CN117422527B - Brand new commodity cold starting method based on user preference perception - Google Patents
Brand new commodity cold starting method based on user preference perception Download PDFInfo
- Publication number
- CN117422527B CN117422527B CN202311526419.6A CN202311526419A CN117422527B CN 117422527 B CN117422527 B CN 117422527B CN 202311526419 A CN202311526419 A CN 202311526419A CN 117422527 B CN117422527 B CN 117422527B
- Authority
- CN
- China
- Prior art keywords
- commodity
- user
- cooperative
- characteristic
- content
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 22
- 230000008447 perception Effects 0.000 title claims abstract description 18
- 230000004931 aggregating effect Effects 0.000 claims abstract description 6
- 238000005457 optimization Methods 0.000 claims abstract description 3
- 230000003993 interaction Effects 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 230000002195 synergetic effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000006467 substitution reaction 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a brand new commodity cold start method based on user preference perception, which comprises the following steps: acquiring user cooperative characteristics, commodity cooperative characteristics and commodity content characteristics; based on the user cooperative characteristics and commodity cooperative characteristics, establishing a relationship model of the user and the commodity; based on commodity cooperative features and commodity content features, aligning the content features and the cooperative features of the same commodity to obtain a commodity level feature alignment model; aggregating the cooperative characteristics of all commodities except the target commodity in the same user purchase record to obtain user preference; aligning the user preference with the commodity content characteristic to obtain a group level characteristic alignment model; and carrying out joint loss optimization on the relation model of the user and the commodity, the commodity level characteristic alignment model and the group level characteristic alignment model to obtain a joint loss model, and recommending the original commodity and the brand new commodity through the cooperative characteristic and the content characteristic. The invention can better solve the problem of cold start of brand new commodities.
Description
Technical Field
The invention belongs to the technical field of recommendation algorithms, and particularly relates to a brand new commodity cold start method based on user preference perception.
Background
With the explosive growth of information, a great number of brand new commodities exist in the platform every day in the electronic commerce field. Although recommendation systems based on collaborative filtering algorithms have met with significant success in personalizing recommendation tasks. These collaborative filtering-based systems are intended to learn high quality collaborative features from historical interactions (e.g., clicks, browses, collections, etc.) to represent users and merchandise. However, these methods cannot be used to characterize these brand new products without any historical interaction information, which creates a cold start problem for the brand new products.
Most existing methods characterize these new goods by introducing their multimedia information as content features. However, these methods only adopt the alignment policy of commodity level to align the content features and the cooperative features of the commodity, and the method is still not fully applicable to solving the problem of cold start of brand new commodity lacking cooperative features. Therefore, the invention provides a brand new commodity cold start method based on user preference perception, which is used for solving the problem of brand new commodity cold start.
Disclosure of Invention
In order to solve the technical problems, the invention provides a brand new commodity cold start method based on user preference perception, which is based on a comparison learning method, and solves the brand new commodity cold start problem by respectively establishing an alignment strategy between the same commodity content characteristics and the cooperative characteristics at a single commodity level and an alignment strategy between the user preference and the commodity content characteristics at a group level.
In order to achieve the above object, the present invention provides a brand new method for cold starting of goods based on user preference perception, comprising:
Acquiring user cooperative characteristics, commodity cooperative characteristics and commodity content characteristics;
Based on the user cooperative characteristics and the commodity cooperative characteristics, establishing a relationship model of the user and the commodity;
Based on the commodity cooperative features and the commodity content features, aligning the content features of the same commodity with the cooperative features to obtain a commodity level feature alignment model;
aggregating the cooperative characteristics of all commodities except the target commodity in the same user purchase record to obtain user preference;
aligning the user preference with the commodity content feature to obtain a group level feature alignment constraint model;
Performing joint loss optimization on the relation model of the user and the commodity, the commodity level characteristic alignment model and the group level characteristic alignment model to obtain a joint loss model;
and acquiring the content characteristics of the brand-new commodity through the joint loss model and recommending the brand-new commodity.
