive feature data.
Optionally, the triggering the multi-task product recommendation model to process the user feature data, the above interaction feature data, and the product feature data to update the product recommendation sequence of each product recommendation block included in the e-commerce software includes:
triggering the multitask commodity recommendation model to call an input feature layer to process the user feature data, the above interactive feature data and the commodity feature data to obtain user feature discrete data corresponding to the user feature data, the above interactive feature discrete data corresponding to the above interactive feature data and commodity feature discrete data corresponding to the commodity feature data;
the multitask commodity recommendation model calls a feature embedding layer to carry out low-dimensional embedding mapping processing on the user feature discrete data, the above interactive feature discrete data and the commodity feature discrete data respectively to obtain a user feature matrix corresponding to the user feature data, an above interactive feature matrix corresponding to the above interactive feature data and a commodity feature matrix corresponding to the commodity feature data;
the multitask commodity recommendation model calls a feature extraction layer to perform feature splicing on the user feature matrix, the above interactive feature matrix and the commodity feature matrix to obtain commodity recommendation features corresponding to the user;
inputting the commodity recommendation characteristics to a multitask recommendation layer in the multitask commodity recommendation model, and enabling the multitask recommendation layer to process the commodity recommendation characteristics to obtain a global commodity sequence of each commodity recommendation block, wherein each global commodity sequence comprises at least one commodity, and each commodity comprises an estimated click rate and an estimated conversion rate of the commodity;
obtaining a fusion score of each commodity in each global commodity sequence based on the estimated click rate and the estimated conversion rate of each commodity in each global commodity sequence;
and for each commodity recommendation block, sorting commodities in the global commodity sequence of the commodity recommendation block according to the sequence of fusion scores from high to low, and selecting the first N commodities to form the commodity recommendation sequence of the commodity recommendation block, wherein N is greater than or equal to 1.
The above method, optionally, further includes:
after the E-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to the commodity recommendation layout block corresponding to the commodity, data generated when the user carries out interaction behaviors on the display page are collected, and the data are stored in a preset historical behavior database.
An article recommendation device comprising:
the electronic commerce system comprises an acquisition unit, a display unit and a display unit, wherein the acquisition unit is used for acquiring user characteristic data, above interactive characteristic data and commodity characteristic data of electronic commerce software when a commodity display instruction sent by a user through the electronic commerce software is received, and the above interactive characteristic data is data acquired by the electronic commerce software based on the commodity display instruction and used for interaction between the user and the electronic commerce software;
the sending unit is used for sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multitask commodity recommendation model;
the triggering unit is used for triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software;
and the rendering unit is used for sending each commodity recommendation sequence to the e-commerce software, so that the e-commerce software renders each commodity in each commodity recommendation sequence to a display page corresponding to the commodity recommendation block corresponding to the commodity, and the commodity is recommended to the user.
The above apparatus, optionally, the obtaining unit includes:
the first obtaining subunit is configured to obtain user identification information of the user, obtain attribute information and historical interactive behavior data of the user based on the user identification information, filter the historical interactive behavior data to obtain qualified historical interactive behavior data, and combine the attribute information and the qualified historical interactive behavior data into the user feature data;
the analysis subunit is used for analyzing the commodity display instruction, determining the instruction type of the commodity display instruction, acquiring interactive data corresponding to the instruction type, and taking the interactive data as the interactive characteristic data;
and the determining subunit is used for determining each commodity recommendation block contained in the E-commerce software, acquiring the commodity to-be-displayed information of each commodity recommendation block, and forming the commodity characteristic data based on the commodity to-be-displayed information.
The above apparatus, optionally, the parsing subunit includes:
the acquisition module is used for analyzing the commodity display instruction, acquiring a type identifier in the commodity display instruction, and judging whether the instruction type of the commodity display instruction is a skip type or not based on the type identifier;
the determining module is used for determining interactive data generated when the user executes an interactive behavior on a page currently displayed by the E-commerce software if the instruction type of the commodity display instruction is a skip type, and taking the interactive data as the above interactive feature data;
and the generation module is used for determining that the instruction type of the commodity display instruction is a starting type if the instruction type of the commodity display instruction is not a skip type, generating data representing that interaction between the user and the E-commerce software is zero, and taking the data as the above interactive feature data.
The above apparatus, optionally, the triggering unit includes:
the triggering subunit is used for triggering the multitask commodity recommendation model to call an input feature layer to process the user feature data, the above interactive feature data and the commodity feature data to obtain user feature discrete data corresponding to the user feature data, the above interactive feature discrete data corresponding to the above interactive feature data and commodity feature discrete data corresponding to the commodity feature data;
the processing subunit is used for the multitask commodity recommendation model calling a feature embedding layer to perform low-dimensional embedding mapping processing on the user feature discrete data, the above interactive feature discrete data and the commodity feature discrete data respectively to obtain a user feature matrix corresponding to the user feature data, an above interactive feature matrix corresponding to the above interactive feature data and a commodity feature matrix corresponding to the commodity feature data;
the characteristic splicing subunit is used for calling a characteristic extraction layer by the multi-task commodity recommendation model to perform characteristic splicing on the user characteristic matrix, the above interactive characteristic matrix and the commodity characteristic matrix to obtain a commodity recommendation characteristic corresponding to the user;
the input subunit is configured to input the commodity recommendation feature to a multitask recommendation layer in the multitask commodity recommendation model, so that the multitask recommendation layer processes the commodity recommendation feature to obtain a global commodity sequence of each commodity recommendation block, where each global commodity sequence includes at least one commodity, and each commodity includes an estimated click rate and an estimated conversion rate of the commodity;
the second obtaining subunit is used for obtaining a fusion score of each commodity in each global commodity sequence based on the estimated click rate and the estimated conversion rate of each commodity in each global commodity sequence;
and the sorting subunit is configured to sort, for each commodity recommendation block, commodities in the global commodity sequence of the commodity recommendation block according to a sequence from high to low fusion scores, and then select the first N commodities to form the commodity recommendation sequence of the commodity recommendation block, where N is greater than or equal to 1.
The above apparatus, optionally, further comprises:
and the acquisition unit is used for acquiring data generated when the user carries out interaction behavior on the display page after the E-commerce software renders each commodity included in each commodity recommendation sequence to the display page corresponding to the commodity recommendation block corresponding to the commodity, and storing the data in a preset historical behavior database.
A storage medium comprising stored instructions, wherein the instructions, when executed, control a device on which the storage medium is located to perform a merchandise recommendation method as described above.
An electronic device comprising a memory, and one or more instructions, wherein the one or more instructions are stored in the memory and configured to be executed by one or more processors to perform a method of merchandise recommendation as described above.
Compared with the prior art, the invention has the following advantages:
the invention provides a commodity recommendation method and device, a storage medium and electronic equipment, wherein the method comprises the following steps: when a commodity display instruction sent by a user through E-commerce software is received, acquiring user characteristic data, the above interactive characteristic data and commodity characteristic data of the E-commerce software; sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to the multitask commodity recommendation model; triggering a multi-task commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software; and sending each commodity recommendation sequence to the E-commerce software, so that the E-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation layout block corresponding to the commodity, and recommends the commodity to the user. The multi-dimensional feature data are provided for the multi-task commodity recommendation model, the current shopping preference and interest of the user can be deeply mined, analysis is performed from multiple angles, the commodity which best meets the current shopping preference and interest is recommended to the user, and the click rate and the conversion rate of the recommended commodity are improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart of a method for recommending a commodity according to an embodiment of the present invention;
fig. 2 is a flowchart of another method of a commodity recommendation method according to an embodiment of the present invention;
FIG. 3 is a flowchart of another method of a merchandise recommendation method according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a commodity recommending apparatus according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
In this application, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
When commodities are recommended to a user traditionally, the commodities are recommended to the user only based on historical browsing data of the user, and the commodities recommended in the mode are difficult to meet the latest shopping interest and preference of the user, so that the click rate and the conversion rate of the commodities recommended to the user are greatly reduced. In view of this, the invention provides a commodity recommendation method, which is used for accurately recommending commodities meeting the shopping interest and preference of a user to the user, so as to improve the click rate and the conversion rate of the commodities recommended to the user.
The method provided by the invention can be applied to a commodity recommendation system composed of a plurality of general or special computing devices or configurations, the execution subject of the method is a processor or an actuator of the commodity recommendation system, and referring to fig. 1, a method flow chart of the commodity recommendation method provided by the embodiment of the invention is specifically described as follows:
s101, when a commodity display instruction sent by a user through E-commerce software is received, obtaining user characteristic data, above interactive characteristic data and commodity characteristic data of the E-commerce software, wherein the above interactive characteristic data are acquired by the E-commerce software based on the commodity display instruction, and the user and the E-commerce software interact with each other.
In the method provided by the embodiment of the invention, two situations exist when a user sends a commodity display instruction through E-commerce software, one situation is that the E-commerce software is started, and the user needs to enter other commodity recommending modules of the E-commerce software; the other is that the E-commerce software is not started, and a user needs to start the E-commerce software; further, the E-commerce software is various shopping software or software for selling commodities.
The specific process of acquiring the feature data of the user, the above interactive feature data, and the commodity feature data of the e-commerce software is described as follows with reference to fig. 2:
s201, obtaining user identification information of the user, obtaining attribute information and historical interactive behavior data of the user based on the user identification information, filtering the historical interactive behavior data to obtain qualified historical interactive behavior data, and forming the attribute information and the qualified historical interactive behavior data into the user characteristic data.
The user identification information includes but is not limited to user code, user identification number and other information, wherein the user identification number may be an identity card number or a mobile phone number of the user, and the user code may be an identity number which is allocated to the user by the system when the user registers the e-commerce software. The attribute information of the user includes but is not limited to the sex, age, income, academic calendar and other information of the user; the historical interactive behavior data includes, but is not limited to, a commodity clicking sequence, a commodity collecting sequence, a commodity ordering sequence and the like of a user within a preset time period, the historical interactive behavior data can be obtained from a preset historical behavior database, the preset time period is set according to actual requirements, and specifically, if the preset time period is the first 5 days of the current use date of the e-commerce software, and the current use date of the exemplary e-commerce software is 2021 year, 5 month, 11 days, the preset time period is 2021 year, 5 month, 6 # to 10 # and is the preset time period.
And acquiring attribute information of the user in a user attribute library according to the user identification information, and acquiring historical interactive behavior data of the user in a historical behavior database according to the user identification information.
Click data of each commodity clicked by a user in a preset time period is contained in the click commodity sequence; the collected commodity sequence comprises the collected data of each commodity collected by the user in a preset time period; the order placing commodity sequence comprises order placing data of each commodity purchased by the user in a preset time period.
When filtering the historical interactive behavior data, screening each click data, each collection data and each order placing data in the historical interactive behavior data through a set screening strategy, deleting each unqualified data, and forming each qualified data into qualified historical interactive behavior data; specifically, different screening conditions exist for different types of data, for example, click data is illustrated here, and when the frequency of clicking on a certain product by a user in a certain time period is higher than a preset value, the click data of the product by the user in the time period is determined as ineligible data. The purpose of filtering the historical interactive behavior data is to filter data generated by the user's click cheating behavior and the order-swiping behavior, ensure the health of the data environment, and improve the accuracy rate of the commodities recommended to the user to meet the purchasing interest and demand of the user.
S202, analyzing the commodity display instruction, determining the instruction type of the commodity display instruction, acquiring interactive data corresponding to the instruction type, and taking the interactive data as the interactive feature data.
Determining an instruction type of the commodity display instruction so as to obtain interactive data corresponding to the instruction type, wherein the instruction type is associated with a condition that a user sends the commodity display instruction, and further, a specific process of obtaining the interactive data corresponding to the instruction type of the commodity display instruction is as follows:
analyzing the commodity display instruction to obtain a type identifier in the commodity display instruction, and judging whether the instruction type of the commodity display instruction is a skip type or not based on the type identifier;
if the instruction type of the commodity display instruction is a skip type, determining interactive data generated when the user carries out interactive behavior on a page currently displayed by the E-commerce software, and taking the interactive data as the above interactive feature data;
and if the instruction type of the commodity display instruction is not the jump type, determining that the instruction type of the commodity display instruction is the starting type, generating data representing that the interaction between the user and the E-commerce software is zero, and taking the data as the above interactive feature data.
The type identifier in the invention is used for representing the instruction type of the commodity display instruction, specifically, the instruction type of the commodity display instruction has two types, one is a jump type, and the other is a starting type; when the instruction type of the commodity display instruction is a skip type, the instruction type indicates that the user has opened the E-commerce software, the E-commerce software displays commodities to the user, and at the moment, the user needs to enter display pages of other commodity recommendation modules of the E-commerce software; when the instruction type of the commodity display instruction is a starting instruction, the instruction indicates that the user just needs to open the E-commerce software, and the E-commerce software displays the recommended commodity for the user.
When judging whether the instruction type of the commodity display instruction is a skip type based on the type identifier, judging whether the type identifier is the skip identifier, and if the type identifier is the skip identifier, determining that the instruction type of the commodity display instruction is the skip type; if the type identification is not the jump identification, the type identification is determined to be a starting identification, and the instruction type of the commodity display instruction is determined to be a starting type.
When the instruction type of the commodity display instruction is a skip type, determining interactive data generated when a user executes an interactive behavior on a page currently displayed by E-commerce software, and using the interactive data as the above interactive feature data, further, the interactive behavior executed by the user on the page currently displayed by the E-commerce software can include two types, one type is a click behavior, the other type is a skip behavior, further, the interactive data generated by the click behavior is the above click commodity data, and the above click commodity data includes data of each commodity clicked by the user on the page currently displayed by the E-commerce software; and the interaction data generated by the skipping action is skipped commodity data, and the skipped commodity data comprises data of commodities skipped by the user on the current displayed page of the E-commerce software.
Further, when the instruction type of the commodity display instruction is not the skip type, determining that the instruction type of the commodity display instruction is the start type, at this time, the commodity display instruction indicates that the user is about to open the e-commerce software, and at this time, interactive data representing that the interaction between the user and the e-commerce software is zero needs to be generated, namely, the interactive data is blank data.
S203, determining each commodity recommendation block contained in the E-commerce software, acquiring commodity to-be-displayed information of each commodity recommendation block, and forming the commodity characteristic data based on each commodity to-be-displayed information.
The commodity recommending plate is specifically a commodity main plate or a commodity advertisement recommending plate of E-commerce software, the information to be displayed of commodities of different commodity recommending plates is different, the different commodity recommending plates correspond to different types of commodities, and the commodities corresponding to the commodity recommending plates are crossed, namely one commodity can correspond to a plurality of commodity recommending plates. The information to be displayed of the commodity comprises information such as commodity names, commodity categories and commodity discrete codes of commodities corresponding to the commodity recommendation plates, and commodity characteristic data is formed on the basis of the information to be displayed of the commodities of the commodity recommendation plates.
In the method provided by the embodiment of the present invention, the execution sequence of the three steps S201, S202, and S203 in fig. 2 may be changed and executed, and the three steps may also be executed in parallel.
The method and the device acquire the attribute information and the historical interactive behavior data of the user based on the user identification information of the user, filter the historical interactive behavior data and ensure the health of a data environment; the above interactive feature data are obtained based on the commodity display instruction, so that interactive data executed by the user in the front-sequence commodity recommendation block can be obtained, and the latest shopping demand and shopping interest of the user can be obtained; the commodity feature data of the E-commerce software is obtained in a multi-path recall mode, and commodity feature software of global commodities of the E-commerce software can be obtained; therefore, multi-dimensional characteristic data can be obtained, and various characteristic data are provided for commodities recommended by a user through subsequent analysis.
S102, sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multitask commodity recommendation model.
The multi-task commodity recommendation model is a model for performing overall combined modeling on each commodity recommendation block in E-commerce software by adopting a multi-task sequencing framework, the data of the commodity recommendation blocks are integrated, the overall multi-task modeling of the multi-block can be completed through a single model, and information such as commodity estimated click rate and estimated conversion rate under each commodity recommendation block is output.
The user characteristic data, the above interactive characteristic data and the commodity characteristic data are sent to the multi-task commodity recommendation model, so that the characteristic data can be provided for the multi-task commodity recommendation model from multiple dimensions, and the probability that commodities recommended by the multi-task commodity recommendation model meet the shopping requirements and interests of the user is improved.
S103, triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update the commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software.
The multitask commodity recommendation model is composed of an input feature layer, a feature embedding layer, a feature extraction layer and a multitask recommendation layer, different layers have different functions, the multitask commodity recommendation model processes user feature data, above interactive feature data and commodity feature data to update a commodity recommendation sequence of each commodity recommendation block included in E-commerce software, and the specific description is as follows with reference to fig. 3:
s301, triggering the multi-task commodity recommendation model to call an input feature layer to process the user feature data, the above interactive feature data and the commodity feature data to obtain user feature discrete data corresponding to the user feature data, the above interactive feature discrete data corresponding to the above interactive feature data and the commodity feature discrete data corresponding to the commodity feature data.
The user characteristic data comprises attribute information and qualified historical interactive behavior data, and the attribute information comprises but is not limited to the information of the age, the gender, the academic calendar, the city and the like of the user; the qualified historical interaction behavior data includes, but is not limited to, a click commodity sequence, a collection commodity sequence, an order placing commodity sequence, and the like. Wherein, the characteristic data of age and the like with size relation adopts a maximum and minimum normalization lead-in model; discrete characteristics such as gender, academic calendar, city, qualified historical interactive behavior data and the like are subjected to discrete numerical value coding according to the corresponding dimension representation space size, so that user characteristic discrete data corresponding to the user characteristic data can be obtained; for example, assuming that 1000w different commodities exist in the commodity library, 1-1000w integer values are respectively used to perform discrete numerical coding on the commodities in the commodity library, and the commodities not covered by the commodity mapping set in the sample set are uniformly filled with the commodity missing value by using an integer 0, while the gender has two categories, that is, the gender is 0, and the gender is 1, so that the gender is discrete numerical coding is performed.
The commodity characteristic data comprises but is not limited to commodity names, commodity categories, discrete codes of commodities to be estimated and the like, the commodity names are respectively subjected to word segmentation from word and word granularity, discrete mapping coding is carried out on term (representing words and expressions) after word segmentation, meanwhile, the commodity categories are classified according to the categories such as electronic products, fresh products and the like for category mapping, the discrete codes of the commodities to be estimated are subjected to discrete numerical coding by using a commodity mapping table which is the same as qualified historical interactive behavior data, and therefore the discrete data of the commodity characteristics corresponding to the commodity characteristic data can be obtained.
The above interactive feature data include, but are not limited to, a user timestamp, the above clicked commodity data, the above skipped commodity data, and the like, and discrete numerical value coding is performed on commodities in the above clicked commodity data and the above skipped commodity data by using a commodity mapping table the same as the user feature data, so that the above interactive feature discrete data corresponding to the above interactive feature data can be obtained; further, for products directly presented to the user in interfaces such as a product master block or a product search result page, the above interactive feature data of the user in the current terminal session is lacked, the above interactive feature data of the scene with the missing interactive behavior is represented by a code [0], that is, the above click product data and the above skip product data in the above interactive feature data of the scene with the missing interactive behavior are represented by a code [0 ].
S302, the multitask commodity recommendation model calls a feature embedding layer to perform low-dimensional embedding mapping processing on the user feature discrete data, the above interactive feature discrete data and the commodity feature discrete data respectively to obtain a user feature matrix corresponding to the user feature data, an above interactive feature matrix corresponding to the above interactive feature data and a commodity feature matrix corresponding to the commodity feature data.
The user characteristic discrete data, the commodity characteristic discrete data and the interactive characteristic discrete data are high-dimensional discrete input signals; the multitask commodity recommendation model is used for performing embedding (embedding layer for mapping sparse high-dimensional discrete vectors into low-dimensional dense vectors) mapping on user characteristic discrete data, commodity characteristic discrete data and the above interactive characteristic discrete data output from the input characteristic layer, namely performing low-dimensional embedding mapping processing, so that high-dimensional discrete input signals are converted into low-dimensional dense signals, further, after the low-dimensional embedding mapping processing is performed on the user characteristic discrete data, the commodity characteristic discrete data and the above interactive characteristic discrete data, a user characteristic matrix corresponding to the user characteristic data, an above interactive characteristic matrix corresponding to the above interactive characteristic data and a commodity characteristic matrix corresponding to the commodity characteristic data are obtained, and therefore the data can be mapped into a two-dimensional embedding matrix. The feature embedding layer maps each discrete data into a corresponding feature matrix, so that the generalization effect of the multi-task commodity recommendation model is improved.
S303, calling a feature extraction layer by the multi-task commodity recommendation model to perform feature splicing on the user feature matrix, the above interactive feature matrix and the commodity feature matrix to obtain commodity recommendation features corresponding to the user.
The feature extraction layer reduces the user feature matrix, the above interactive feature matrix and the commodity feature matrix into one-dimensional embedding vectors by using modules such as a pooling layer, an LSTM or a transform according to different dimensions, and performs feature splicing with the one-dimensional embedding vectors such as gender and commodities to be estimated and floating point features such as age of the user, so as to obtain commodity recommendation features corresponding to the user.
Wherein, the LSTM is a long-short term memory network, and english is called as: long Short-Term Memory; the transform is a sequence-to-sequence model architecture proposed in the article "Attention All you need"; embedding is an embedding layer for mapping sparse high-dimensional discrete vectors into low-dimensional dense vectors.
S304, inputting the commodity recommendation characteristics to a multitask recommendation layer in the multitask commodity recommendation model, and enabling the multitask recommendation layer to process the commodity recommendation characteristics to obtain a global commodity sequence of each commodity recommendation block, wherein each global commodity sequence comprises at least one commodity, and each commodity comprises an estimated click rate and an estimated conversion rate of the commodity.
The multitask recommendation layer carries out multitask sequencing modeling by adopting a share-bottom or MMOE and ESMM multitask sequencing model learning framework to extract high-order characteristics, and a CTR estimation task and a CVR estimation task of each commodity recommendation block are respectively constructed on the multitask classification layer; wherein, MMOE is called Multi-gate texture-of-Expert in English, and is a Multi-task learning architecture provided for Google; the ESMM is called an Enterre Space Multi-Task Model in English, and provides a multitask learning framework for Alibama. When the commodity recommendation block is a main commodity block, the CTR estimation task and the CVR estimation task of the commodity recommendation block respectively correspond to different classifiers; when the commodity recommendation block is a commodity advertisement recommendation block, the CTR estimation task and the CVR estimation task of the commodity recommendation block respectively correspond to different classifiers, wherein the commodity advertisement recommendation block and the commodity advertisement recommendation block correspond to different classifiers, namely, the multi-task classification layer comprises 4 classifiers.
Furthermore, CTR is click rate, and the CVR estimation task is used for outputting the estimated click rate of each commodity corresponding to the commodity recommended block; and the CVR is a conversion rate, and the CVR estimation task is used for outputting the estimated conversion rate of each commodity corresponding to the commodity recommended block.
The multi-task recommendation layer processes the commodity features to obtain a global commodity sequence of each commodity recommendation block, wherein the global commodity sequence comprises at least one commodity, each commodity in the global commodity sequence is sequenced according to the estimated conversion rate and the estimated click rate of the commodity, and the number of the commodities in the global commodity sequence can be set according to actual requirements.
S305, obtaining a fusion score of each commodity in each global commodity sequence based on the estimated click rate and the estimated conversion rate of each commodity in each global commodity sequence.
For each commodity in each global upper column, calculating a fusion score of the commodity according to the estimated click rate and the estimated conversion rate of the commodity; the formula applied by the calculation process is as follows:
score = a CTR + b CVR, where score represents the fusion score of the commodity, CTR is the estimated click rate of the commodity, CVR is the estimated conversion rate of the commodity, a and b are preset weight values, and the specific values of a and b can be set according to actual requirements.
S306, for each commodity recommendation block, sorting commodities in the global commodity sequence of the commodity recommendation block according to the sequence of fusion scores from high to low, and selecting the first N commodities to form the commodity recommendation sequence of the commodity recommendation block, wherein N is greater than or equal to 1.
After the fusion score of each commodity in the global commodity sequence of each commodity recommendation block is obtained, the commodity recommendation sequence of the commodity recommendation block is obtained according to the fusion score of the commodity, and when the commodity recommendation sequence of the commodity recommendation block is obtained, one of the modes can be as follows: arranging all commodities in the global commodity sequence from high to low according to the fusion score, and selecting the first N commodities to form a commodity recommendation sequence, wherein N is greater than or equal to 1, and further, N is any positive integer; the other mode is as follows: and for each commodity in the global commodity sequence of the commodity recommendation block, sequentially selecting the commodities according to the sequence from high to low of the fusion score, stopping selecting the commodities when the number of the selected commodities reaches a preset value, and arranging the selected commodities according to the sequence from high to low of the fusion score, thereby obtaining the commodity recommendation sequence of the commodity recommendation block, wherein the value range of the preset value is a positive integer greater than 1. According to the invention, the information fed back by the user to the commodities recommended by different versions is deeply fused, and the data sets of multiple versions are used for training the multi-task commodity recommendation model, so that the condition that the commodities recommended by the versions with sparse data are over-fitted is effectively avoided, and the good feeling of the user to the commodities recommended by the system can be improved.
During training of the multi-task commodity recommendation model, collected data are classified according to different plate blocks, data corresponding to each plate block are classified according to different data attributes, the data attributes include but are not limited to display attributes, click attributes, order placing attributes and the like, training data sets and testing data sets of different plate blocks are respectively constructed according to different dimensional characteristics and dimensional labels, the dimensional characteristics comprise user dimensions, recommended commodity dimensions, upper interactive behavior dimensions and the like, and the dimensional labels comprise click labels, order placing labels and the like; and mixing the training sets of the plates, independently constructing independent click CTR (click through ratio) tasks and ordering CVR (constant value response) task classification layers after sharing the characteristic representation layers for the plates based on a multi-task learning framework, performing global joint training on click rate estimation and conversion rate estimation tasks of different plates by using the mixed training sets, and finishing the training of the multi-task commodity recommendation model after the multi-task commodity recommendation model is stably trained.
In the method provided by the embodiment of the invention, the user characteristic data, the above interactive characteristic data and the commodity characteristic data are provided for the multitask commodity recommendation model, so that the multidimensional characteristic data is realized, the multitask commodity recommendation model is analyzed from multiple dimensions and multiple aspects, particularly, the above interactive characteristic data comprises the characteristic data of the interaction of the user in the previous edition, the commodity can be recommended to the user to the maximum extent according to the recent shopping interest and hobbies of the user, the commodity recommended to the user can better meet the requirements of the user, and the click rate and the conversion rate of the commodity recommended to the user can be further improved.
S104, sending each commodity recommendation sequence to the E-commerce software, and enabling the E-commerce software to render each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation layout block corresponding to the commodity so as to recommend the commodity to the user.
In the method provided by the embodiment of the invention, the commodity recommendation system sends each commodity recommendation sequence to the E-commerce software, so that the E-commerce software renders each commodity contained in each commodity recommendation sequence to the display page corresponding to the commodity recommendation block corresponding to the commodity, and the commodity is recommended to a user. Optionally, after each commodity recommendation sequence is sent to the e-commerce software, when the user triggers the interface of the e-commerce software to be converted into the interface of the commodity recommendation block selected by the user, the e-commerce software renders each commodity in the commodity recommendation sequence corresponding to the commodity recommendation block on the interface so as to recommend the commodity to the user.
Further, after the E-commerce software renders each commodity in each commodity recommendation sequence to a display interface corresponding to a commodity recommendation plate corresponding to the commodity, data generated when a user carries out interaction behaviors on the display interface is collected, wherein the interaction behaviors include but are not limited to clicking, skipping, ordering and other behaviors, and the generated data is stored in a preset historical behavior database so as to update the behavior data of the user; and recording each commodity recommended to the user as a recommendation record in a log database so as to analyze the commodity recommended to the user according to the recommendation record.
In the method provided by the embodiment of the invention, when a commodity display instruction sent by a user through E-commerce software is received, user characteristic data, the above interactive characteristic data and commodity characteristic data of the E-commerce software are obtained, wherein the above interactive characteristic data are data of interaction between the user and the E-commerce software, which are acquired by the E-commerce software based on the commodity display instruction; sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multi-task commodity recommendation model; triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software; and sending each commodity recommendation sequence to the e-commerce software, so that the e-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation block corresponding to the commodity, and recommends the commodity to the user. The multi-dimensional feature data are provided for the multi-task commodity recommendation model, so that the multi-task commodity recommendation model analyzes commodities to be recommended to the user from multiple dimensions, the current preference information of the user can be mined from the feature data of the multiple dimensions, and then the commodities recommended to the user can meet the recent shopping interest and preference of the user, so that the click rate and the conversion rate of the recommended commodities are improved.
Corresponding to the method shown in fig. 1, the present invention further provides a commodity recommendation device, which is applied to various commodity recommendation systems, each commodity recommendation system is composed of various computer terminals or various intelligent devices, the device is used for supporting the application of the method shown in fig. 1 in real life, the structural schematic diagram of the device is shown in fig. 4, and the following description specifically illustrates:
an obtaining unit 401, configured to obtain user feature data, above-mentioned interaction feature data, and commodity feature data of an e-commerce software when a commodity display instruction sent by a user through the e-commerce software is received, where the above-mentioned interaction feature data is data of interaction between the user and the e-commerce software, which is acquired by the e-commerce software based on the commodity display instruction;
a sending unit 402, configured to send the user feature data, the above interaction feature data, and the commodity feature data to a preset multitask commodity recommendation model;
a triggering unit 403, configured to trigger the multitask commodity recommendation model to process the user feature data, the above interaction feature data, and the commodity feature data, so as to update a commodity recommendation sequence of each commodity recommendation block included in the e-commerce software;
the rendering unit 404 is configured to send each of the commodity recommendation sequences to the e-commerce software, so that the e-commerce software renders each commodity included in each of the commodity recommendation sequences to a display page corresponding to a commodity recommendation layout block corresponding to the commodity, so as to recommend the commodity to the user.
In the device provided by the embodiment of the invention, when a commodity display instruction sent by a user through E-commerce software is received, user characteristic data, the above interactive characteristic data and commodity characteristic data of the E-commerce software are obtained; sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multitask commodity recommendation model; triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software; and sending each commodity recommendation sequence to the e-commerce software, so that the e-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation block corresponding to the commodity, and recommends the commodity to the user. By providing the multi-dimensional characteristic data to the multi-task commodity recommendation model, the multi-task commodity recommendation model analyzes commodities to be recommended to the user from multiple dimensions, the current preference information of the user can be mined from the multi-dimensional characteristic data, and then the commodities recommended to the user can meet the recent shopping interests and hobbies of the user, so that the click rate and the conversion rate of the recommended commodities are improved.
In the apparatus provided in the embodiment of the present invention, the obtaining unit 401 may be configured to:
the first obtaining subunit is configured to obtain user identification information of the user, obtain attribute information and historical interactive behavior data of the user based on the user identification information, filter the historical interactive behavior data to obtain qualified historical interactive behavior data, and combine the attribute information and the qualified historical interactive behavior data into the user feature data;
the analysis subunit is used for analyzing the commodity display instruction, determining the instruction type of the commodity display instruction, acquiring interactive data corresponding to the instruction type, and taking the interactive data as the interactive characteristic data;
and the determining subunit is used for determining each commodity recommendation block contained in the E-commerce software, acquiring the commodity to-be-displayed information of each commodity recommendation block, and forming the commodity characteristic data based on each commodity to-be-displayed information.
In the apparatus provided in the embodiment of the present invention, the parsing subunit may be configured to:
the acquisition module is used for analyzing the commodity display instruction, acquiring a type identifier in the commodity display instruction, and judging whether the instruction type of the commodity display instruction is a skip type or not based on the type identifier;
the determining module is used for determining interactive data generated when the user executes an interactive behavior on a page currently displayed by the E-commerce software if the instruction type of the commodity display instruction is a skip type, and taking the interactive data as the above interactive feature data;
and the generation module is used for determining that the instruction type of the commodity display instruction is a starting type if the instruction type of the commodity display instruction is not a jump type, generating data representing that the interaction between the user and the E-commerce software is zero, and taking the data as the above interactive characteristic data.
In the apparatus provided in the embodiment of the present invention, the triggering unit 403 may be configured to:
the triggering subunit is used for triggering the multitask commodity recommendation model to call an input feature layer to process the user feature data, the above interactive feature data and the commodity feature data to obtain user feature discrete data corresponding to the user feature data, the above interactive feature discrete data corresponding to the above interactive feature data and commodity feature discrete data corresponding to the commodity feature data;
the processing subunit is used for the multitask commodity recommendation model calling a feature embedding layer to respectively perform low-dimensional embedding mapping processing on the user feature discrete data, the above interactive feature discrete data and the commodity feature discrete data to obtain a user feature matrix corresponding to the user feature data, an above interactive feature matrix corresponding to the above interactive feature data and a commodity feature matrix corresponding to the commodity feature data;
the characteristic splicing subunit is used for calling a characteristic extraction layer by the multi-task commodity recommendation model to perform characteristic splicing on the user characteristic matrix, the above interactive characteristic matrix and the commodity characteristic matrix to obtain a commodity recommendation characteristic corresponding to the user;
the input subunit is configured to input the commodity recommendation feature to a multitask recommendation layer in the multitask commodity recommendation model, so that the multitask recommendation layer processes the commodity recommendation feature to obtain a global commodity sequence of each commodity recommendation block, where each global commodity sequence includes at least one commodity, and each commodity includes an estimated click rate and an estimated conversion rate of the commodity;
the second obtaining subunit is used for obtaining a fusion score of each commodity in each global commodity sequence based on the estimated click rate and the estimated conversion rate of each commodity in each global commodity sequence;
and the sorting subunit is used for sorting the commodities in the global commodity sequence of the commodity recommendation block according to the sequence of the fusion scores from high to low for each commodity recommendation block, and selecting the first N commodities to form the commodity recommendation sequence of the commodity recommendation block, wherein N is greater than or equal to 1.
In the apparatus provided in the embodiment of the present invention, the apparatus may be further configured to:
and the acquisition unit is used for acquiring data generated when the user carries out interaction behavior on the display page after the E-commerce software renders each commodity included in each commodity recommendation sequence to the display page corresponding to the commodity recommendation block corresponding to the commodity, and storing the data in a preset historical behavior database.
The embodiment of the invention also provides a storage medium, which comprises a stored instruction, wherein when the instruction runs, the device where the storage medium is located is controlled to execute the commodity recommendation method.
An electronic device is provided in an embodiment of the present invention, and the structural diagram of the electronic device is shown in fig. 5, which specifically includes a memory 501 and one or more instructions 502, where the one or more instructions 502 are stored in the memory 501, and are configured to be executed by one or more processors 503 to perform the following operations according to the one or more instructions 502:
when a commodity display instruction sent by a user through E-commerce software is received, acquiring user characteristic data, the above interactive characteristic data and commodity characteristic data of the E-commerce software, wherein the above interactive characteristic data is data of interaction between the user and the E-commerce software, which is acquired by the E-commerce software based on the commodity display instruction;
sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multitask commodity recommendation model;
triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software;
and sending each commodity recommendation sequence to the E-commerce software, so that the E-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation version block corresponding to the commodity, and recommends the commodity to the user.
The specific implementation procedures and derivatives thereof of the above embodiments are within the scope of the present invention.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the system or system embodiments are substantially similar to the method embodiments and therefore are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for related points. The above-described system and system embodiments are only illustrative, wherein the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A method for recommending an article, comprising:
when a commodity display instruction sent by a user through E-commerce software is received, obtaining user characteristic data, above interactive characteristic data and commodity characteristic data of the E-commerce software, wherein the above interactive characteristic data are data acquired by the E-commerce software based on the commodity display instruction and used for interaction between the user and the E-commerce software;
sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multitask commodity recommendation model;
triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software;
and sending each commodity recommendation sequence to the e-commerce software, so that the e-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation block corresponding to the commodity, and recommends the commodity to the user.
2. The method of claim 1, wherein the obtaining of the user characteristic data, the above interaction characteristic data, and the commodity characteristic data of the e-commerce software comprises:
acquiring user identification information of the user, acquiring attribute information and historical interactive behavior data of the user based on the user identification information, filtering the historical interactive behavior data to obtain qualified historical interactive behavior data, and forming the attribute information and the qualified historical interactive behavior data into the user characteristic data;
analyzing the commodity display instruction, determining the instruction type of the commodity display instruction, acquiring interactive data corresponding to the instruction type, and taking the interactive data as the above interactive feature data;
determining each commodity recommendation block contained in the E-commerce software, acquiring commodity to-be-displayed information of each commodity recommendation block, and forming the commodity characteristic data based on the commodity to-be-displayed information.
3. The method according to claim 2, wherein the analyzing the commodity display instruction, determining an instruction type of the commodity display instruction, obtaining interaction data corresponding to the instruction type, and using the interaction data as the above interaction feature data comprises:
analyzing the commodity display instruction to obtain a type identifier in the commodity display instruction, and judging whether the instruction type of the commodity display instruction is a skip type or not based on the type identifier;
if the instruction type of the commodity display instruction is a skip type, determining interactive data generated when the user executes an interactive behavior on a page currently displayed by the E-commerce software, and taking the interactive data as the above interactive feature data;
and if the instruction type of the commodity display instruction is not the jump type, determining that the instruction type of the commodity display instruction is the starting type, generating data representing that the interaction between the user and the E-commerce software is zero, and taking the data as the above interactive feature data.
4. The method according to claim 1, wherein the triggering the multitask commodity recommendation model to process the user characteristic data, the above interaction characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block included in the e-commerce software comprises:
triggering the multitask commodity recommendation model to call an input feature layer to process the user feature data, the above interactive feature data and the commodity feature data to obtain user feature discrete data corresponding to the user feature data, the above interactive feature discrete data corresponding to the above interactive feature data and the commodity feature discrete data corresponding to the commodity feature data;
the multitask commodity recommendation model calls a feature embedding layer to carry out low-dimensional embedding mapping processing on the user feature discrete data, the above interactive feature discrete data and the commodity feature discrete data respectively to obtain a user feature matrix corresponding to the user feature data, an above interactive feature matrix corresponding to the above interactive feature data and a commodity feature matrix corresponding to the commodity feature data;
the multitask commodity recommendation model calls a feature extraction layer to perform feature splicing on the user feature matrix, the above interactive feature matrix and the commodity feature matrix to obtain commodity recommendation features corresponding to the user;
inputting the commodity recommendation characteristics to a multitask recommendation layer in the multitask commodity recommendation model, and enabling the multitask recommendation layer to process the commodity recommendation characteristics to obtain a global commodity sequence of each commodity recommendation block, wherein each global commodity sequence comprises at least one commodity, and each commodity comprises an estimated click rate and an estimated conversion rate of the commodity;
obtaining a fusion score of each commodity in each global commodity sequence based on the estimated click rate and the estimated conversion rate of each commodity in each global commodity sequence;
and for each commodity recommendation block, sorting commodities in the global commodity sequence of the commodity recommendation block according to the sequence of fusion scores from high to low, and selecting the first N commodities to form the commodity recommendation sequence of the commodity recommendation block, wherein N is greater than or equal to 1.
5. The method of claim 1, further comprising:
after the E-commerce software renders each commodity included in each commodity recommendation sequence to a display page corresponding to a commodity recommendation section corresponding to the commodity, data generated when the user carries out interaction on the display page is collected, and the data are stored in a preset historical behavior database.
6. An article recommendation device, comprising:
the electronic commerce system comprises an acquisition unit, a display unit and a display unit, wherein the acquisition unit is used for acquiring user characteristic data, above interactive characteristic data and commodity characteristic data of electronic commerce software when a commodity display instruction sent by a user through the electronic commerce software is received, and the above interactive characteristic data is data acquired by the electronic commerce software based on the commodity display instruction and used for interaction between the user and the electronic commerce software;
the sending unit is used for sending the user characteristic data, the above interactive characteristic data and the commodity characteristic data to a preset multi-task commodity recommendation model;
the triggering unit is used for triggering the multitask commodity recommendation model to process the user characteristic data, the above interactive characteristic data and the commodity characteristic data so as to update a commodity recommendation sequence of each commodity recommendation block contained in the E-commerce software;
and the rendering unit is used for sending each commodity recommendation sequence to the e-commerce software, so that the e-commerce software renders each commodity in each commodity recommendation sequence to a display page corresponding to the commodity recommendation version block corresponding to the commodity, and the commodity is recommended to the user.
7. The apparatus of claim 6, wherein the obtaining unit comprises:
the first obtaining subunit is configured to obtain user identification information of the user, obtain attribute information and historical interactive behavior data of the user based on the user identification information, filter the historical interactive behavior data to obtain qualified historical interactive behavior data, and combine the attribute information and the qualified historical interactive behavior data into the user feature data;
the analysis subunit is used for analyzing the commodity display instruction, determining the instruction type of the commodity display instruction, acquiring interactive data corresponding to the instruction type, and taking the interactive data as the interactive characteristic data;
and the determining subunit is used for determining each commodity recommendation block contained in the E-commerce software, acquiring the commodity to-be-displayed information of each commodity recommendation block, and forming the commodity characteristic data based on each commodity to-be-displayed information.
8. The apparatus of claim 7, wherein the parsing subunit comprises:
the acquisition module is used for analyzing the commodity display instruction, acquiring a type identifier in the commodity display instruction, and judging whether the instruction type of the commodity display instruction is a skip type or not based on the type identifier;
the determining module is used for determining interactive data generated when the user executes an interactive behavior on a page currently displayed by the E-commerce software if the instruction type of the commodity display instruction is a skip type, and taking the interactive data as the above interactive feature data;
and the generation module is used for determining that the instruction type of the commodity display instruction is a starting type if the instruction type of the commodity display instruction is not a jump type, generating data representing that the interaction between the user and the E-commerce software is zero, and taking the data as the above interactive characteristic data.
9. A storage medium, characterized in that the storage medium comprises stored instructions, wherein when the instructions are executed, the storage medium is controlled to execute the commodity recommendation method according to any one of claims 1 to 5.
10. An electronic device comprising a memory and one or more instructions, wherein the one or more instructions are stored in the memory and configured to be executed by the one or more processors to perform the merchandise recommendation method of any one of claims 1-5.
CN202110666837.XA 2021-06-16 2021-06-16 Commodity recommendation method and device, storage medium and electronic equipment Pending CN115482054A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110666837.XA CN115482054A (en) 2021-06-16 2021-06-16 Commodity recommendation method and device, storage medium and electronic equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110666837.XA CN115482054A (en) 2021-06-16 2021-06-16 Commodity recommendation method and device, storage medium and electronic equipment

Publications (1)

Publication Number Publication Date
CN115482054A true CN115482054A (en) 2022-12-16

Family

ID=84420330

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110666837.XA Pending CN115482054A (en) 2021-06-16 2021-06-16 Commodity recommendation method and device, storage medium and electronic equipment

Country Status (1)

Country Link
CN (1) CN115482054A (en)

Similar Documents

Publication Publication Date Title
CN109697629B (en) Product data pushing method and device, storage medium and computer equipment
CN110008397B (en) Recommendation model training method and device
CN110427560A (en) A kind of model training method and relevant apparatus applied to recommender system
CN111400599A (en) User group portrait generation method, device and system
CN110580489B (en) Data object classification system, method and equipment
CN111767466A (en) Recommendation information recommendation method and device based on artificial intelligence and electronic equipment
CN111967914A (en) User portrait based recommendation method and device, computer equipment and storage medium
CN112711713B (en) Interest point recommendation and display method and device, computer equipment and storage medium
CN113592535A (en) Advertisement recommendation method and device, electronic equipment and storage medium
CN112200538A (en) Data processing method, device, equipment and storage medium
CN111191133A (en) Service search processing method, device and equipment
CN113157752A (en) Scientific and technological resource recommendation method and system based on user portrait and situation
CN111861605A (en) Business object recommendation method
CN113327132A (en) Multimedia recommendation method, device, equipment and storage medium
CN111325614B (en) Recommendation method and device of electronic object and electronic equipment
CN112015970A (en) Product recommendation method, related equipment and computer storage medium
CN115482054A (en) Commodity recommendation method and device, storage medium and electronic equipment
CN115375484A (en) Matrix decomposition-based insurance product extraction method and device, equipment and medium
CN115965089A (en) Machine learning method for interface feature display across time zones or geographic regions
CN115048579A (en) Method, device and equipment for searching materials
CN116764236A (en) Game prop recommending method, game prop recommending device, computer equipment and storage medium
CN113515701A (en) Information recommendation method and device
CN115455276A (en) Method and device for recommending object, computer equipment and storage medium
CN111414538A (en) Text recommendation method and device based on artificial intelligence and electronic equipment
CN106484710B (en) Dynamic data processing method and device and information display method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination