CN117350816A - Independent station commodity recommendation method and device, equipment and medium thereof - Google Patents
Independent station commodity recommendation method and device, equipment and medium thereof Download PDFInfo
- Publication number
- CN117350816A CN117350816A CN202311388440.4A CN202311388440A CN117350816A CN 117350816 A CN117350816 A CN 117350816A CN 202311388440 A CN202311388440 A CN 202311388440A CN 117350816 A CN117350816 A CN 117350816A
- Authority
- CN
- China
- Prior art keywords
- commodity
- recommended
- scene
- features
- click
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000000605 extraction Methods 0.000 claims abstract description 112
- 238000006243 chemical reaction Methods 0.000 claims abstract description 95
- 238000012216 screening Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims description 71
- 238000010276 construction Methods 0.000 claims description 23
- 238000012502 risk assessment Methods 0.000 claims description 20
- 230000009471 action Effects 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 description 34
- 239000013598 vector Substances 0.000 description 24
- 230000006399 behavior Effects 0.000 description 22
- 230000006870 function Effects 0.000 description 18
- 230000004913 activation Effects 0.000 description 8
- 238000004891 communication Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 3
- 238000002955 isolation Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000008447 perception Effects 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000001413 cellular effect Effects 0.000 description 2
- 239000002360 explosive Substances 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000011324 bead Substances 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 239000002537 cosmetic Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- KJONHKAYOJNZEC-UHFFFAOYSA-N nitrazepam Chemical compound C12=CC([N+](=O)[O-])=CC=C2NC(=O)CN=C1C1=CC=CC=C1 KJONHKAYOJNZEC-UHFFFAOYSA-N 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/09—Supervised learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Business, Economics & Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- General Business, Economics & Management (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to a method for recommending commodities in an independent station, a device, equipment and a medium thereof in the technical field of electronic commerce, wherein the method comprises the following steps: acquiring scene specific characteristics, commodity basic characteristics and commodity cross characteristics, user characteristics and shop cross characteristics of a commodity to be recommended, adopting a scene extraction layer in a preset commodity recommendation model to determine target scene extraction characteristics according to the scene specific characteristics, the user characteristics and the commodity basic characteristics, adopting a shop extraction layer in the commodity recommendation model to determine target shop extraction characteristics according to the shop cross characteristics and the commodity cross characteristics, adopting a task output layer in the commodity recommendation model to determine click rate and click conversion rate of the commodity to be recommended in the scene to be recommended according to the target scene extraction characteristics and the target shop extraction characteristics, and screening out a part of commodity to be recommended to construct a recommended commodity list according to the click rate and the click conversion rate of the commodity to be recommended, and applying the commodity list to the scene to be recommended. It can be seen that the merchandise that attracts the user can be recommended.
Description
Technical Field
The present disclosure relates to the field of electronic commerce technologies, and in particular, to a method for recommending commodities in an independent station, and a corresponding apparatus, computer device, and computer readable storage medium thereof.
Background
With the explosive development of electronic commerce, the number of commodity categories at independent stations and the number of commodities in each category are rapidly increasing. However, this explosive information growth causes consumers to spend a lot of time looking for their desired goods, having to face a lot of irrelevant product information, thus putting the dilemma of information overload. This problem not only runs off the consumer, but also limits the potential for conversion enhancement at the stand alone station. Therefore, the independent station actively adopts a commodity recommendation strategy to assist a user to quickly make a purchase decision, improve the purchase conversion rate, reduce the information overload problem, provide more intelligent and efficient shopping experience, and increase commodity sales of the independent station.
In the conventional technology, a deep learning algorithm is generally adopted to model the relevance between a recommended scene and commodities, and commodities suitable for recommendation in the recommended scene are determined according to the relevance, on one hand, because the relevance only relates to the relationship between the commodities and the recommended scene, the accuracy of the relevance is difficult to ensure, and the corresponding recommended commodities are difficult to ensure to be capable of sufficiently attracting users, on the other hand, each recommended scene is generally required to be respectively modeled, so that various recommended scenes are isolated from each other, and the mutual available partial relevance among the various recommended scenes cannot be shared.
In view of the shortcomings of the traditional technology, the applicant has long conducted research in the related field, and has developed a new way for solving the problems in the field of database information processing.
Disclosure of Invention
It is a primary object of the present application to solve at least one of the above problems and provide a method for recommending independent station commodities, and a corresponding apparatus, computer device, and computer-readable storage medium thereof.
In order to meet the purposes of the application, the application adopts the following technical scheme:
an independent station commodity recommendation method provided in accordance with one of the objects of the present application includes the steps of:
acquiring scene specific characteristics of a scene to be recommended, commodity basic characteristics and commodity cross characteristics of commodities to be recommended, user characteristics and shop cross characteristics;
a scene extraction layer in a preset commodity recommendation model is adopted to determine target scene extraction features according to the scene unique features, the user features and commodity basic features;
determining target shop extraction features according to the shop cross features and the commodity cross features by adopting a shop extraction layer in the commodity recommendation model;
determining the click rate and click conversion rate of the commodity to be recommended in the scene to be recommended according to the target scene extraction characteristics and the target store extraction characteristics by adopting a task output layer in the commodity recommendation model;
And screening out partial commodity to be recommended according to the click rate and click conversion rate of the commodity to be recommended to construct a recommended commodity list, and applying the list to the scene to be recommended.
On the other hand, the independent station commodity recommending device provided by adapting to one of the purposes of the application comprises a characteristic acquiring module, a scene extracting module, a shop extracting module, a task extracting module and a list constructing module, wherein the characteristic acquiring module is used for acquiring scene specific characteristics of a scene to be recommended, commodity basic characteristics and commodity cross characteristics of commodities to be recommended, user characteristics and shop cross characteristics; the scene extraction module is used for determining target scene extraction features according to the scene unique features, the user features and the commodity basic features by adopting a scene extraction layer in a preset commodity recommendation model; the store extraction module is used for determining target store extraction characteristics according to the store cross characteristics and the commodity cross characteristics by adopting a store extraction layer in the commodity recommendation model; the task extraction module is used for determining the click rate and click conversion rate of the commodity to be recommended in the scene to be recommended according to the target scene extraction characteristics and the target shop extraction characteristics by adopting a task output layer in the commodity recommendation model; the list construction module is used for screening out a part of commodity construction recommended commodity list to be recommended according to the clicking rate and clicking conversion rate of the commodity to be recommended, and applying the commodity list to the scene to be recommended.
In yet another aspect, a computer device is provided, adapted for one of the purposes of the present application, comprising a central processor and a memory, the central processor being adapted to invoke the steps of running a computer program stored in the memory to perform the stand alone commodity recommendation method described herein.
In yet another aspect, a computer readable storage medium adapted to another object of the present application is provided, in which a computer program implemented according to the stand alone commodity recommendation method is stored in the form of computer readable instructions, which when invoked by a computer, perform the steps comprised by the method.
The technical solution of the present application has various advantages, including but not limited to the following aspects:
according to the method, a preset commodity recommendation model is adopted, scene specific features of a scene to be recommended, commodity basic features and commodity cross features of commodities to be recommended, user features and shop cross features are used as inputs, a scene extraction layer in the model determines target scene extraction features according to the scene specific features, the user features and the commodity basic features, a shop extraction layer in the model determines target shop extraction features according to the shop cross features and the commodity cross features, a task output layer in the model determines click rate and click conversion rate of the commodities to be recommended under the scene to be recommended according to the target scene extraction features and the target shop extraction features, and a part of commodity to be recommended construction recommendation commodity list is screened out according to the click rate and the click conversion rate of the commodities to be recommended and is applied to the scene to be recommended. On one hand, the correlation between the commodity to be recommended and the store can be modeled by the store extraction layer of the model, and the correlation between the commodity to be recommended, the scene to be recommended and the user can be modeled by the scene extraction layer of the model, so that the click rate and the click conversion rate of the commodity to be recommended in the scene to be recommended can be accurately determined by the task extraction layer based on the correlation of multiple dimensions, and accordingly, the commodity to be recommended which attracts the user is ensured to be preferred from the commodity to be recommended, and the user is hopeful to be promoted to purchase the commodity to be recommended. On the other hand, the data isolation among different independent stations can be ensured, and personalized commodity recommendation which is only applicable to each independent station can be realized.
Drawings
The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow chart of an exemplary embodiment of a stand alone commodity recommendation method according to the present application;
FIG. 2 is a schematic diagram of a model structure of a commodity recommendation model according to an embodiment of the present application;
FIG. 3 is a flow diagram of constructing a store crossover feature in an embodiment of the present application;
FIG. 4 is a schematic flow chart of constructing a commodity intersection feature in an embodiment of the present application;
FIG. 5 is a schematic flow chart of training a commodity recommendation model according to an embodiment of the present application;
FIG. 6 is a flow diagram of scene specific features for constructing multiple recommended scenes in an embodiment of the present application;
FIG. 7 is a schematic flow chart of a method for constructing a recommendation list for goods according to an embodiment of the present application;
FIG. 8 is a flowchart illustrating determining weights corresponding to click rate and click conversion rate according to an embodiment of the present application;
FIG. 9 is a schematic block diagram of a stand alone commodity recommendation device of the present application;
fig. 10 is a schematic structural diagram of a computer device used in the present application.
Detailed Description
Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only for the purpose of illustrating the present application and are not to be construed as limiting the present application.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. The term "and/or" as used herein includes all or any element and all combination of one or more of the associated listed items.
It will be understood by those skilled in the art that all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs unless defined otherwise. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
As used herein, "client," "terminal device," and "terminal device" are understood by those skilled in the art to include both devices that include only wireless signal receivers without transmitting capabilities and devices that include receiving and transmitting hardware capable of two-way communication over a two-way communication link. Such a device may include: a cellular or other communication device such as a personal computer, tablet, or the like, having a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; a PCS (Personal Communications Service, personal communication system) that may combine voice, data processing, facsimile and/or data communication capabilities; a PDA (Personal Digital Assistant ) that can include a radio frequency receiver, pager, internet/intranet access, web browser, notepad, calendar and/or GPS (Global Positioning System ) receiver; a conventional laptop and/or palmtop computer or other appliance that has and/or includes a radio frequency receiver. As used herein, "client," "terminal device" may be portable, transportable, installed in a vehicle (aeronautical, maritime, and/or land-based), or adapted and/or configured to operate locally and/or in a distributed fashion, at any other location(s) on earth and/or in space. As used herein, a "client," "terminal device," or "terminal device" may also be a communication terminal, an internet terminal, or a music/video playing terminal, for example, a PDA, a MID (Mobile Internet Device ), and/or a mobile phone with music/video playing function, or may also be a device such as a smart tv, a set top box, or the like.
The hardware referred to by the names "server", "client", "service node" and the like in the present application is essentially an electronic device having the performance of a personal computer, and is a hardware device having necessary components disclosed by von neumann's principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device, and a computer program is stored in the memory, and the central processing unit calls the program stored in the external memory to run in the memory, executes instructions in the program, and interacts with the input/output device, thereby completing a specific function.
It should be noted that the concept of "server" as referred to in this application is equally applicable to the case of a server farm. The servers should be logically partitioned, physically separate from each other but interface-callable, or integrated into a physical computer or group of computers, according to network deployment principles understood by those skilled in the art. Those skilled in the art will appreciate this variation and should not be construed as limiting the implementation of the network deployment approach of the present application.
One or several technical features of the present application, unless specified in the plain text, may be deployed either on a server to implement access by remotely invoking an online service interface provided by the acquisition server by a client, or directly deployed and run on the client to implement access.
The neural network model cited or possibly cited in the application can be deployed on a remote server and used for implementing remote call on a client, or can be deployed on a client with sufficient equipment capability for direct call unless specified in a clear text, and in some embodiments, when the neural network model runs on the client, the corresponding intelligence can be obtained through migration learning so as to reduce the requirement on the running resources of the hardware of the client and avoid excessively occupying the running resources of the hardware of the client.
The various data referred to in the present application, unless specified in the plain text, may be stored either remotely in a server or in a local terminal device, as long as it is suitable for being invoked by the technical solution of the present application.
Those skilled in the art will appreciate that: although the various methods of the present application are described based on the same concepts so as to be common to each other, the methods may be performed independently of each other unless specifically indicated otherwise. Similarly, for each of the embodiments disclosed herein, the concepts presented are based on the same inventive concept, and thus, the concepts presented for the same description, and concepts that are merely convenient and appropriately altered although they are different, should be equally understood.
The various embodiments to be disclosed herein, unless the plain text indicates a mutually exclusive relationship with each other, the technical features related to the various embodiments may be cross-combined to flexibly construct a new embodiment, so long as such combination does not depart from the inventive spirit of the present application and can satisfy the needs in the art or solve the deficiencies in the prior art. This variant will be known to the person skilled in the art.
The method for recommending the commodity of the independent station can be programmed into a computer program product and can be deployed in a client or a server for operation, for example, in the exemplary application scene of the application, the method can be deployed in the server of an electronic commerce platform, and therefore the method can be executed by accessing an interface opened after the computer program product is operated and performing man-machine interaction with the process of the computer program product through a graphical user interface.
Referring to fig. 1, in an exemplary embodiment, the method for recommending a commodity in an independent station of the present application includes the following steps:
step S1100, acquiring scene specific features of a scene to be recommended, commodity basic features and commodity cross features of commodities to be recommended, user features and shop cross features;
The independent station is a novel official network (website) which is built based on the SaaS technology platform, has independent domain names, proprietary content, data and rights and interests, has independent management main authority and management main responsibility, is supported by social cloud computing capability, and can independently and freely butt-joint third party software tools, propaganda popularization media and channels. An online store is typically set up and operated on the stand-alone station by a commodity seller so that the online store can display its commodity on the stand-alone station, providing the commodity buyer with e-commerce services such as browsing the commodity on the online store, adding shopping carts, settling accounts, logistic delivery, placing orders, after-sales services, and the like. Because online shops on different independent stations are independent entities, and all data of the online shops are isolated from each other, the online shops on the independent stations are recommended on the premise of guaranteeing data isolation among the independent stations, and commodity recommendation is carried out according to self-recommended scenes, commodities, shops and corresponding characteristics of users of the online shops on the independent stations. The user of the present application refers to a commodity purchaser.
The commodity description text and the commodity picture of each commodity in the online shops of the independent stations are generally set by corresponding commodity sellers, and after the setting is completed, the unique identification of the commodity corresponding to the commodity description text and the commodity picture of each commodity is stored in a commodity database for calling and deleting. Therefore, according to the unique identification of the commodity, the commodity description text and the commodity picture related to the commodity can be obtained from the commodity database. The unique identification of the article is used to uniquely refer to a single article to distinguish between various articles, such as article IDs.
The scene to be recommended can be any one of various recommended scenes in the electronic commerce field, wherein the various recommended scenes comprise a home page recommended scene, an item detail page recommended scene, a shopping cart recommended scene, a settlement page recommended scene and a search result page-free recommended scene.
The first page recommended scene is a scene to be recommended, and a user usually browses recommended commodities in the first page without explicit purchase intention. Under the home page recommendation scene, the commodity to be recommended is more relevant to any one or more of a hot search commodity keyword, a hot sales commodity keyword and a hot shopping commodity keyword, and is more hopeful to attract a user to click the commodity to be recommended, and execute any one or more conversion behaviors such as direct ordering, purchasing, shopping cart adding, collecting, forwarding and the like. Accordingly, any multiple of the hot search commodity keywords, the hot sale commodity keywords and the hot shopping commodity keywords are obtained and spliced with the unique identification of the scene to be recommended, and the scene specific characteristics of the scene to be recommended are obtained. The hot search commodity keywords, the hot sale commodity keywords and the hot shopping commodity keywords are keyword texts which are correspondingly extracted from commodity description texts of the hot search commodity, the hot sale commodity and the hot shopping commodity respectively. The hot-searched commodity is a commodity with a relatively large number of times compared with other commodities as a search result in an online store, the hot-sold commodity is a commodity with a relatively large sales volume compared with other commodities in the online store, and the hot-purchased commodity is a commodity with a relatively large number of times compared with other commodities in the online store. The commodity description text comprises all texts used for describing corresponding commodities, such as commodity titles, commodity categories, commodity labels, commodity details, commodity sales areas, logistics transportation modes, logistics transportation contractors and the like, and the commodity details are usually used for describing information of detail aspects, such as sizes, colors, materials, weights, capacities and the like, of the commodities. The keyword text extracted from the commodity description text of the corresponding commodity can be realized by a keyword extraction algorithm or can be manually extracted by an operation technician, the keyword extraction algorithm can be TF-IDF, textRank, RAKE, LDA, a deep learning model and the like, and the keyword extraction algorithm can be realized by one of the technicians according to the requirements. The unique identification of the scene to be recommended is used to distinguish from other recommended scenes, and can be flexibly set by one skilled in the art based on the disclosure herein.
The recommendation scene of the to-be-recommended scene as the commodity detail page is usually that a user knows about the commodity in the commodity detail page under the condition of a certain purchase intention or purchase interest and browses other recommended commodities. Under the commodity detail page recommendation scene, the higher the similarity degree, the correlation degree and the matching degree of the commodity to be recommended and the displayed commodity are, the more hopefully the user is attracted to click the commodity to be recommended, and any one or any plurality of conversion behaviors such as direct ordering purchase, shopping cart addition, collection, forwarding and the like are executed. Accordingly, the commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the commodity to be recommended and any plurality of commodity basic characteristics of the commodity matched with the displayed commodity are acquired from the current commodity detail page, and are spliced with the unique identification of the scene to be recommended, so that the scene characteristic characteristics of the scene to be recommended are acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, the word segmentation is obtained by splicing unique identifiers of the commodities, the word segmentation can be realized by adopting a word segmentation algorithm, and the word segmentation algorithm can be realized by using a Jieba word segmentation, an n-gram word segmentation, a WordPiece word segmentation, a CRF word segmentation, named entity recognition, an e-commerce dictionary matching word segmentation and the like, and the skilled in the art can flexibly change one realization. The similarity between the displayed commodity and the commodity to be recommended can be determined according to commodity pictures and commodity description texts corresponding to the displayed commodity and the commodity to be recommended, and the displayed commodity and the commodity to be recommended can be flexibly realized by a person skilled in the art or can be realized by reference to the disclosure of the subsequent part of embodiments. The commodity picture comprises any one of pictures used for displaying the appearance of the corresponding commodity, such as a commodity head picture, a commodity detail picture and the like, and the commodity head picture is recommended to be used as the commodity picture of the application.
For a shopping cart recommendation scene, a user refers to or places an order to purchase a displayed commodity in the shopping cart under the explicit purchase intention, and browses other recommended commodities. Under the shopping cart recommendation scene, the higher any one or more of the similarity degree, the correlation degree and the correlation degree between the commodity to be recommended and the displayed commodity and the current commodity purchased by the user in a history mode, the higher any one or more conversion behaviors such as attracting the user to click the commodity to be recommended and executing direct ordering purchase, adding shopping carts, collecting, forwarding and the like are expected. Accordingly, the commodity basic characteristics of the displayed commodity in the current shopping cart, the similarity between the displayed commodity and the commodity to be recommended, and any multiple items of the commodity basic characteristics of the current user historical purchased commodity are acquired and spliced with the unique identification of the scene to be recommended, so that the scene specific characteristics of the scene to be recommended are acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, and the unique identification of the commodity is spliced. The similarity between the displayed commodity and the commodity to be recommended can be determined according to commodity pictures and commodity description texts corresponding to the displayed commodity and the commodity to be recommended, and the displayed commodity and the commodity to be recommended can be flexibly realized by a person skilled in the art or can be realized by reference to the disclosure of the subsequent part of embodiments. The commodity picture comprises any one of pictures used for displaying the appearance of the corresponding commodity, such as a commodity head picture, a commodity detail picture and the like, and the commodity head picture is recommended to be used as the commodity picture of the application.
The recommended scene for the settlement page is usually that the user places an order for the displayed commodity in the settlement page under the explicit purchase intention and browses other commodities. Under the settlement page recommendation scene, the higher the similarity degree, the correlation degree and the preferential collocation degree of the commodity to be recommended and the displayed commodity are, the more hopefully the user is attracted to click the commodity to be recommended, and any one or any plurality of conversion behaviors such as direct ordering purchase, shopping cart addition, collection, forwarding and the like are executed. Accordingly, any multiple items of commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the commodity to be recommended and the commodity basic characteristics of the preferential commodity matched with the displayed commodity to reach preferential conditions in the current settlement page are acquired and spliced with the unique identification of the scene to be recommended, so that the scene specific characteristics of the scene to be recommended are acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, and the unique identification of the commodity is spliced. The similarity between the displayed commodity and the commodity to be recommended can be determined according to commodity pictures and commodity description texts corresponding to the displayed commodity and the commodity to be recommended, and the displayed commodity and the commodity to be recommended can be flexibly realized by a person skilled in the art or can be realized by reference to the disclosure of the subsequent part of embodiments. The commodity picture comprises any one of pictures used for displaying the appearance of the corresponding commodity, such as a commodity head picture, a commodity detail picture and the like, and the commodity head picture is recommended to be used as the commodity picture of the application.
The page recommendation scene without the search result for the scene to be recommended is usually that a user browses other recommended commodities under the condition that the user has an explicit purchase intention but does not search for the commodities meeting the purchase intention. And under the non-search result page recommendation scene, the higher any one or more of the correlation degree between the commodity to be recommended and the commodity clicked by the current user in a history way, the correlation degree between the commodity purchased by the current user in a history way and the correlation degree between the commodity displayed in the current shopping cart is, the more the user is expected to be attracted to click the recommended commodity, and any one or more conversion behaviors such as direct ordering purchase, shopping cart adding, collection, forwarding and the like are executed. Accordingly, the commodity basic characteristics of the commodity clicked by the user in the history, the commodity basic characteristics of the commodity purchased by the user in the history, the commodity basic characteristics of the commodity displayed in the shopping cart in the history and the unique identification of the scene to be recommended are acquired and spliced, so that the scene specific characteristics of the scene to be recommended are acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, and the unique identification of the commodity is spliced.
Any commodity in the online store can be used as the commodity to be recommended. Acquiring commodity description text of the commodity to be recommended from the commodity database according to the unique identifier of the commodity to be recommended, after each text in the commodity description text is spliced, word segmentation is carried out on the corresponding spliced text by adopting a word segmentation algorithm, and the unique identifier on the corresponding word segmentation sequence splice is used as commodity basic characteristics of the commodity to be recommended.
The personal information set when the user registers in the independent station and the receiving address set when shopping are obtained, wherein the personal information comprises any plurality of items such as gender, education level, age/birth date, occupation and the like, the corresponding age is calculated according to the birth date in the personal information, administrative division part texts are determined according to the receiving address to serve as the location, and the personal information and the location are spliced to obtain the user portrait characteristics of the user. Exemplary examples: 28. master, female, finance, guangzhou city sea bead district, guangdong province, china.
Historical behavioral data corresponding to each user's online store accessing the independent station may be recorded and stored in a user database in association with each user's unique identification of that user. According to the current unique identifier such as ID of the user, accessing a user database to acquire historical behavior data of the user, on one hand, according to actions such as praying, forwarding, collecting, adding shopping carts, purchasing, clicking browsing and the like of commodities in an online store in the historical behavior data, determining brands and commodity classes of commodities related to any multiple items as brand preferences respectively, commodity class preferences such as Chanel, gucci, lipstick, package and cosmetics, determining a purchasing period according to commodity time of purchasing in the historical behavior data, such as three times per month, determining a payment sum and the number of purchased commodities according to order data of purchasing in the historical behavior data, determining a consumption level according to the payment sum and the number of purchased commodities, in one embodiment, presetting a stepped increase threshold corresponding to the payment sum and the number of purchased commodities, and mapping the consumption level corresponding to each class, and determining a corresponding consumption level when the payment sum and the number of purchased commodities reach the corresponding threshold, such as the consumption level: consumption level is low [ first hierarchy: ten thousand to hundred thousand, 100-500 goods ], in consumption level [ second tier: hundred thousand to one hundred thousand, 500-5000, high consumption level [ third level: more than half a million yuan, more than 5000, one skilled in the art can flexibly set the stepwise increase threshold corresponding to the total amount paid and the number of items purchased in accordance with the disclosure herein. And splicing the brand preference, the commodity class preference, the consumption level and the purchase period to obtain the first user behavior characteristic of the current user. On the other hand, according to the click behaviors and purchase behaviors of the commodities in the various recommended scenes in the historical behavior data, a keyword extraction algorithm is adopted to extract keywords from commodity description texts of the commodities which are related to the click behaviors and are clicked by the user as click keywords, a keyword extraction algorithm is adopted to extract keywords from commodity description texts of the commodities which are related to the purchase behaviors and are purchased after being clicked by the user as purchase keywords, unique identifiers of recommended scenes where the corresponding commodities are spliced by the click keywords are obtained to obtain click scene data pairs, unique identifiers of recommended scenes where the corresponding commodities are spliced by the purchase keywords are obtained to obtain purchase scene data pairs, and all the click scene data pairs and the purchase scene data pairs are spliced to obtain second user behavior characteristics of the current user. And splicing the first user behavior characteristic, the second user behavior characteristic and the user portrait characteristic of the user to obtain the user characteristic.
Any of the channels, commodity categories, commodity attributes, commodity categories, commodity prices, commodity sales areas, logistics transportation modes and logistics transportation contractors of the to-be-recommended commodity are correspondingly crossed with any of the main channels, main commodity categories, main commodity attributes, main commodity brands, main commodity price intervals, main commodity sales areas, main logistics transportation modes and main logistics transportation contractors of the on-line shops, however, the higher the crossed overlapping degree is, the more hopeful the user to click the to-be-recommended commodity, and any one or any plurality of conversion actions such as direct order purchase, shopping cart addition, collection, forwarding and the like are executed. Accordingly, commodity cross features are constructed according to any plurality of cross items of the commodity to be recommended, in addition, store cross features are constructed according to any plurality of cross items of the online store at present, and a person skilled in the art can flexibly and flexibly realize the construction of the commodity cross features and the store cross features, and can also realize the construction according to the disclosure of the embodiments in the follow-up part. The main channels are channels with more advertisement expenditure in all channels corresponding to marketing advertisements of commodity put in by merchant users. The main commodity is a commodity with more sales and/or click browsing quantity in all commodities of the online store. The main camp commodity class, the main camp commodity attribute, the main camp commodity brand, the main camp commodity price interval, the main camp commodity sales area, the main camp logistics transportation mode and the main camp logistics transportation contractor are the commodity class, the commodity attribute, the commodity class, the commodity price, the commodity sales area, the logistics transportation mode and the logistics transportation contractor corresponding to all main camp commodities, and the most frequent occurrence occurs.
Step 1200, determining target scene extraction features by using a scene extraction layer in a preset commodity recommendation model according to the scene unique features, the user features and commodity basic features;
referring to fig. 2, the model structure of the commodity recommendation model includes a shared embedding layer 200, a store extraction layer 201, a scene extraction layer 202, a task extraction layer 203, a first task tower 204, and a second task tower 205. The commodity recommendation model is trained to a convergence state in advance, and the capability of determining the click rate and click conversion rate of commodities in various recommendation scenes is obtained.
The specific scene characteristics, the user characteristics and the commodity basic characteristics are input into the sharing embedding layer 200 of the commodity recommendation model to carry out embedding vector representation, the embedding vector representation corresponding to each characteristic is obtained and input into the scene extraction layer 202, valuable shared information between the scene to be recommended and other recommended scenes is determined by a scene sharing expert network in the scene extraction layer 202, the shared information representation is output, the specific information of the scene to be recommended is determined by the scene specific expert network in the scene extraction layer 202, the specific information representation is output, the specific information of other various recommended scenes is determined by a scene perception attention network in the scene extraction layer 202, the importance weights of the specific information of the scene to be recommended to the specific information representation are output, the weighted sum obtained by multiplying the importance weights of the specific information of the other recommended scenes is output, and the output results corresponding to the scene sharing network, the scene specific expert network and the scene perception attention network are connected to form the target scene extraction characteristics.
The scene sharing network and the scene specific expert network are the same in network structure, and are each MoE (mixing-of-Experts) comprising a plurality of sub expert networks and a single gating network, wherein the gating network is realized based on the linear change of a Softmax activation function, and the sub expert networks are composed of a ReLU activation function and an MLP (Multilayer perceptron, multi-layer perceptron). The scene perception attention network is a MoE formed by combining a plurality of MoEs and a single gating network, each MoE outputs the unique information representation of the unique information of the corresponding other single recommended scene to the unique information representation of the scene to be recommended, and the gating network is used for calculating the importance weights of the unique information of other various recommended scenes to the unique information representation of the scene to be recommended after the embedded representation of the unique identifier of the scene to be recommended is subjected to the linear change of a Softmax activation function.
Step S1300, determining target shop extraction features according to the shop cross features and the commodity cross features by adopting a shop extraction layer in the commodity recommendation model;
the shop cross feature and the commodity cross feature are input to a shared embedding layer 200 of the commodity recommendation model to perform embedding vector representation, the embedding vector representation corresponding to each feature is obtained and input to a shop extraction layer 201, the shop mixing expert network of the shop extraction layer 201 determines the cross information of the online shop and the commodity to be recommended, and the cross information representation is output and is taken as a target shop extraction feature.
The network structure of the store hybrid expert network is that the MoE comprises a plurality of sub-expert networks and a single gating network, the gating network is realized based on the linear change of a Softmax activation function, and the sub-expert networks consist of a ReLU activation function and an MLP.
Step 1400, determining the click rate and click conversion rate of the commodity to be recommended in the scene to be recommended by adopting a task output layer in the commodity recommendation model according to the target scene extraction features and the target store extraction features;
the target scene extraction feature and the target store extraction feature are connected and then input into a task specific expert network of a first task in a task extraction layer 203, which is used for determining the click rate of the commodity to be recommended, a task sharing expert network of a second task and the first task, which are used for determining the click conversion rate of the commodity to be recommended, and a gating network, which is used for the first task, wherein the task specific expert network determines the specific information of the first task in the scene to be recommended, the task specific expert network outputs the specific information representation, the task sharing expert network determines the sharing information of the first task and the second task in the scene to be recommended, the sharing information representation is output, the gating network determines the weights of the specific information representation and the sharing information representation, the weighted sum obtained by multiplying the corresponding weights is output, and the weighted sum is input into a first task tower 204, so that the click rate of the commodity to be recommended is determined.
The target scene extraction feature and the target store extraction feature are connected and then input into a task extraction layer 203 to be used for determining a task specific expert network of the second task, a task sharing expert network used for serving the second task and the first task, and a gating network 205 used for serving the second task, the task specific expert network determines specific information of the second task in the scene to be recommended, the specific information representation is output, the task sharing expert network determines sharing information of the first task and the second task in the scene to be recommended, the sharing information representation is output, the gating network determines weights of the specific information representation and the sharing information representation, the weighted sum obtained by multiplying the specific information representation and the sharing information representation by the corresponding weights is output, and the weighted sum is input into a second task tower 205 to determine the click conversion rate of the commodity to be recommended.
The task output layer comprises a task extraction layer 203, a first task tower 204 and a second task tower 205, the task extraction layer is CGC (Customized Gate Control) network comprising a task specific expert network serving the first task, a task specific expert network serving the second task, a task sharing network serving the first task and the second task, a gating network serving the first task and a gating network serving the second task, each expert network is composed of a ReLU activation function and an MLP, and each gating network is realized based on linear changes of a Softmax activation function and a linear function. The first task tower 204 and the second task tower 205 are each composed of an MLP network and an activation function.
And S1500, screening out a part of commodity structure recommended commodity list to be recommended according to the click rate and click conversion rate of the commodity to be recommended, and applying the commodity list to the scene to be recommended.
It is not easy to understand that for each commodity in the online store, the clicking rate and clicking conversion rate of each commodity to be recommended can be determined, in one embodiment, the clicking rate and clicking conversion rate of each commodity to be recommended are multiplied to obtain recommendation scores, all recommendation scores are ranked according to the sequence from high to low of the recommendation scores, the commodity to be recommended corresponding to the N recommendation scores which are ranked in front is selected as the recommended commodity, the recommended commodity is ranked according to the current sequence to form a recommended commodity list, and N can be set according to requirements. And displaying the recommended commodities in the to-be-recommended commodity list in sequence in a waterfall flow layout mode in a commodity display area of a front-end display interface of the to-be-recommended scene of the independent station.
As can be appreciated from the exemplary embodiments of the present application, the technical solution of the present application has various advantages, including but not limited to the following aspects:
according to the method, a preset commodity recommendation model is adopted, scene specific features of a scene to be recommended, commodity basic features and commodity cross features of commodities to be recommended, user features and shop cross features are used as inputs, a scene extraction layer in the model determines target scene extraction features according to the scene specific features, the user features and the commodity basic features, a shop extraction layer in the model determines target shop extraction features according to the shop cross features and the commodity cross features, a task output layer in the model determines click rate and click conversion rate of the commodities to be recommended under the scene to be recommended according to the target scene extraction features and the target shop extraction features, and a part of commodity to be recommended construction recommendation commodity list is screened out according to the click rate and the click conversion rate of the commodities to be recommended and is applied to the scene to be recommended. On one hand, the correlation between the commodity to be recommended and the store can be modeled by the store extraction layer of the model, and the correlation between the commodity to be recommended, the scene to be recommended and the user can be modeled by the scene extraction layer of the model, so that the click rate and the click conversion rate of the commodity to be recommended in the scene to be recommended can be accurately determined by the task extraction layer based on the correlation of multiple dimensions, and accordingly, the commodity to be recommended which attracts the user is ensured to be preferred from the commodity to be recommended, and the user is hopeful to be promoted to purchase the commodity to be recommended. On the other hand, the data isolation among different independent stations can be ensured, and personalized commodity recommendation which is only applicable to each independent station can be realized.
Referring to fig. 3, in a further embodiment, step S1100, obtaining scene specific features of a scene to be recommended, commodity basic features and commodity cross features of a commodity to be recommended, user features, and shop cross features, includes the following steps:
step S1110, acquiring main channels of marketing advertisements of online shops to construct shop marketing advertisement features;
commodity sellers typically select multiple channels to deliver marketing advertisements for commodities in online stores, obtaining benefits of adapting to audience of different channels, increasing competitiveness of sales markets, reducing risk of losing one channel, etc., where the multiple channels may be any one or more of social media platform, search engine, advertisement banner of current independent station, etc.
And acquiring advertisement expenditure corresponding to the various channels, sorting the advertisement expenditure according to the order of the advertisement expenditure from high to low, screening N channels with the top sorting as main channels, splicing all the main channels to obtain marketing advertisement characteristics of a store, wherein N can be set according to the needs. It will be appreciated that these primary channels are relatively high in advertising expenditure relative to other channels, typically because advertising in these primary channels achieves better results, i.e., the corresponding merchandise is more attractive to the merchandise buyer.
Step S1120, acquiring any plurality of main commodity categories, main commodity attributes, main commodity brands, main commodity price intervals and main commodity sales areas of online shops, and constructing main commodity characteristics of the shops;
and acquiring commodity categories, commodity attributes, commodity brands, commodity prices and commodity sales areas in commodity description texts of all commodities from the commodity database according to unique identifiers of all commodities in the online shops.
In one embodiment, sales of all commodities in the online store are obtained, all sales are ordered according to the order of the sales from high to low, N commodities with the top order are selected as main commodities, and N can be set as required.
In one embodiment, the click browsing amount of all commodities in the online store is obtained, all the click browsing amounts are ranked according to the order of the click browsing amount from high to low, N commodities with the top ranking are screened out to be used as main commodities, and N can be set according to requirements.
It will be appreciated that the principal commodity in an online store is attractive to commodity buyers relative to other commodities.
In a recommended embodiment, acquiring sales and click browsing amounts of all commodities in an online store, sorting all sales according to the order of the sales from high to low, and screening N with the top sorting 1 Personal products are used as hot-sell products, and in addition, the products are browsed by clickingSorting all click browsing amounts in a high-to-low order, and screening N with top sorting 2 The individual commodities are used as hot commodity, and the commodity which is both the hot commodity and the hot commodity is determined to be the main commodity, wherein the N is 1 、N 2 Can be set as required.
And determining that the maximum value and the minimum value in the commodity prices of all the main commodities are correspondingly used as the maximum value and the minimum value in the main commodity price interval, and obtaining the main commodity price interval. And screening out commodity types, commodity attributes, commodity brands and commodity sales areas of all the main commodities, wherein the commodity types, commodity attributes, commodity brands and commodity sales areas with highest occurrence frequency correspond to each other and are used as main commodity types, main commodity attributes, main commodity brands and main commodity sales areas for splicing, so that the main commodity characteristics of the store are obtained.
Step S1130, any one or all of a main camp logistics transportation mode and a main camp logistics transportation contractor of an online store are obtained, and store main camp logistics characteristics are constructed;
and acquiring a logistics transportation mode and a logistics transportation contractor in the commodity description text of each main commodity from the commodity description database according to the unique identifiers of all the main commodities.
And screening out the logistics transportation modes of all the main commodities, and the logistics transportation mode with the highest occurrence frequency among the logistics transportation contractors, wherein the logistics transportation contractors are correspondingly used as the main logistics transportation mode and the main logistics transportation contractors for splicing, so that the main logistics characteristics of the store are obtained.
And step 1140, splicing the store marketing advertisement feature, the store main commodity feature and the store main commodity logistics feature into a store cross feature.
And splicing the marketing advertisement feature of the store, the commodity feature of the store owner and the commodity feature of the store owner to obtain the cross feature of the store.
In this embodiment, a process for constructing a shop cross feature is disclosed, in which a shop marketing advertisement feature, a shop owner commodity feature, and a shop owner commodity feature that constitute the shop cross feature are constructed on the premise that a commodity can attract a commodity buyer. The overlapping degree of the feature intersection of the commodity to be recommended and the shop intersection feature is ensured, and the click rate and click conversion rate of the commodity to be recommended can be accurately influenced.
Referring to fig. 4, in a further embodiment, step S1100, obtaining scene specific features of a scene to be recommended, commodity basic features and commodity cross features of a commodity to be recommended, user features, and shop cross features, includes the following steps:
S1101, acquiring a channel for delivering marketing advertisements of goods to be recommended to construct commodity marketing advertisement characteristics;
and acquiring all channels selected by the marketing advertisements of the commodity to be recommended in the online store, which are put in by the merchant user, and splicing to obtain the commodity marketing advertisement.
Step 1102, acquiring any multiple items of commodity class, commodity attribute, commodity brand, commodity price and commodity sales area of the commodity to be recommended, and constructing commodity display characteristics;
and acquiring commodity categories, commodity attributes, commodity brands, commodity prices and commodity sales areas in commodity description texts of the commodities to be recommended from the commodity database according to the unique identification of the commodities to be recommended, and splicing to obtain commodity display characteristics.
Step S1103, any one or all of a logistics transportation mode and a logistics transportation contractor of the commodity to be recommended are obtained, and commodity logistics characteristics are constructed;
and acquiring a logistics transportation mode and a logistics transportation underwriter in a commodity description text of the commodity to be recommended from the commodity database according to the unique identification of the commodity to be recommended, and splicing to obtain commodity logistics characteristics.
And step 1104, splicing the commodity marketing advertisement feature, the commodity display feature and the commodity logistics feature into commodity cross features.
And splicing the commodity marketing advertisement feature, the commodity display feature and the commodity logistics feature to obtain the commodity cross feature.
In this embodiment, a process of constructing a commodity intersection feature is disclosed, in which a corresponding intersection of the commodity intersection feature and a store intersection feature is ensured.
Referring to fig. 5, in a further embodiment, before obtaining the scene specific features of the scene to be recommended, the commodity basic features and commodity cross features of the commodity to be recommended, the user features, and the shop cross features, the method includes the following steps:
step S1000, obtaining scene unique features of various recommended scenes, and commodity basic features, commodity cross features, user features and shop cross features of corresponding recommended commodities in each recommended scene;
the plurality of recommendation scenes comprise a home page recommendation scene, an item detail page recommendation scene, a shopping cart recommendation scene, a settlement page recommendation scene and a search result page recommendation scene.
For users without explicit buying intention, the home page recommending scene is usually aimed at, in order to attract the interests of the users to the commodities in the home page, the home page is recommended to manually screen commodities which are highly related to any one or more of the hot-search commodity keywords, the hot-sell commodity keywords and the hot-add shopping commodity keywords in an online store, and accordingly, before the commodities are recommended in the home page recommending scene, the hot-search commodity keywords, the hot-add shopping commodity keywords and the unique identification of the home page recommending scene can be acquired first and spliced, and the scene characteristic features of the home page recommending scene are obtained. The hot search commodity keywords, the hot sale commodity keywords and the hot shopping commodity keywords are keyword texts which are correspondingly extracted from commodity description texts of the hot search commodity, the hot sale commodity and the hot shopping commodity respectively. The hot-searched commodity is a commodity with a relatively large number of times compared with other commodities as a search result in an online store, the hot-sold commodity is a commodity with a relatively large sales volume compared with other commodities in the online store, and the hot-purchased commodity is a commodity with a relatively large number of times compared with other commodities in the online store. The keyword text extracted from the commodity description text of the corresponding commodity can be realized by a keyword extraction algorithm or can be manually extracted by an operation technician, the keyword extraction algorithm can be TF-IDF, textRank, RAKE, LDA, a deep learning model and the like, and the keyword extraction algorithm can be realized by one of the technicians according to the requirements.
In order to attract the interest of the user to the commodities other than the displayed commodity in the commodity detail page, the commodity detail page recommends the user to manually screen out the commodity which is highly related to any one or more of the similarity degree, the correlation degree and the matching degree of the displayed commodity in the online store, so that the commodity basic feature of the current displayed commodity, the similarity between the displayed commodity and the manually screened recommended commodity and the unique identification of the commodity detail page recommends the scene, and the scene characteristic feature of the scene to be recommended can be obtained before the commodity is recommended in the commodity detail page recommends the scene. The similarity between the displayed commodity and the recommended commodity can be determined according to commodity pictures and commodity description texts corresponding to the displayed commodity and the recommended commodity, and the method can be flexibly realized by a person skilled in the art or can be realized by reference to the disclosure of the follow-up part of the embodiments. The commodity picture comprises any one of pictures used for displaying the appearance of the corresponding commodity, such as a commodity head picture, a commodity detail picture and the like, and the commodity head picture is recommended to be used as the commodity picture of the application.
For the shopping cart recommendation scene is usually oriented to users with clear buying intention, in order to attract the interest of the users to the commodities except the displayed commodities in the shopping cart, the shopping cart is recommended to the users to manually screen out any one or more commodities with higher similarity, correlation degree and correlation degree with the current user historical purchased commodities in the online shops, so that the commodity basic characteristics of the displayed commodities in the current shopping cart, the similarity between the displayed commodities and the manually screened recommended commodities and the unique identification of the shopping cart recommendation scene can be acquired before the commodities are recommended in the shopping cart recommendation scene, and the scene unique characteristics of the shopping cart recommendation scene can be obtained by splicing any multiple items in the commodity basic characteristics of the current user historical purchased commodities. The similarity between the displayed commodity and the recommended commodity can be determined according to commodity pictures and commodity description texts corresponding to the displayed commodity and the recommended commodity, and the method can be flexibly realized by a person skilled in the art or can be realized by reference to the disclosure of the follow-up part of the embodiments. The commodity picture comprises any one of pictures used for displaying the appearance of the corresponding commodity, such as a commodity head picture, a commodity detail picture and the like, and the commodity head picture is recommended to be used as the commodity picture of the application.
In order to attract the interest of the user to the commodities other than the displayed commodity in the settlement page, the settlement page recommends and manually screens out any one or more commodities with higher similarity, correlation and preferential collocation degree from the online store to the user, so that the commodity basic characteristics of the displayed commodity in the current settlement page, the similarity between the displayed commodity and the manually screened recommended commodity and the preferential commodity basic characteristics of the preferential commodity matched with the displayed commodity to reach preferential conditions can be acquired before the commodity is recommended in the settlement page recommendation scene, and the scene specific characteristics of the settlement page recommendation scene can be obtained. The similarity between the displayed commodity and the recommended commodity can be determined according to commodity pictures and commodity description texts corresponding to the displayed commodity and the recommended commodity, and the method can be flexibly realized by a person skilled in the art or can be realized by reference to the disclosure of the follow-up part of the embodiments. The commodity picture comprises any one of pictures used for displaying the appearance of the corresponding commodity, such as a commodity head picture, a commodity detail picture and the like, and the commodity head picture is recommended to be used as the commodity picture of the application.
In order to attract the interest of users to the commodities in the search-free result page, the search-free result page is recommended to the users to manually screen out any one or more commodities with higher correlation degree between the commodities and the current user historical click commodity in an online store, correlation degree between the commodities and the current user historical purchase commodity and correlation degree between the commodities displayed in a current shopping cart, so that the commodity basic characteristics of the current user historical click commodity, the commodity basic characteristics of the current user historical purchase commodity and the unique identification of the search-free result page recommendation scene can be firstly obtained under the search-free result page recommendation scene, and then the scene characteristic characteristics of the search-free result page recommendation scene can be obtained.
Before recommending goods in various recommendation scenes, firstly acquiring unique identifiers of the manually screened recommended goods in the corresponding recommendation scenes from the goods database, acquiring a goods description text of the goods to be recommended, splicing various texts in the goods description text, and then performing word segmentation on the corresponding spliced texts by using a word segmentation algorithm to acquire the unique identifiers which are spliced by the corresponding word segmentation sequences and serve as basic characteristics of the goods of the recommended goods. In addition, according to steps S1101-1104, any multiple items of a channel, a commodity class, a commodity attribute, a commodity class, a commodity price, a commodity sales area, a logistics transportation mode and a logistics transportation contractor of the recommended commodity, which are put in the marketing advertisement, are obtained, and the commodity cross characteristic of the recommended commodity is constructed. And according to the steps S1110-1140, obtaining any plurality of main channels, main commodity categories, main commodity attributes, main commodity brands, main commodity price intervals, main commodity sales areas, main commodity logistics transportation modes and main commodity logistics transportation contractors of the online shops, and constructing the shop cross characteristics.
Step S1010, constructing a plurality of training samples by using scene specific features of various recommended scenes, commodity basic features, commodity cross features, user features and shop cross features of corresponding recommended commodities in the recommended scenes, and constructing a training set by using all the training samples and the supervision labels thereof according to whether the user clicks the recommended commodities and whether to execute conversion actions after clicking the recommended commodities in the recommended scenes of the training samples;
and taking a plurality of input texts, which are formed by the scene specific features of various recommended scenes, commodity basic features, commodity cross features, user features and shop cross features of corresponding recommended commodities in the recommended scenes, as training samples.
The transformation behavior can be any one or more of direct ordering, purchasing, adding shopping carts, collecting, forwarding and the like. After recommending goods in various recommendation scenes, for each training sample, confirming that a user clicks the corresponding recommended goods in the recommendation scene of the training sample and then executes conversion behaviors after clicking the corresponding recommended goods, and marking the click rate and click conversion rate of the recommended goods corresponding to the training sample as supervision labels of [1 (click rate), 1 (click conversion rate) ]; confirming that a user does not click on a corresponding recommended commodity in a recommended scene of the training sample, and marking the click rate and click conversion rate of the recommended commodity corresponding to the training sample as a supervision tag of [0 (click rate), 0 (click conversion rate) ]; and confirming that the user does not execute conversion actions after clicking the corresponding recommended commodity and clicking the corresponding recommended commodity under the recommended scene of the training sample, and marking the click rate and click conversion rate of the recommended commodity corresponding to the training sample as a supervision tag of [1 (click rate), 0 (click conversion rate) ].
And step 1020, training a commodity recommendation model by adopting a training set to acquire the capability of determining the click rate and click conversion rate of the commodity under various recommendation scenes.
Invoking a single training sample in the training set, inputting the single training sample into a commodity recommendation model, determining target scene extraction characteristics by a scene extraction layer in the commodity recommendation model according to scene characteristic characteristics, user characteristics and commodity basic characteristics in the training sample, determining target shop extraction characteristics by a shop extraction layer in the commodity recommendation model according to shop cross characteristics and commodity cross characteristics in the training sample, and determining predicted click rate and predicted click conversion rate of recommended commodities corresponding to the training sample under the corresponding recommended scenes of the training sample by a task output layer in the commodity recommendation model according to the target scene extraction characteristics and the target shop extraction characteristics in the training sample. Calculating the predicted click rate and the predicted click conversion rate loss value according to the supervision labels of the training samples by adopting a preset cross entropy loss value function, and when the loss values corresponding to the training samples corresponding to various recommendation scenes reach a preset threshold value, indicating that the commodity recommendation model is trained to a convergence state, so that the commodity recommendation model training can be terminated; when the loss value corresponding to the training sample corresponding to any recommended scene does not reach the preset threshold value, the commodity recommendation model is indicated to be not converged, gradient update is carried out on the model according to the loss value, the weight parameters of each link of the model are corrected through back propagation to enable the model to further approach convergence, and then other training samples are continuously called to carry out iterative training on the commodity recommendation until the model is trained to a convergence state. The preset threshold may be flexibly set by a person skilled in the art.
In this embodiment, by constructing a training set including training samples corresponding to multiple recommendation scenes, and training the commodity recommendation model to a convergence state using the training set, it is ensured that the model can accurately determine the click rate and click conversion rate of the commodity in multiple recommendation scenes.
Referring to fig. 6, in a further embodiment, step S1000, obtaining scene specific features of a plurality of recommended scenes includes the following steps:
step S1001, acquiring any multiple of hot search commodity keywords, hot sale commodity keywords and hot shopping commodity keywords, and constructing scene specific features of a home page recommended scene;
for users without explicit buying intention, the home page recommending scene is usually aimed at, and in order to attract the interest of the users to the commodities in the home page, the home page recommends to the users to manually screen commodities which are highly related to any one or more of the hot search commodity keywords, the hot sales commodity keywords and the hot shopping commodity keywords in the online store.
Therefore, before recommending commodities in the home page recommending scene, the number of times, sales volume and shopping cart adding times of all commodities in the online shops are acquired, and the number of times, the sales volume and the shopping cart adding times of all the searching results are correspondingly corresponding to each other in the order from high to low according to the number of times, sales volume and the number of times of shopping carts adding After the times of adding shopping carts are ranked, N which is ranked at the front in the times of being used as all search results is screened out 1 The corresponding commodities are used as hot search commodities, and N which are ranked at the front in all sales are screened out 2 The corresponding commodities are used as hot sale commodities, and N which is ranked at the front in the times of adding shopping carts is screened out 3 The corresponding commodity is used as a hot-add commodity. The N is 1 、N 2 、N 3 Can be set as required.
And extracting the hot search commodity keywords, the hot sale commodity keywords and the hot shopping commodity keywords from commodity description texts of the hot search commodity, the hot sale commodity and the hot shopping commodity by adopting a keyword extraction algorithm. And splicing any multiple items of the hot search commodity keywords, the hot sale commodity keywords and the hot shopping commodity keywords with the unique identification of the top page recommended scene to obtain the scene unique characteristics of the top page recommended scene.
The keyword extraction algorithm can be TF-IDF, textRank, RAKE, LDA, a deep learning model and the like, and can be realized alternatively as required by a person skilled in the art.
Step S1002, acquiring any multiple items of commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and commodity basic characteristics of the matched commodity in the commodity detail page, and constructing scene specific characteristics of a commodity detail page recommended scene;
For the commodity detail page recommendation scene, a user facing a certain degree of purchase intention or purchase interest is usually recommended to manually select commodities which are highly related to any one or more of the similarity degree, the relatedness degree and the matching degree of the displayed commodities in an online shop from the commodity detail page to the user in order to attract the interest of the user to the commodities except the displayed commodities in the commodity detail page.
Therefore, before the commodity is recommended in the commodity detail page recommendation scene, the commodity basic characteristics of the commodity displayed in the current commodity detail page, the commodity basic characteristics of the matched commodity which is the commodity displayed in a matching way, the commodity description text and commodity picture of the displayed commodity and the commodity description text and commodity picture of the manually screened recommended commodity are acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, and the unique identification of the commodity is spliced.
And a preset text encoder is adopted, commodity description text of the displayed commodity is taken as input, deep semantic information of the commodity description text is extracted, and a display text feature vector for vectorizing the deep semantic information is output. And extracting deep semantic information of the commodity description text by taking the commodity description text of the recommended commodity as input by adopting the text encoder, and outputting a recommended text feature vector for vectorizing and representing the deep semantic information. And calculating the vector distance between the display text feature vector and the recommended text feature vector by adopting a preset vector distance algorithm to serve as the text similarity.
And a preset picture encoder is adopted, commodity pictures of the displayed commodity are taken as input, deep semantic information of the commodity pictures is extracted, and a display picture feature vector for vectorizing the deep semantic information is output. And extracting deep semantic information of the commodity picture by taking the commodity picture of the recommended commodity as input by adopting the text encoder, and outputting a recommended picture feature vector for vectorizing the deep semantic information. And calculating the vector distance between the feature vector of the display picture and the feature vector of the recommended picture by adopting a preset vector distance algorithm, and taking the vector distance as the picture similarity.
And multiplying the text similarity and the picture similarity to obtain the similarity between the displayed commodity and the recommended commodity.
The text encoder is pre-trained to converge and learned to determine the capabilities of the corresponding vectors based on the semantics of the extracted text. The text encoder can adopt a model in the NLP processing field, the recommended model is a BERT model, and any other model such as Transfomer Encoder, roBERTa, XLM-RoBERTa, MPNet, biBiLSTM, GPT and the like can also be adopted. The training of these models is known to those skilled in the art and will not be described in detail. The picture encoder is trained in advance to converge, and the capability of determining the corresponding vector based on the semantics of the extracted picture is learned. The picture encoder may adopt a model in the field of picture processing, the recommended model is ViT (Vision Transformer) model, or any other model such as CNN model, deep convolution model EfficientNet, denseNet, resnet, etc., and the training of these models is known to those skilled in the art, so that detailed description will not be given here. The vector distance algorithm can be any one of common algorithms such as cosine similarity algorithm, euclidean distance algorithm, pelson coefficient algorithm, jacquard similarity algorithm and the like.
And splicing any plurality of the commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and the commodity basic characteristics of the matched commodity with the unique identification of the commodity detail page recommendation scene to obtain the scene unique characteristics of the commodity detail page recommendation scene.
Step S1003, acquiring any plurality of commodity basic characteristics of the displayed commodity in the shopping cart, the similarity between the displayed commodity and the recommended commodity and commodity basic characteristics of the commodity purchased by the user in history, and constructing scene specific characteristics of a shopping cart recommended scene;
in order to attract the interest of the user to the commodities other than the displayed commodities in the shopping cart, the shopping cart is recommended to manually screen out any one or more commodities in the online shops, such as the similarity degree, the correlation degree and the correlation degree with the current user historical purchased commodities.
Therefore, before the commodity is recommended in the shopping cart recommendation scene, the commodity basic characteristics of the commodity displayed in the current shopping cart, the commodity basic characteristics of the commodity purchased in the current user history, the commodity description text and commodity picture of the displayed commodity, and the commodity description text and commodity picture of the manually screened recommended commodity are acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, and the unique identification of the commodity is spliced.
According to the relevant disclosure of step S1002, the similarity between the displayed commodity and the recommended commodity is determined according to the commodity description text and the commodity picture of the displayed commodity and the commodity description text and the commodity picture of the recommended commodity manually screened.
And splicing any plurality of commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and the commodity basic characteristics of the commodity purchased by the user in history with the unique identification of the shopping cart recommendation scene to obtain scene specific characteristics of the shopping cart recommendation scene.
Step S1004, acquiring any multiple items of commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and commodity basic characteristics of the preferential commodity in the settlement page, and constructing scene specific characteristics of a settlement page recommended scene;
in order to attract the interest of the user to the commodities other than the displayed commodity in the settlement page, the settlement page is recommended to manually screen out any one or more commodities with higher similarity degree, correlation degree and preferential collocation degree from the displayed commodity in the online store.
Therefore, before the commodity is recommended in the settlement page recommendation scene, the commodity basic characteristics of the commodity displayed in the current settlement page, the commodity basic characteristics of the preferential commodity matched with the displayed commodity to reach preferential conditions, the commodity description text and commodity picture of the displayed commodity, and the commodity description text and commodity picture of the manually screened recommended commodity can be acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, and the unique identification of the commodity is spliced.
According to the relevant disclosure of step S1002, the similarity between the displayed commodity and the recommended commodity is determined according to the commodity description text and the commodity picture of the displayed commodity and the commodity description text and the commodity picture of the recommended commodity manually screened.
And splicing any plurality of commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and the commodity basic characteristics of the preferential commodity with the unique identification of the settlement page recommendation scene to obtain the scene unique characteristics of the settlement page recommendation scene.
Step S1005, acquiring any multiple items of commodity basic features of historical click commodities, commodity basic features of historical purchase commodities and commodity basic features of displayed commodities in shopping carts of users, constructing scene specific features of a non-search result page recommendation scene, and constructing scene specific features of the non-search result page recommendation scene.
For users who have clear buying intention but do not search for goods meeting the buying intention, the non-search result page recommending scene is usually used for recommending and manually screening any one or more higher goods among the correlation degree between the non-search result page and the current user historical click goods in the online store, the correlation degree between the non-search result page and the current user historical purchase goods and the correlation degree between the non-search result page and the displayed goods in the current shopping cart in order to attract the interest of the users in the goods in the non-search result page.
Therefore, between the recommended commodities in the non-search result page recommendation scene, the commodity basic characteristics of the commodity clicked by the current user in history, the commodity basic characteristics of the commodity purchased by the current user in history and the commodity basic characteristics of the commodity displayed in the current shopping cart are acquired first to be spliced with the unique identification of the non-search result page recommendation scene, so that the scene specific characteristics of the non-search result page recommendation scene are acquired. The commodity basic feature is a word segmentation sequence obtained by word segmentation after splicing various texts in commodity description texts of corresponding commodities, and the unique identification of the commodity is spliced.
In this embodiment, the construction of scene-specific features of multiple recommended scenes is disclosed, so that the scene-specific features can be ensured to accurately influence the click rate and click conversion rate of recommended commodities.
Referring to fig. 7, in a further embodiment, step S1500 of screening out a list of recommended articles of a part of the construction of the articles to be recommended according to the click rate and click conversion rate of the articles to be recommended includes the following steps:
step S1510, calculating a recommendation score to be corrected according to weights corresponding to the click rate and click conversion rate matching of the commodity to be recommended;
it can be understood that the click rate and the click conversion rate have different influence degrees on the commodity purchase of the user, and in order to reasonably quantify the influence degrees corresponding to the click rate and the click conversion rate, the respective weights of the click rate and the click conversion rate are respectively set. The sum of the weights corresponding to the click rate and the click conversion rate is 1, and can be determined by a person skilled in the art according to priori knowledge, the weight of the exemplary click rate is 0.3, and the weight of the click conversion rate is 0.7.
And multiplying the click rate of the commodity to be recommended by the weight of the commodity to be recommended and the click conversion rate by the weight of the commodity to be recommended, and then summing to obtain the recommendation score to be corrected.
Step S1520, determining a risk coefficient according to the commodity basic characteristics of the commodity to be recommended by adopting a preset risk assessment model;
the risk factor quantification represents a likelihood of the item to be recommended being at risk.
Acquiring commodity description text of the commodity to be recommended from the commodity database according to the unique identifier of the commodity to be recommended, after each text in the commodity description text is spliced, word segmentation is carried out on the corresponding spliced text by adopting a word segmentation algorithm, and the unique identifier on the corresponding word segmentation sequence splice is used as commodity basic characteristics of the commodity to be recommended.
In an embodiment, the risk assessment model includes a text feature extraction module and a classifier followed by the text feature extraction module, where the text feature extraction module may be a model suitable for extracting text features in the NLP field, and the recommended model is a BERT model, or any other model, such as Transfomer Encoder, roBERTa, XLM-RoBERTa, MPNet, biLSTM, GPT, etc. The classifier is suitable for two classification tasks, and the selection can be MLP (feedforward neural network) or FC (fully connected layer).
And training the risk assessment model in advance until convergence, and obtaining the capability of determining the risk coefficient according to the commodity basic characteristics of the commodity. For the training process of the risk assessment model, acquiring commodity basic characteristics of a plurality of commodities with risks and commodity basic characteristics of a plurality of commodities without risks which are manually determined in advance and respectively used as training samples, and marking supervision labels of the training samples according to whether the training samples have risks or not, wherein the supervision labels of the training samples are marked as 1 according to whether the training samples have risks or not by using the training samples; the commodity basic feature of the training sample is not at risk, and the supervision label of the training sample is marked as 0. And forming a training set by using all training samples and supervision labels thereof. Obtaining a single training sample of the training set, inputting the training sample into a risk assessment model, extracting deep semantic information of the training sample by a text feature extraction module in the risk assessment model, determining a text feature vector representing the deep semantic information in a vectorization mode, mapping the text feature vector into preset binary categories by a classifier in the risk assessment model, wherein the binary categories comprise positive categories representing that input texts are at risk, negative categories representing that the input texts are not at risk, and obtaining prediction probabilities mapped into the positive categories in the binary categories as prediction risk coefficients. Invoking a preset cross entropy loss function or a mean square error loss function, wherein the cross entropy loss function or the mean square error loss function can be flexibly set by a person skilled in the art according to priori knowledge or experimental experience, calculating a loss value of the prediction risk coefficient according to a supervision label of the training sample, and when the loss value reaches a preset threshold value, indicating that the risk assessment model is trained to a convergence state, so that the risk assessment model training can be terminated; and when the loss value does not reach the preset threshold value, indicating that the risk assessment model is not converged, then implementing gradient update on the risk assessment model according to the loss value, correcting the weight parameters of each link of the risk assessment model through back propagation to enable the risk assessment model to further approach convergence, and then continuing to call other training samples to implement iterative training on the risk assessment model until the risk assessment model is trained to a convergence state. The preset threshold can be set as required.
Inputting commodity basic features of the commodity to be recommended into the risk assessment model, extracting deep semantic information of the commodity basic features by a text feature extraction module in the risk assessment model, determining text feature vectors representing the deep semantic information in a vectorization mode, mapping the text feature vectors into preset binary categories by a classifier in the risk assessment model, and obtaining probability of positive categories mapped into the binary categories as risk coefficients.
Step S1530, correcting the recommendation score to be corrected by adopting the risk coefficient of the commodity to be recommended to obtain a recommendation score;
when the risk coefficient exceeds a preset threshold, the risk coefficient is modified to be 1, and the preset threshold can be set as required by a person skilled in the art according to the disclosure. When the risk coefficient does not exceed the preset threshold value, the original value of the risk coefficient is reserved and is not changed.
And calculating a risk coefficient of subtracting the commodity to be recommended, and multiplying the risk coefficient by the recommendation score to be corrected to obtain a recommendation score.
Step S1540, screening out the commodities to be recommended, the recommendation scores of which meet preset conditions, as recommended commodities, and constructing a recommended commodity list.
And sorting all the recommended scores according to the sequence from high to low of the recommended scores aiming at the recommended scores of all the commodities to be recommended, screening out the commodities to be recommended corresponding to N recommended scores which are ranked at the front and have the recommended scores greater than 0 as recommended commodities, and sorting the recommended commodities according to the current sequence to form a recommended commodity list, wherein N can be set according to requirements.
In the embodiment, the to-be-corrected recommendation score is calculated through the weight corresponding to the click rate and click conversion rate matching of the to-be-recommended commodity, the risk coefficient is determined according to the commodity basic characteristics of the to-be-recommended commodity by the risk assessment model, the to-be-corrected recommendation score is obtained, the recommendation commodity list is constructed by optimizing the recommended commodity with the better recommendation score, the reliability and accuracy of the recommendation score can be ensured, the user is prevented from buying the at-risk recommended commodity, and the user satisfaction is improved.
Referring to fig. 8, in a further embodiment, before calculating the recommendation score to be corrected according to the weight corresponding to the click rate and click conversion rate matching of the commodity to be recommended, step S1510 includes the following steps:
step S1501, obtaining click browsing amount and click conversion amount of a user in a preset time period;
The click browsing amount is the number of times that a user clicks a commodity to browse details of the commodity, the click conversion amount is the number of times that the user executes conversion actions after clicking the commodity, and the conversion actions can be any one or any plurality of direct ordering purchase, shopping cart addition, collection, forwarding and the like.
The preset time period can be set as required, for example, any one of the last year, the last half year, the last 30 days, the last week, the last three days and the like.
Step S1502, determining the click rate of the commodity to be recommended and the weight corresponding to the click conversion rate according to the difference between the click browsing amount and the click conversion amount.
For the gap between the click through volume and the click conversion volume, the gap may be quantized as a ratio or a difference between the click through volume and the click conversion volume.
It can be understood that the click rate and the click conversion rate have different influence degrees on the commodity purchase of the user, and in order to reasonably quantify the influence degrees corresponding to the click rate and the click conversion rate, the respective weights of the click rate and the click conversion rate are respectively set. And the sum of the weights corresponding to the click rate and the click conversion rate is 1.
When the quantized value of the gap exceeds a preset threshold, the user will click the commodity and will execute the transformation action after clicking the commodity is weaker, namely the click rate of the commodity to be recommended is higher than the click conversion rate in the influence degree of the click conversion rate on commodity purchase of the user, so that the weight of the click rate can be set to be larger than the weight of the click conversion rate, the numerical value of each weight can be set as required, the weight of the click rate is 0.7, and the weight of the click conversion rate is 0.3.
When the quantized value of the gap does not exceed the preset threshold, the user's willingness to click the commodity is weak, but the willingness to execute the transformation behavior after clicking the commodity is strong, that is, the click conversion rate of the commodity to be recommended is higher than the click rate, so that the click conversion rate weight is larger than the click rate weight, the numerical value of each weight can be set as required, the exemplary example is that the click rate weight is 0.3, and the click conversion rate weight is 0.7. The preset threshold may be set as desired by one skilled in the art based on the disclosure herein.
In this embodiment, the click rate of the commodity to be recommended and the weight corresponding to the click conversion rate are determined according to the difference between the click browsing amount and the click conversion amount of the user in the preset time period, so that the subsequently obtained recommendation score can more accurately reflect the attraction of the commodity to be recommended to the user, and the recommended commodity which is expected to be promoted to be purchased by the user is selected accordingly.
Referring to fig. 9, an independent station commodity recommendation device provided for adapting to one of the purposes of the present application is a functional implementation of an independent station commodity recommendation method of the present application, and on the other hand, the device provided for adapting to one of the purposes of the present application includes a feature acquisition module 1100, a scene extraction module 1200, a store extraction module 1300, a task extraction module 1400, and a list construction module 1500, where the feature acquisition module 1100 is configured to acquire a scene specific feature of a scene to be recommended, a commodity basic feature and a commodity cross feature of a commodity to be recommended, a user feature, and a store cross feature; the scene extraction module 1200 is configured to determine a target scene extraction feature according to the scene characteristic feature, the user feature and the commodity basic feature by using a scene extraction layer in a preset commodity recommendation model; a store extraction module 1300, configured to determine a target store extraction feature according to the store intersection feature and the commodity intersection feature by using a store extraction layer in the commodity recommendation model; the task extraction module 1400 is configured to determine a click rate and a click conversion rate of the commodity to be recommended in the scene to be recommended according to the target scene extraction feature and the target store extraction feature by using a task output layer in the commodity recommendation model; the list construction module 1500 is configured to screen out a part of to-be-recommended commodity to construct a recommended commodity list according to the click rate and click conversion rate of the to-be-recommended commodity, and apply the list to the to-be-recommended scene.
In a further embodiment, the feature acquisition module 1100 includes: the first store feature construction submodule is used for acquiring main channels of marketing advertisements put in online stores to construct store marketing advertisement features; the second store characteristic construction submodule is used for acquiring any plurality of main commodity categories, main commodity attributes, main commodity brands, main commodity price intervals and main commodity sales areas of the online store and constructing store main commodity characteristics; the third store characteristic construction submodule is used for acquiring any one or all of a main camp logistics transportation mode and a main camp logistics transportation contractor of the online store to construct store main camp logistics characteristics; and the store cross feature construction submodule is used for splicing the store cross feature by the store marketing advertisement feature, the store main commodity feature and the store main commodity feature.
In a further embodiment, the feature acquisition module 1100 includes: the first commodity feature construction submodule is used for acquiring a channel for putting marketing advertisements of commodities to be recommended to construct commodity marketing advertisement features; the second commodity characteristic construction submodule is used for acquiring any multiple items of commodity class, commodity attribute, commodity brand, commodity price and commodity sales area of the commodity to be recommended and constructing commodity display characteristics; the third commodity characteristic construction submodule is used for acquiring any one or all of a commodity transportation mode and a commodity transportation contractor of the commodity to be recommended and constructing commodity logistics characteristics; and the commodity cross feature construction sub-module is used for splicing the commodity cross features according to the commodity marketing advertisement features, the commodity display features and the commodity logistics features.
In a further embodiment, before the feature acquisition module 1100, the method includes: the feature acquisition sub-module is used for acquiring scene specific features of various recommended scenes, commodity basic features, commodity cross features, user features and store cross features of corresponding recommended commodities in each recommended scene; the training set construction submodule is used for constructing a plurality of training samples by using scene specific characteristics of various recommended scenes, commodity basic characteristics, commodity cross characteristics, user characteristics and shop cross characteristics of corresponding recommended commodities in the recommended scenes, and constructing a training set by using all the training samples and the supervision labels thereof according to whether the user clicks the recommended commodities and whether the user executes conversion actions after clicking the recommended commodities in the recommended scenes of the training samples to correspondingly label the supervision labels of the training samples; the model training sub-module is used for training the commodity recommendation model by adopting a training set, so that the capability of determining the click rate and click conversion rate of the commodity under various recommendation scenes is learned.
In a further embodiment, the feature acquisition sub-module includes: the home page recommending feature unit is used for acquiring any multiple of a hot search commodity keyword, a hot sale commodity keyword and a hot shopping commodity keyword and constructing scene special features of a home page recommending scene; the detail recommending feature unit is used for acquiring any plurality of items of commodity basic features of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and commodity basic features of the matched commodity in the commodity detail page, and constructing scene special features of a commodity detail page recommending scene; the shopping cart recommendation feature unit is used for acquiring any multiple items of commodity basic features of displayed commodities in the shopping cart, similarity between the displayed commodities and recommended commodities and commodity basic features of historical purchased commodities of a user, and constructing scene specific features of a shopping cart recommendation scene; the settlement recommendation feature unit is used for acquiring any plurality of commodity basic features of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and commodity basic features of the preferential commodity in the settlement page, and constructing scene specific features of a settlement page recommendation scene; the non-search result recommending feature unit is used for acquiring any multiple items of commodity basic features of historical click commodities, commodity basic features of historical purchase commodities and commodity basic features of displayed commodities in shopping carts, constructing scene special features of a non-search result page recommending scene, and constructing scene special features of the non-search result page recommending scene.
In a further embodiment, the list construction module 1500 includes: the score calculating sub-module is used for calculating a recommendation score to be corrected according to the weight corresponding to the click rate and click conversion rate matching of the commodity to be recommended; the risk evaluation sub-module is used for determining a risk coefficient according to the commodity basic characteristics of the commodity to be recommended by adopting a preset risk evaluation model; the scoring correction sub-module is used for correcting the recommendation score to be corrected by adopting the risk coefficient of the commodity to be recommended to obtain a recommendation score; and the list construction sub-module is used for screening the commodities to be recommended, the recommendation scores of which meet preset conditions, as recommended commodities, and constructing a recommended commodity list.
In a further embodiment, before the score computation sub-module, the method includes: the user behavior data acquisition sub-module is used for acquiring click browsing quantity and click conversion quantity of a user in a preset time period; and the weight determining submodule is used for determining the click rate of the commodity to be recommended and the weight corresponding to the click conversion rate according to the difference between the click browsing amount and the click conversion amount.
In order to solve the technical problems, the embodiment of the application also provides computer equipment. As shown in fig. 10, the internal structure of the computer device is schematically shown. The computer device includes a processor, a computer readable storage medium, a memory, and a network interface connected by a system bus. The computer readable storage medium of the computer device stores an operating system, a database and computer readable instructions, the database can store a control information sequence, and when the computer readable instructions are executed by a processor, the processor can realize an independent station commodity recommendation method. The processor of the computer device is used to provide computing and control capabilities, supporting the operation of the entire computer device. The memory of the computer device may have stored therein computer readable instructions that, when executed by the processor, cause the processor to perform the stand alone commodity recommendation method of the present application. The network interface of the computer device is for communicating with a terminal connection. It will be appreciated by those skilled in the art that the structure shown in fig. 10 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
The processor in this embodiment is configured to execute specific functions of each module and its sub-modules in fig. 9, and the memory stores program codes and various types of data required for executing the above modules or sub-modules. The network interface is used for data transmission between the user terminal or the server. The memory in this embodiment stores program codes and data required for executing all modules/sub-modules in the individual station commodity recommendation device according to the present invention, and the server can call the program codes and data of the server to execute the functions of all sub-modules.
The present application also provides a storage medium storing computer readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the stand alone commodity recommendation method of any of the embodiments of the present application.
Those skilled in the art will appreciate that implementing all or part of the above-described methods of embodiments of the present application may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed, may comprise the steps of embodiments of the methods described above. The storage medium may be a computer readable storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a random access Memory (Random Access Memory, RAM).
In summary, the click rate and click conversion rate of the commodity to be recommended in the scene to be recommended can be accurately determined, so that the recommended commodity which can attract the user is preferably selected, and the user is hopeful to be promoted to purchase the recommended commodity.
Those of skill in the art will appreciate that the various operations, methods, steps in the flow, actions, schemes, and alternatives discussed in the present application may be alternated, altered, combined, or eliminated. Further, other steps, means, or steps in a process having various operations, methods, or procedures discussed in this application may be alternated, altered, rearranged, split, combined, or eliminated. Further, various operations, methods, steps, means, or arrangements of procedures found in the prior art with open sources in this application may also be alternated, altered, rearranged, split, combined, or deleted.
The foregoing is only a partial embodiment of the present application, and it should be noted that, for a person skilled in the art, several improvements and modifications can be made without departing from the principle of the present application, and these improvements and modifications should also be considered as the protection scope of the present application.
Claims (10)
1. The commodity recommending method for the independent station is characterized by comprising the following steps of:
Acquiring scene specific characteristics of a scene to be recommended, commodity basic characteristics and commodity cross characteristics of commodities to be recommended, user characteristics and shop cross characteristics;
a scene extraction layer in a preset commodity recommendation model is adopted to determine target scene extraction features according to the scene unique features, the user features and commodity basic features;
determining target shop extraction features according to the shop cross features and the commodity cross features by adopting a shop extraction layer in the commodity recommendation model;
determining the click rate and click conversion rate of the commodity to be recommended in the scene to be recommended according to the target scene extraction characteristics and the target store extraction characteristics by adopting a task output layer in the commodity recommendation model;
and screening out partial commodity to be recommended according to the click rate and click conversion rate of the commodity to be recommended to construct a recommended commodity list, and applying the list to the scene to be recommended.
2. The stand-alone commodity recommendation method according to claim 1, wherein acquiring scene-specific characteristics of a scene to be recommended, commodity basic characteristics and commodity cross characteristics of a commodity to be recommended, user characteristics, and shop cross characteristics, comprises the steps of:
Acquiring main channels of marketing advertisements put in online shops to construct marketing advertisement features of the shops;
acquiring any multiple items of main commodity category, main commodity attribute, main commodity brand, main commodity price interval and main commodity sales area of an online store, and constructing store main commodity characteristics;
acquiring any one or all of main camp logistics transportation modes and main camp logistics transportation contractors of an online store, and constructing store main camp logistics characteristics;
and splicing the store marketing advertisement feature, the store main commodity feature and the store main commodity logistics feature into a store cross feature.
3. The stand-alone commodity recommendation method according to claim 1, wherein acquiring scene-specific characteristics of a scene to be recommended, commodity basic characteristics and commodity cross characteristics of a commodity to be recommended, user characteristics, and shop cross characteristics, comprises the steps of:
acquiring a channel for putting marketing advertisements of goods to be recommended to construct a commodity marketing advertisement feature;
acquiring any multiple items of commodity category, commodity attribute, commodity brand, commodity price and commodity sales area of the commodity to be recommended, and constructing commodity display characteristics;
any one or all of a logistics transportation mode and a logistics transportation contractor of the commodity to be recommended are obtained, and commodity logistics characteristics are constructed;
And splicing the commodity marketing advertisement feature, the commodity display feature and the commodity logistics feature into commodity cross features.
4. The stand-alone commodity recommendation method according to claim 1, wherein before acquiring the scene unique feature of the scene to be recommended, the commodity basic feature and commodity cross feature of the commodity to be recommended, the user feature, the shop cross feature, comprising the steps of:
acquiring scene unique features of various recommended scenes, and commodity basic features, commodity cross features, user features and store cross features of corresponding recommended commodities in each recommended scene;
constructing a plurality of training samples by using scene specific features of various recommended scenes, commodity basic features, commodity cross features, user features and shop cross features of corresponding recommended commodities in the recommended scenes, and constructing a training set by using all the training samples and the supervision labels thereof according to whether a user clicks the recommended commodities and whether to execute conversion actions after clicking the recommended commodities in the recommended scenes of the training samples to correspondingly label the supervision labels of the training samples;
and training the commodity recommendation model by adopting a training set, so that the capability of determining the click rate and click conversion rate of the commodity under various recommendation scenes is learned.
5. The stand-alone commodity recommendation method according to claim 4, wherein obtaining scene-specific features of a plurality of recommended scenes comprises the steps of:
acquiring any multiple of hot search commodity keywords, hot sale commodity keywords and hot shopping commodity keywords, and constructing scene specific features of a home page recommended scene;
acquiring any multiple items of commodity basic characteristics of the displayed commodity, similarity between the displayed commodity and the recommended commodity and commodity basic characteristics of the matched commodity in the commodity detail page, and constructing scene special characteristics of a commodity detail page recommended scene;
acquiring any multiple items of commodity basic characteristics of the displayed commodity in the shopping cart, similarity between the displayed commodity and the recommended commodity and commodity basic characteristics of the commodity purchased by the user in history, and constructing scene specific characteristics of a shopping cart recommended scene;
acquiring any plurality of commodity basic characteristics of the displayed commodity, the similarity between the displayed commodity and the recommended commodity and commodity basic characteristics of the preferential commodity in the settlement page, and constructing scene specific characteristics of a settlement page recommended scene;
and acquiring any multiple items of commodity basic characteristics of historical clicked commodities, commodity basic characteristics of historical purchased commodities and commodity basic characteristics of displayed commodities in shopping carts of users, constructing scene specific characteristics of a scene recommended by the pages without search results, and constructing scene specific characteristics of the scene recommended by the pages without search results.
6. The method for recommending commodities in an independent station according to claim 1, wherein the step of screening out a part of the commodity to be recommended to construct a recommended commodity list according to the click rate and the click conversion rate of the commodity to be recommended comprises the following steps:
calculating a recommendation score to be corrected according to the weight corresponding to the click rate and click conversion rate matching of the commodity to be recommended;
determining a risk coefficient according to the commodity basic characteristics of the commodity to be recommended by adopting a preset risk assessment model;
correcting the recommendation score to be corrected by adopting the risk coefficient of the commodity to be recommended to obtain a recommendation score;
and screening the commodities to be recommended, the recommendation scores of which meet preset conditions, as recommended commodities, and constructing a recommended commodity list.
7. The method for recommending stand alone commodity according to claim 6, wherein before calculating the recommendation score to be corrected according to the weight corresponding to the click rate and click conversion rate matching of the commodity to be recommended, comprising the steps of:
acquiring click browsing quantity and click conversion quantity of a user in a preset time period;
and determining the click rate of the commodity to be recommended and the weight corresponding to the click conversion rate according to the gap between the click browsing amount and the click conversion amount.
8. An independent station commodity recommendation device, characterized by comprising:
the feature acquisition module is used for acquiring scene specific features of a scene to be recommended, commodity basic features and commodity cross features of commodities to be recommended, user features and shop cross features;
the scene extraction module is used for determining target scene extraction features according to the scene unique features, the user features and the commodity basic features by adopting a scene extraction layer in a preset commodity recommendation model;
the store extraction module is used for determining target store extraction characteristics according to the store cross characteristics and the commodity cross characteristics by adopting a store extraction layer in the commodity recommendation model;
the task extraction module is used for determining the click rate and click conversion rate of the commodity to be recommended in the scene to be recommended according to the target scene extraction characteristics and the target shop extraction characteristics by adopting a task output layer in the commodity recommendation model;
the list construction module is used for screening out a part of commodity construction recommended commodity list to be recommended according to the clicking rate and clicking conversion rate of the commodity to be recommended, and applying the commodity list to the scene to be recommended.
9. A computer device comprising a central processor and a memory, characterized in that the central processor is arranged to invoke a computer program stored in the memory for performing the steps of the method according to any of claims 1 to 7.
10. A computer-readable storage medium, characterized in that it stores in the form of computer-readable instructions a computer program implemented according to the method of any one of claims 1 to 7, which, when invoked by a computer, performs the steps comprised by the corresponding method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311388440.4A CN117350816A (en) | 2023-10-24 | 2023-10-24 | Independent station commodity recommendation method and device, equipment and medium thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311388440.4A CN117350816A (en) | 2023-10-24 | 2023-10-24 | Independent station commodity recommendation method and device, equipment and medium thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117350816A true CN117350816A (en) | 2024-01-05 |
Family
ID=89366380
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311388440.4A Pending CN117350816A (en) | 2023-10-24 | 2023-10-24 | Independent station commodity recommendation method and device, equipment and medium thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117350816A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117994010A (en) * | 2024-04-07 | 2024-05-07 | 深圳市正东兴通讯设备有限公司 | Intelligent screen information recommendation method based on big data |
-
2023
- 2023-10-24 CN CN202311388440.4A patent/CN117350816A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117994010A (en) * | 2024-04-07 | 2024-05-07 | 深圳市正东兴通讯设备有限公司 | Intelligent screen information recommendation method based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6356744B2 (en) | Method and system for displaying cross-website information | |
| US10354184B1 (en) | Joint modeling of user behavior | |
| US10180979B2 (en) | System and method for generating suggestions by a search engine in response to search queries | |
| CN111784455A (en) | Article recommendation method and recommendation equipment | |
| CN106600302A (en) | Hadoop-based commodity recommendation system | |
| CN110020094A (en) | A kind of methods of exhibiting and relevant apparatus of search result | |
| US11580585B1 (en) | User-preferred item attributes | |
| CN108897823B (en) | Personalized commodity retrieval method and device based on deep learning attention mechanism | |
| CN114663197A (en) | Commodity recommendation method and device, equipment, medium and product thereof | |
| CN113793182B (en) | Commodity object recommendation method and device, equipment, medium and product thereof | |
| CN117350816A (en) | Independent station commodity recommendation method and device, equipment and medium thereof | |
| CN113627995A (en) | Commodity recommendation list updating method and device, equipment, medium and product thereof | |
| CN113327151A (en) | Commodity object recommendation method and device, computer equipment and storage medium | |
| CN115545832A (en) | Commodity search recommendation method and device, equipment and medium thereof | |
| CN114862480A (en) | Advertisement putting orientation method and its device, equipment, medium and product | |
| CN116976920A (en) | Commodity shopping guide method and device, equipment and medium thereof | |
| CN116089745A (en) | Information recommendation method, device, electronic equipment and computer readable storage medium | |
| CN113971599A (en) | Advertisement putting and selecting method and device, equipment, medium and product thereof | |
| KR101473692B1 (en) | System and Method for Intelligent Golf Booking and Join by Context-awareness based on Sensor and Personal Information Learning of Smart Phone User | |
| CN116823404A (en) | Commodity combination recommendation method, device, equipment and medium thereof | |
| Ou et al. | An Intelligent Recommendation System for Real Estate Commodity. | |
| CN110781399A (en) | Cross-platform information pushing method and device | |
| CN115293818A (en) | Advertisement putting and selecting method and device, equipment and medium thereof | |
| CN113704630A (en) | Information pushing method and device, readable storage medium and electronic equipment | |
| Chew et al. | Recommender System for Retail Domain: An Insight on Techniques and Evaluations |
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 |