Optionally, acquiring the user cooperation feature, the commodity cooperation feature and the commodity content feature includes:
Adopting a double-branch structure, wherein the double-branch structure comprises an interaction branch and a content branch, wherein the interaction branch respectively establishes an embedded matrix U, I for a user and a commodity, a row vector of U represents a user cooperative characteristic, and a row vector of I represents a commodity cooperative characteristic;
The interaction branch acquires the user cooperative characteristics and the commodity cooperative characteristics according to historical interaction information of the commodity and the user;
And the content branches acquire the content characteristics of the commodity according to commodity picture information.
Optionally, the acquiring the content feature of the commodity according to the commodity picture information by the content branch includes:
encoding commodity pictures based on ResNet models to obtain f-dimension content characteristics v q;
Based on the content characteristics v q of the f dimension, acquiring the content characteristics of the commodity by using a two-layer fully connected network:
cq=W2·δ(W1·vq+b1)+b2
where W 1、W2 and b 1、b2 are the linear mapping matrix and bias, respectively, of the fully connected neural network, δ (·) represents the LeaklyRelu function, c q is the commodity content feature, and v q is the f-dimensional content feature.
Optionally, based on the user cooperative feature and the commodity cooperative feature, establishing a relationship model between the user and the commodity includes:
If interaction exists between the user h and the commodity q, the user cooperative feature u h is used as an anchor vector, the commodity cooperative feature i q is used as a positive sample, and the uncorrelated commodity cooperative feature set of the commodity q The cooperative characteristic of the commodity in (3) is taken as a negative sample, and a first user and commodity comparison sample subset {(uh,iq),(uh,iq,1),(uh,iq,2),...,(uh,iq,K)}; is constructed, wherein i q,j is the cooperative characteristic of commodity j which is irrelevant to commodity q,/>For a set of users interacting with item q, K represents/>Number of negative samples included in,/>A commodity set interacted with the user j;
Taking the commodity cooperative feature i q as an anchor vector, taking the user cooperative feature u h as a positive sample and collecting The cooperative characteristic of the commodity is a negative sample, and a second user and commodity comparison sample subset {(iq,uh),(iq,uh,1),(iq,uh,2),...,(iq,uh,L)}; is constructed, wherein/>Representing the set of goods purchased by user h, L represents/>The number of negative samples included in the method, u h,t, is the cooperative characteristic of the user t which is irrelevant to the user h;
Constructing a user and commodity comparison sample set based on the first user and commodity comparison sample subset and the second user and commodity comparison sample set;
And based on the user and commodity comparison sample set, establishing a relationship model of the user and commodity by using a comparison loss function.
Optionally, the relationship model between the user and the commodity is:
Wherein, Constraint model for relationship between user and commodity,/>Information collection/>, for interaction between user h and commodity qIn (2), beta is a super parameter, i q,j is a cooperative feature of commodity j unrelated to commodity q, and u h,t is a cooperative feature of user t unrelated to user h.
Optionally, obtaining the commodity level feature alignment model includes:
taking the commodity cooperative characteristic i q as an anchor vector, taking the commodity content characteristic c q as a positive sample and taking the uncorrelated content characteristic set of the commodity q The content features of the commodity in (a) are taken as negative samples, and a first commodity level feature alignment comparison sample set {(iq,cq),(iq,cq,1),(iq,cq,2),...,(iq,cq,Z)}; is constructed, wherein/>For a user collection who purchased commodity j, c q,j is the content feature of commodity j that is not related to commodity q, Z is the collection/>The number of negative samples;
Taking the commodity content characteristic c q as an anchor vector, the commodity cooperative characteristic i q as a positive sample, and the uncorrelated commodity cooperative characteristic set of the commodity q The cooperative characteristic of the commodity is a negative sample, and a second commodity level characteristic alignment comparison sample set is constructed {(cq,iq),(cq,iq,1),(cq,iq,2),...,(cq,iq,K)};
Constructing a commodity level feature alignment sample set based on the first commodity level feature alignment sample subset and the second commodity level feature alignment sample subset;
And acquiring the commodity level feature alignment model based on the commodity level feature alignment comparison sample set.
Optionally, the commodity level feature alignment constraint model is:
Wherein, For the commodity level feature alignment constraint model, α is a superparameter, and c q,j is the content feature of commodity j that is not related to commodity q.
Optionally, aggregating the collaborative features of all the commodities except the target commodity in the same user purchase record, and acquiring the user preference includes:
By integrating user h histories except commodity q The cooperative characteristics of the original commodities are aggregated, and the user preference p h is obtained:
Wherein i j is the collaborative feature of commodity j.
Optionally, obtaining the group level feature alignment model includes:
Based on the user preference p h, obtaining a preference perception comparison sample set {(ph,cq),(ph,cq,1),(ph,cq,2),...,(ph,cq,Z)};
Based on the preference perception comparison sample set, aligning the user preference with the commodity content characteristic to obtain the group level characteristic alignment model;
optionally, the joint loss model is:
Wherein lambda, eta and mu are super parameters, theta is all the learnable parameters in the model, Respectively a relation constraint model of a user and commodities, a commodity level characteristic alignment constraint model and a group level characteristic alignment constraint model,/>Is a joint loss model.
The invention has the technical effects that: the invention establishes the relationship between the content characteristics of the target commodity and the cooperative characteristics of a plurality of commodities with the same cooperative signals from the group level by proposing preference perception contrast loss. And the recommendation effect of the original commodity and the brand new commodity is balanced better by the combination of the commodity relation establishment of the user, commodity level characteristic alignment and group level characteristic alignment, so that the problem of cold start of the brand new commodity is solved better.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:
FIG. 1 is a schematic diagram of an overall technical route of a brand new method for cold starting merchandise based on user preference perception according to an embodiment of the present invention;
Fig. 2 is a schematic diagram of an overall model structure of a model proposed by a brand new commodity cold start method based on user preference perception according to an embodiment of the present invention.
Detailed Description
It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other. The application will be described in detail below with reference to the drawings in connection with embodiments.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
As shown in fig. 1-2, in this embodiment, a brand new method for cold starting of commodities based on user preference perception is provided, and the whole architecture can be divided into two parts, wherein the two parts include user cooperation features, commodity cooperation features and acquisition of commodity content features, and establishment of a joint loss. The joint loss comprises three parts: a relationship constraint between a user and a commodity, an alignment constraint of two features of the same commodity at the commodity level, and an alignment constraint between user preferences at the group level and commodity content. The method specifically comprises the following steps:
Step1, feature coding: acquiring cooperative characteristics of a user and commodities according to the historical interaction information; acquiring content characteristics of the commodity according to commodity picture information;
step2, establishing commodity relation of the user: establishing a relation between a user and a commodity according to the historical information, and modeling whether the relation exists between the user and the commodity or not by utilizing contrast learning;
Step 3, commodity level characteristic alignment: constructing a comparison sample set for commodity level alignment in a dual mode, and aligning the content features and the cooperative features of the same commodity through a comparison learning function;
Step 4, group level feature alignment: and (3) by aggregating the cooperative characteristics of all commodities except the target commodity in the same user purchase record as user preferences, establishing a user preference perception comparison sample set, and aligning the user preferences with the commodity content characteristics by utilizing a comparison learning function.
Step 1 feature encoding section. In the characteristic coding stage, the invention adopts a double-branch structure. In the double-branch structure, one branch is used for acquiring the cooperative characteristics of the user and the cooperative characteristics of the commodity from the interactive information of the commodity and the user, and is named as an interactive branch; the other branch is used for acquiring the content characteristics of the commodity from the commodity image and is named as a content branch.
In the interaction branch, the invention respectively establishes two embedded matrixes for users and commodities Wherein the row vector of U represents a user cooperative feature, the row vector of I represents a commodity cooperative feature, and d, M and N are the dimension of the user vector, the number of users and the number of commodities respectively. In the content branch, the commodity image is firstly input into a ResNet model which is trained, and a content feature v q with f dimension is obtained. After that, the content characteristics/>, of the commodity are further obtained by utilizing a two-layer fully connected networkThe specific mode is as follows:
Cq=W2·δ(W1·Vq+b1)+b2
Wherein the method comprises the steps of And/>The linear mapping matrix and the bias, delta (·) of the fully connected neural network represent LeaklyRelu functions, respectively.
And 2, establishing a commodity relation of the user. If interaction exists between the h user and the q commodity, the cooperative characteristic of the q commodityThe cooperative features of other commodities are closer to those of the h user/>Here,/>Representing a set of users interacting with item q, where (U) refers to the set being a set of users,/>A set of items purchased by user h is represented, where (I) means that the set is a set of items. The user and commodity comparison sample set comprises two subsets, wherein one subset takes u h as an anchor vector, i q as a positive sample and an uncorrelated commodity cooperative feature set of commodity qThe synergistic features of the commodity in (a) are negative samples, the subset can be expressed as {(uh,iq),(uh,iq,1),(uh,iq,2),...,(uh,iq,K)}; another subset, i g is taken as an anchor vector, u h is taken as a positive sample, and the uncorrelated user synergistic feature set/>As a negative example, the subset may be represented as {(iq,uh),(iq,uh,1),(iq,uh,2),...,(iq,uh,L)}, where L represents/>The number of negative samples contained in the sample, K, represents/>Number of negative samples included in,/>For a commodity set that has interacted with user j, u h,t is the collaborative feature of user t that is not related to user h. Based on the user and commodity comparison sample set, the relationship constraint (relationship model of the user and commodity) between the user and the commodity is established by utilizing the comparison loss functionThe following are provided:
Wherein, Information collection/>, for interaction between user h and commodity qIn (2), beta is a super parameter, i q,j is a cooperative feature of commodity j unrelated to commodity q, and u h,t is a cooperative feature of user t unrelated to user h.
Step 3 commodity level feature alignment part. In order to maximize the mutual information of the content features and the collaborative features of the same commodity, the commodity level feature alignment comparison sample set contains two subsets. The first subset takes the cooperative characteristic i q of q commodities as an anchor vector, and the corresponding content characteristic of the cooperative characteristic i q As a positive sample, the uncorrelated content feature set of commodity qThe content of the commodity in (c) is characterized as a negative sample, and the subset can be expressed as {(iq,cq),(iq,cq,1),(iq,cq,2),…,(iq,cq,Z)},/>For a user collection that purchased commodity h, Z is collection/>The number of negative samples; the other subset takes the content characteristic c q of q commodity as an anchor vector and the corresponding cooperative characteristic i q as a positive sample, and the set/>The subset can be expressed as {(cq,iq),(cq,iq,1),(cq,iq,2),...,(cq,iq,K)}., based on the commodity level feature alignment comparison sample set, the established relationship constraint (commodity level feature alignment constraint model)/>, of commodity content features and commodity cooperative features of the commodity levelThe following are provided:
Where α is a superparameter and c q,j is the content feature of commodity j that is not related to commodity q.
Step 4 group level feature alignment. In order to mine the relation between the content characteristics of the target commodity and the cooperative characteristics of the original commodity from the group hierarchy, user preferences are introduced, and a preference perception comparison sample set is constructed. Wherein the preference of user hCommodity collection/>, can be interacted through user h history except commodity qThe original commodity cooperative characteristics are polymerized, and the specific calculation mode is as follows:
After this, it is assumed that when an interaction occurs between the h user and the q merchandise, the content features c q of the q merchandise will be compared to the set The content characteristics of the other items in (a) are closer to the preferences p h of the h user. The set of preference-aware comparison examples can be represented as {(ph,cq),(ph,cq,1),(ph,cq,2),...,(ph,cq,Z)}., on the basis of which alignment between user preferences and merchandise content features is accomplished using the following constraints, a group-level feature alignment model (group-level feature alignment constraint) is obtained:
Where i j is the collaborative feature of commodity j, and c q,j is the content feature of commodity j that is not related to commodity q.
Finally, the three constraints are optimized in the form of joint loss. By the method, high-quality content features can be generated for brand-new commodities and used as characterization results thereof for recommendation, and the problem of cold start of the brand-new commodities is solved. Joint lossThe concrete representation is as follows:
wherein λ, η and μ are all hyper-parameters. Θ is all the learnable parameters in the model.
The invention provides a brand new commodity cold starting method perceived by user preference, which not only aligns the content characteristics and the cooperative characteristics of the same commodity from a commodity level, but also models the user preference from a group level and aligns the user preference with the content characteristics of the commodity, thereby improving the characterization quality of the brand new commodity and better solving the problem of brand new commodity cold starting. The invention can realize the effect of recommending original commodity and brand new commodity according to the cooperative characteristic and the content characteristic respectively, and solves the problem of cold start of brand new commodity.
The present application is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present application are intended to be included in the scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims (1)
1. A brand new commodity cold start method based on user preference perception is characterized by comprising the following steps:
Acquiring user cooperative characteristics, commodity cooperative characteristics and commodity content characteristics;
Based on the user cooperative characteristics and the commodity cooperative characteristics, establishing a relationship model of the user and the commodity;
Based on the commodity cooperative features and the commodity content features, aligning the content features of the same commodity with the cooperative features to obtain a commodity level feature alignment model;
aggregating the cooperative characteristics of all commodities except the target commodity in the same user purchase record to obtain user preference;
Aligning the user preference with the commodity content feature to obtain a group level feature alignment model;
Performing joint loss optimization on the relation model of the user and the commodity, the commodity level characteristic alignment model and the group level characteristic alignment model to obtain a joint loss model;
acquiring content characteristics of brand-new commodities and recommending the brand-new commodities through the joint loss model;
The acquiring the user cooperative characteristic, the commodity cooperative characteristic and the commodity content characteristic comprises the following steps:
Adopting a double-branch structure, wherein the double-branch structure comprises an interaction branch and a content branch, wherein the interaction branch respectively establishes an embedded matrix U, I for a user and a commodity, a row vector of U represents a user cooperative characteristic, and a row vector of I represents a commodity cooperative characteristic;
The interaction branch acquires the user cooperative characteristics and the commodity cooperative characteristics according to historical interaction information of the commodity and the user;
the content branches acquire the content characteristics of the commodity according to commodity picture information;
The content branching comprises the steps of:
encoding commodity pictures based on ResNet models to obtain f-dimension content characteristics v q;
Based on the content characteristics v q of the f dimension, acquiring the content characteristics of the commodity by using a two-layer fully connected network:
cq=W2·δ(W1·vq+b1)+b2
Wherein, W 1、W2 and b 1、b2 are the linear mapping matrix and bias of the fully connected neural network, respectively, δ (·) represents LeaklyRelu function, c q is commodity content feature, and v q is f-dimensional content feature;
Based on the user cooperative features and the commodity cooperative features, establishing a relationship model between the user and the commodity comprises:
If interaction exists between the user h and the commodity q, the user cooperative feature u h is used as an anchor vector, the commodity cooperative feature i q is used as a positive sample, and the uncorrelated commodity cooperative feature set of the commodity q The cooperative characteristic of the commodity in (3) is taken as a negative sample, and a first user and commodity comparison sample subset {(uh,iq),(uh,iq,1),(uh,iq,2),...,(uh,iq,K)}; is constructed, wherein i q,j is the cooperative characteristic of commodity j which is irrelevant to commodity q,/>For a set of users interacting with item q, K represents/>Number of negative samples included in,/>A user collection for purchasing commodity j;
Taking the commodity cooperative feature i q as an anchor vector, the user cooperative feature u h as a positive sample, and an uncorrelated user cooperative feature set of a user h The cooperative characteristic of the commodity is a negative sample, and a second user and commodity comparison sample subset {(iq,uh),(iq,uh,1),(iq,uh,2),...,(iq,uh,L)}; is constructed, wherein/>Representing the set of goods purchased by user h,/>For a collection of items that have interacted with user j, L represents/>The number of negative samples included in the method, u h,t, is the cooperative characteristic of the user t which is irrelevant to the user h;
Constructing a user and commodity comparison sample set based on the first user and commodity comparison sample subset and the second user and commodity comparison sample set;
based on the user and commodity comparison sample set, a comparison loss function is utilized to establish a relationship model of the user and commodity;
the relation model of the user and the commodity is as follows:
Wherein, Constraint model for relationship between user and commodity,/>Information collection/>, for interaction between user h and commodity qΒ is a super parameter, i q,j is a cooperative feature of commodity j unrelated to commodity q, u h,t is a cooperative feature of user t unrelated to user h;
Acquiring the commodity level feature alignment model comprises:
taking the commodity cooperative characteristic i q as an anchor vector, taking the commodity content characteristic c q as a positive sample and taking the uncorrelated content characteristic set of the commodity q As a negative sample, a first commodity level feature alignment comparison sample set {(iq,cq),(iq,cq,1),(iq,cq,2),...,(iq,vq,Z)}; is constructed in which,For a user collection who purchased commodity j, c q,j is the content feature of commodity j that is not related to commodity q, Z is the collection/>The number of negative samples;
Taking the commodity content characteristic c q as an anchor vector, the commodity cooperative characteristic i q as a positive sample, and the uncorrelated commodity cooperative characteristic set of the commodity q The cooperative characteristic of the commodity is a negative sample, and a second commodity level characteristic alignment comparison sample set is constructed {(cq,iq),(cq,iq,1),(cq,iq,2),...,(cq,iq,K)};
Constructing a commodity level feature alignment sample set based on the first commodity level feature alignment sample subset and the second commodity level feature alignment sample subset;
acquiring the commodity level feature alignment model based on the commodity level feature alignment comparison sample set;
the commodity level feature alignment constraint model is:
Wherein, Aligning constraint models for commodity level features, wherein alpha is a super parameter, and c q,j is the content feature of commodity j which is irrelevant to commodity q;
aggregating the collaborative features of all but the target commodity in the same user purchase record, wherein the obtaining the user preference comprises the following steps:
By integrating user h histories except commodity q The cooperative characteristics of the original commodities are aggregated, and the user preference p h is obtained:
Wherein i j is the collaborative feature of commodity j;
obtaining the group level feature alignment model includes:
Based on the user preference p h, obtaining a preference perception comparison sample set {(ph,cq),(ph,cq,1),(ph,cq,2),...,(ph,cq,z)};
Based on the preference perception comparison sample set, aligning the user preference with the commodity content characteristic to obtain the group level characteristic alignment model;
The joint loss model is:
Wherein lambda, eta and mu are super parameters, theta is all the learnable parameters in the model, Respectively a relation constraint model of a user and commodities, a commodity level characteristic alignment constraint model and a group level characteristic alignment constraint model,/>Is a joint loss model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311526419.6A CN117422527B (en) | 2023-11-15 | 2023-11-15 | Brand new commodity cold starting method based on user preference perception |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311526419.6A CN117422527B (en) | 2023-11-15 | 2023-11-15 | Brand new commodity cold starting method based on user preference perception |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117422527A CN117422527A (en) | 2024-01-19 |
CN117422527B true CN117422527B (en) | 2024-05-03 |
Family
ID=89526455
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311526419.6A Active CN117422527B (en) | 2023-11-15 | 2023-11-15 | Brand new commodity cold starting method based on user preference perception |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117422527B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20120003362A (en) * | 2010-07-02 | 2012-01-10 | 아코지토(주) | Recommendation system on personalized favorite programs/channels for internet protocol tv users by using collaborative filtering algorithm |
CN116738053A (en) * | 2023-06-16 | 2023-09-12 | 人民网股份有限公司 | Cross-domain news recommendation system and recommendation method based on text implication |
WO2023185320A1 (en) * | 2022-03-30 | 2023-10-05 | 腾讯科技(深圳)有限公司 | Cold start object recommendation method and apparatus, computer device and storage medium |
-
2023
- 2023-11-15 CN CN202311526419.6A patent/CN117422527B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20120003362A (en) * | 2010-07-02 | 2012-01-10 | 아코지토(주) | Recommendation system on personalized favorite programs/channels for internet protocol tv users by using collaborative filtering algorithm |
WO2023185320A1 (en) * | 2022-03-30 | 2023-10-05 | 腾讯科技(深圳)有限公司 | Cold start object recommendation method and apparatus, computer device and storage medium |
CN116738053A (en) * | 2023-06-16 | 2023-09-12 | 人民网股份有限公司 | Cross-domain news recommendation system and recommendation method based on text implication |
Non-Patent Citations (1)
Title |
---|
基于社交关系拓扑结构的冷启动推荐方法;张亚楠;曲明成;刘宇鹏;;浙江大学学报(工学版);20160515(第05期);第1000-1002页 * |
Also Published As
Publication number | Publication date |
---|---|
CN117422527A (en) | 2024-01-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109767255B (en) | Method for realizing intelligent operation and accurate marketing through big data modeling | |
CN111611472B (en) | Binding recommendation method and system based on graph convolution neural network | |
Wang et al. | Applying consistent fuzzy preference relations to partnership selection | |
Phillips et al. | The knowledge society's origins and current trajectory | |
Xu | Deviation measures of linguistic preference relations in group decision making | |
CN110334284B (en) | Novel recommendation method integrating multi-factor decision making | |
Namvar et al. | Simplifying sensemaking: Concept, process, strengths, shortcomings, and ways forward for information systems in contemporary business environments | |
Kim et al. | Exploring innovation ecosystem of incumbents in the face of technological discontinuities: Automobile firms | |
CN115618128A (en) | Collaborative filtering recommendation system and method based on graph attention neural network | |
CN116128417A (en) | Method and system for identifying and early warning missing part risk of computer hardware inventory | |
CN117422527B (en) | Brand new commodity cold starting method based on user preference perception | |
Alam et al. | Business 4.0-A New Revolution | |
Sturm et al. | How two leading partners learn to tango: The case of IoT-based business model co-innovation between a retailer and an electronics supplier | |
CN110209379B (en) | Intelligent interactive software system and use method | |
WO2023185320A1 (en) | Cold start object recommendation method and apparatus, computer device and storage medium | |
CN115982373A (en) | Knowledge graph recommendation method combining multi-level interactive contrast learning | |
Slama et al. | The digital playbook: A practitioner’s guide to smart, connected products and solutions with AIoT | |
Sharafuddin et al. | The evolution of data analytics through the lens of business cases | |
KR20200045089A (en) | Consensus algorithm of blockchain, cryptographic technique and evolution network in blockchain | |
Stork | Big Data in Hollywood: The New Operating Logic of Media and Entertainment | |
CN117422526B (en) | Prompt-based user cross-domain cold start method | |
KR102570617B1 (en) | System, method and program for transacting power forecasting ai model based on blockchain | |
CN116777400B (en) | Engineering consultation information whole-flow management system and method based on deep learning | |
Soltanzadeh et al. | Challenges of ERP implementation: ERP as a Technology Transfer Project | |
Jenkins | The role of supply chain resource orchestration and supply chain knowledge in improving product launch performance in emerging markets |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |