Fusion sorting recommendation method and system based on multi-path recall

Abstract

The invention relates to a fusion sorting recommendation method and system based on multi-path recall, and belongs to the field of recommendation systems. Comprising the steps of obtaining commodity information and an interaction behavior between a user and a commodity; calculating a first recommendation score of each commodity through a content-based recall mode on the basis of the similarity between the commodity attribute features and the user portraits; calculating a second recommendation score of each commodity through a recall mode based on collaborative filtering based on the user behavior characteristics; based on click rates of different recall modes, performing weighted average on the first recommendation score and the second recommendation score, and calculating a weighted average score of each commodity; sorting the commodities in a descending order according to the weighted average score to obtain a commodity recommendation list; and preprocessing the commodity recommendation list to generate a final recommendation list. And the correlation, diversity and effectiveness of recommendation results are improved.

Description

Technical Field

[0001] The present invention belongs to the field of recommendation systems, and in particular relates to a fusion ranking recommendation method and system based on multi-way recall. Background Art

[0002] In the recommendation system, multi-way recall refers to the use of a variety of different recall strategies or algorithms to screen out a set of products that may meet the user's interests from the product pool. These recall strategies are based on different information sources or algorithm characteristics to provide a richer and more comprehensive set of candidate products. The fusion ranking recommendation method is a technology that comprehensively ranks the candidate product sets generated by multiple recall strategies in the recommendation system.

[0003] As the platform usage time increases, the user and product data on the platform will become very rich, that is, information overload will occur. Recommending a variety of products to the platform and users is of great significance to improving user experience, increasing platform retention and conversion, improving the relevance and diversity of recommendation results, and achieving the platform's profit goals.

[0004] However, traditional recommendation methods have problems such as cold start, sparse data, and relatively simple recommendations. They are unable to meet the needs of diversity in recommendation results and exposure of unpopular products, resulting in a decrease in user experience and the effectiveness of recommendation results, and are unable to better adapt to changes in user personalized needs. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the purpose of the invention is to provide a fusion ranking recommendation method and system based on multi-way recall, which can improve the relevance and diversity of recommendation results, increase user experience and platform retention and conversion, achieve the platform's profit goals, and improve the effectiveness of recommendation results.

[0006] In a first aspect of the present invention, a fusion ranking recommendation method based on multi-way recall is proposed, and the method comprises:

[0007] S1, obtain product information and the interaction between users and products;

[0008] S2, based on the similarity between product attributes and user portraits, the first recommendation score of each product is calculated through content-based recall;

[0009] S3, based on user behavior characteristics, calculate the second recommendation score of each product through a recall method based on collaborative filtering;

[0010] S4, based on the click rates of different recall methods, weighted average the first recommendation score and the second recommendation score to calculate the weighted average score of each product;

[0011] S5, sorting the products in descending order according to the weighted average score to obtain a product recommendation list;

[0012] S6, preprocessing the product recommendation list to generate a final recommendation list.

[0013] Furthermore, the commodity attribute features include tag attribute features, subject attribute features, business domain attribute features and application scenario attribute features.

[0014] Furthermore, S2 specifically includes:

[0015] S201, build user portraits based on the interaction between users and products;

[0016] S202, calculate the Jaccard similarity between the user and the product as the first recommendation score of each product:

[0017]

[0018] Among them, js ij represents the Jaccard similarity between user i and product j, s1 j represents the first recommendation score of recommended product j, f i represents the feature set of user i’s portrait, f j represents the feature set of product j, ∪ represents the union operator, ∩ represents the intersection operator, and || represents the set size.

[0019] Furthermore, user behavior characteristics include rating behavior characteristics, click behavior characteristics, and subscription behavior characteristics.

[0020] Furthermore, S3 specifically includes:

[0021] S301, constructing a rating matrix R based on the rating behavior characteristics;

[0022] S302, decomposing the rating matrix into a user matrix P and a product matrix Q;

[0023] S303, based on the user matrix and the product matrix, calculate the actual rating of each product of the user:

[0024]

[0025] Among them, r ij represents the actual rating of user i on product j, p i represents the potential feature vector of user i in the user matrix, q j represents the potential feature vector of product j in the product matrix, and T represents the transpose;

[0026] S304, constructing objective functions of the user matrix and the product matrix;

[0027] S305, optimizing the user matrix and the product matrix by using a stochastic gradient descent method according to the objective function;

[0028] S306, according to the optimized user matrix and product matrix, calculate the second recommendation score of each product through the inner product:

[0029]

[0030] Among them, s2 j and Both represent the second recommendation score of recommended product j.

[0031] Furthermore, S304 specifically includes:

[0032] S3041, construct the confidence matrix through nonlinear function:

[0033] c ij =1+α·log(1+cl ij )+β·su ij

[0034] Among them, c ij represents the confidence matrix, α represents the influencing factor of click behavior, log represents the logarithmic function, β represents the influencing factor of subscription behavior and β>α, cl ij represents the number of clicks of user i on product j, su ij Indicates whether user i has subscribed to product j, su ij =0 means no subscription, or su ij =1 means subscription;

[0035] S3042, by introducing a time decay factor, reflects the user's current preference:

[0036] γ(t,t ij )=e -τ(t-tij)

[0037] Among them, t represents the current time, t ij represents the time when the interaction occurs between user i and product j, e represents the base of the natural logarithm, and τ represents the time decay parameter used to control the rate of time decay;

[0038] S3043, combined with the time decay factor, update the confidence matrix:

[0039]

[0040] in, Indicates the time decay factor of the click behavior, Indicates the time decay factor of the subscription behavior; represents the time when a click occurs between user i and product j. Indicates the moment when the subscription behavior occurs between user i and product j;

[0041] S3044, introduce the updated confidence matrix into the error term of the explicit feedback score and construct the objective function:

[0042]

[0043] Among them, R represents the set of user-item pairs with ratings, min represents the minimum value, and e ij represents the adjustment of explicit scoring error according to the weight of implicit feedback, λ represents the regularization parameter, || || 2 Represents the square of the norm.

[0044] Furthermore, S305 specifically includes:

[0045] S3051, initialize the potential feature vectors of users and products respectively;

[0046] S3052, by setting the learning rate, controls the step size of each update;

[0047] S3053, by randomly selecting a small batch of users and corresponding product pairs, calculate the gradient of the objective function and update the parameters of the objective function:

[0048]

[0049] p i ←p i +η(e ij ·q j -λp i )

[0050] q j ←q j +η(e ij ·p i -λq j )

[0051] in, represents the symbol for partial derivative, η represents the learning rate, and ← represents the update parameter;

[0052] S3054, repeat the iteration until convergence or the set number of iterations is reached, and then stop the optimization.

[0053] Furthermore, S4 specifically includes:

[0054] S401, using the click rates of different recall methods as their respective weight factors:

[0055]

[0056] Among them, CTR1 represents the weight factor of recalling products based on content, and CTR2 represents the weight factor of recalling products based on collaborative filtering;

[0057] S402, calculating the weighted average score of the products recommended by the content recall method and the products recommended by the collaborative filtering recall method:

[0058]

[0059] Among them, s j ′ represents the weighted average score of product j, s1 j and s2 j They represent the first recommendation score and the second recommendation score generated for product j by the content recall method and the collaborative filtering recall method respectively, and I1 and I2 represent the product sets recalled by the content recall method and the collaborative filtering recall method respectively.

[0060] Furthermore, preprocessing includes filtering, deduplication, paging and complementation.

[0061] A second aspect of the present invention provides a fusion ranking recommendation system based on multi-way recall, comprising: a memory and a processor;

[0062] An application is stored in the memory, and the application is suitable for being executed by the processor to implement the multi-way recall-based fusion ranking recommendation method described in the first aspect.

[0063] The beneficial effects of the present invention are as follows:

[0064] In an embodiment of the present invention, based on commodity attribute characteristics and user behavior characteristics, respectively, two different recall methods are used to calculate the first recommendation score and the second recommendation score of each commodity. Combined with the click-through rates of different recall methods, the first recommendation score and the second recommendation score are weighted averaged to calculate the weighted average score of each commodity and generate a final recommendation list. This method can improve the relevance and diversity of recommendation results, increase user experience and platform retention and conversion, achieve the platform's profit goals, and improve the effectiveness of recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0066] Figure 1It is a flowchart of a fusion ranking recommendation method based on multi-way recall provided by an embodiment of the present invention;

[0067] Figure 2 It is a structural diagram of a fusion ranking recommendation system based on multi-way recall provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0069] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0070] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0071] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0072] The present invention proposes a fusion ranking recommendation method and system based on multi-way recall to solve the problems of cold start, sparse data and relatively single recommendation in traditional recommendation methods, which leads to a decrease in user experience and the effectiveness of recommendation results, and thus cannot better adapt to changes in user personalized needs.

[0073] Method Embodiment

[0074] Reference Figure 1 , showing a flow chart of a fusion ranking recommendation method based on multi-way recall provided in an embodiment of the present invention.

[0075] An embodiment of the present invention provides a fusion ranking recommendation method based on multi-way recall, the method comprising:

[0076] Specifically, the method includes steps S1 to S6.

[0077] S1, obtain product information and the interaction between users and products.

[0078] S2, based on the similarity between product attribute features and user portraits, the first recommendation score of each product is calculated through a content-based recall method.

[0079] Among them, the commodity attribute feature is the basic attribute of the commodity, which is used to describe the characteristics of the commodity. The user portrait is a set of user features built based on the user's historical behavior and preferences, which is used to describe the user's interests, needs and preferences, so as to facilitate the system to make personalized recommendations. The content-based recall method is a recommendation strategy. The first recommendation score refers to the matching score between each commodity and the user calculated based on the content recall method.

[0080] It should be noted that this step can more accurately find products similar to user preferences and recommend content of interest to users.

[0081] In a possible implementation, the commodity attribute features include tag attribute features, subject attribute features, business domain attribute features, and application scenario attribute features.

[0082] In a possible implementation, S2 specifically includes:

[0083] S201, build a user profile based on the interaction between the user and the product.

[0084] Specifically, user portraits can be constructed based on the user's interactive behaviors with products (such as ratings, clicks, and subscriptions), mapping the features of the interactive products to the user to form a user's interest portrait.

[0085] S202, calculate the Jaccard similarity between the user and the product as the first recommendation score of each product:

[0086]

[0087] Among them, js ij represents the Jaccard similarity between user i and product j, s1 j represents the first recommendation score of recommended product j, f i represents the feature set of user i’s portrait, f j represents the feature set of product j, ∪ represents the union operator, ∩ represents the intersection operator, and || represents the set size.

[0088] S3, based on the user behavior characteristics, the second recommendation score of each product is calculated through a recall method based on collaborative filtering.

[0089] The second recommendation score refers to the matching score between each product and the user calculated based on the collaborative filtering recall method.

[0090] It should be noted that the second recommendation score for each product is generated by a collaborative filtering method based on user behavior characteristics, which adds a personalized recommendation dimension based on user behavior to the recommendation system.

[0091] In a possible implementation, the user behavior characteristics include rating behavior characteristics, click behavior characteristics, and subscription behavior characteristics.

[0092] In a possible implementation, S3 specifically includes:

[0093] S301, constructing a rating matrix R based on the rating behavior characteristics.

[0094] S302, decomposing the rating matrix into a user matrix P and a product matrix Q.

[0095] S303, based on the user matrix and the product matrix, calculate the actual rating of each product of the user:

[0096]

[0097] Among them, r ij represents the actual rating of user i on product j, p i represents the potential feature vector of user i in the user matrix, q j represents the potential feature vector of product j in the product matrix, and T represents the transpose.

[0098] S304, constructing objective functions of the user matrix and the product matrix.

[0099] In a possible implementation manner, S304 specifically includes:

[0100] S3041, construct the confidence matrix through nonlinear function:

[0101] c ij =1+α·log(1+cl ij )+β·su ij

[0102] Among them, c ij represents the confidence matrix, α represents the influencing factor of click behavior, log represents the logarithmic function, β represents the influencing factor of subscription behavior and β>α, cl ij represents the number of clicks of user i on product j, su ij Indicates whether user i has subscribed to product j, su ij =0 means no subscription, or su ij =1 means subscription.

[0103] S3042, by introducing a time decay factor, reflects the user's current preference:

[0104]

[0105] Among them, t represents the current time, t ij represents the time when the interaction occurs between user i and product j, e represents the base of the natural logarithm, and τ represents the time decay parameter used to control the rate of time decay.

[0106] S3043, combined with the time decay factor, update the confidence matrix:

[0107]

[0108] in, Indicates the time decay factor of the click behavior, Indicates the time decay factor of the subscription behavior; represents the time when a click occurs between user i and product j. Indicates the time when a subscription occurs between user i and product j.

[0109] S3044, introduce the updated confidence matrix into the error term of the explicit feedback score and construct the objective function:

[0110]

[0111] Among them, R represents the set of user-item pairs with ratings, min represents the minimum value, and e ij represents the adjustment of explicit scoring error according to the weight of implicit feedback, λ represents the regularization parameter, || || 2 Represents the square of the norm.

[0112] S305, optimizing the user matrix and the product matrix according to the objective function by using the stochastic gradient descent method.

[0113] In a possible implementation manner, S305 specifically includes:

[0114] S3051, initialize the potential feature vectors of users and products respectively.

[0115] S3052, by setting the learning rate, controls the step size of each update.

[0116] S3053, by randomly selecting a small batch of users and corresponding product pairs, calculate the gradient of the objective function and update the parameters of the objective function:

[0117]

[0118] p i ←p i +η(e ij ·q j -λp i )

[0119] q j ←q j +η(e ij ·p i -λq j )

[0120] in, Represents the symbol for partial derivative, η represents the learning rate, and ← represents the update parameter.

[0121] S3054, repeat the iteration until convergence or the set number of iterations is reached, and then stop the optimization.

[0122] S306, according to the optimized user matrix and product matrix, calculate the second recommendation score of each product through the inner product:

[0123]

[0124] Among them, s2 j and Both represent the second recommendation score of recommended product j.

[0125] S4, based on the click rates of different recall methods, weighted average the first recommendation score and the second recommendation score to calculate the weighted average score of each product.

[0126] It should be noted that by taking a weighted average of the click-through rates of different recall methods, the recommendation system can combine the advantages of multiple recall strategies to provide more personalized recommendations.

[0127] In a possible implementation, S4 specifically includes:

[0128] S401, using the click rates of different recall methods as their respective weight factors:

[0129]

[0130] Among them, CTR1 represents the weight factor of recalling products based on content, and CTR2 represents the weight factor of recalling products based on collaborative filtering.

[0131] S402, calculating the weighted average score of the products recommended by the content recall method and the products recommended by the collaborative filtering recall method:

[0132]

[0133] Among them, s j ′ represents the weighted average score of product j, s1 j and s2 j They represent the first recommendation score and the second recommendation score generated for product j by the content recall method and the collaborative filtering recall method respectively, and I1 and I2 represent the product sets recalled by the content recall method and the collaborative filtering recall method respectively.

[0134] S5, sorting the products in descending order according to the weighted average score to obtain a product recommendation list.

[0135] It should be noted that this process effectively improves the relevance and accuracy of recommendation results, ensures that users first see the products they may be most interested in, and increases the click-through rate and user satisfaction of the recommendation list.

[0136] S6, preprocessing the product recommendation list to generate a final recommendation list.

[0137] It should be noted that by performing pre-processing steps such as filtering, deduplication, paging and supplementing the recommendation list, the quality of the recommendation list is further improved, making the user experience smoother and richer.

[0138] In a possible implementation, preprocessing includes filtering, deduplication, paging, and supplementation.

[0139] The beneficial effects of the present invention are as follows:

[0140] In an embodiment of the present invention, based on commodity attribute characteristics and user behavior characteristics, respectively, two different recall methods are used to calculate the first recommendation score and the second recommendation score of each commodity. Combined with the click-through rates of different recall methods, the first recommendation score and the second recommendation score are weighted averaged to calculate the weighted average score of each commodity and generate a final recommendation list. This method can improve the relevance and diversity of recommendation results, increase user experience and platform retention and conversion, achieve the platform's profit goals, and improve the effectiveness of recommendation results.

[0141] System Example

[0142] Reference Manual Attached Figure 2 , showing a structural diagram of a fusion ranking recommendation system based on multi-way recall provided in an embodiment of the present invention.

[0143] The present invention proposes a fusion ranking recommendation system 30 based on multi-way recall, including: a memory 303 and a processor 301.

[0144] The memory 303 stores an application program, and the application program is suitable for being executed by the processor 301 to implement the petroleum equipment fault prediction method based on big data of the method embodiment.

[0145] The oil equipment fault prediction system 30 based on big data includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, through a bus 302.

[0146] The structure of the multi-way recall-based fusion ranking recommendation system 30 does not constitute a limitation on the embodiment of the present invention.

[0147] Processor 301 may be a CPU, a general purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0148] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI bus or an EISA bus, etc. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0149] The memory 303 can be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0150] Computer Readable Storage Medium Embodiments

[0151] The present invention provides a computer-readable storage medium having a computer program stored thereon, which can be loaded and executed by a processor to implement the multi-way recall-based fusion ranking recommendation method of the first aspect.

[0152] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A fusion ranking recommendation method based on multi-way recall, characterized in that: The fusion ranking recommendation method based on multi-way recall includes: S1, obtain product information and the interaction between users and products; S2, based on the similarity between product attributes and user portraits, the first recommendation score of each product is calculated through content-based recall; S3, based on user behavior characteristics, calculate the second recommendation score of each product through a recall method based on collaborative filtering; S4, based on the click rates of different recall methods, taking a weighted average of the first recommendation score and the second recommendation score to calculate a weighted average score of each product; S5, sorting the products in descending order according to the weighted average score to obtain a product recommendation list; S6: pre-process the product recommendation list to generate a final recommendation list.

2. The fusion ranking recommendation method based on multi-way recall according to claim 1 is characterized in that: The commodity attribute features include tag attribute features, subject attribute features, business domain attribute features and application scenario attribute features.

3. The fusion ranking recommendation method based on multi-way recall according to claim 1 is characterized in that: The S2 specifically includes: S201, constructing a user profile based on the interaction behavior between the user and the product; S202, calculate the Jaccard similarity between the user and the product as the first recommendation score of each product: Among them, js ij represents the Jaccard similarity between user i and product j, s1 j represents the first recommendation score of recommended product j, f i represents the feature set of user i’s portrait, f j represents the feature set of product j, ∪ represents the union operator, ∩ represents the intersection operator, and || represents the set size.

4. The fusion ranking recommendation method based on multi-way recall according to claim 1 is characterized in that: The user behavior characteristics include rating behavior characteristics, click behavior characteristics and subscription behavior characteristics.

5. The fusion ranking recommendation method based on multi-way recall according to claim 4 is characterized in that: The S3 specifically includes: S301, constructing a scoring matrix R based on the scoring behavior characteristics; S302, decomposing the rating matrix into a user matrix P and a product matrix Q; S303, calculating the actual rating of each product of the user according to the user matrix and the product matrix: Among them, r ij represents the actual rating of user i on product j, p i represents the potential feature vector of user i in the user matrix, q j represents the potential feature vector of product j in the product matrix, and T represents the transpose; S304, constructing objective functions of the user matrix and the product matrix; S305, optimizing the user matrix and the product matrix according to the objective function by using a stochastic gradient descent method; S306, according to the optimized user matrix and product matrix, calculate the second recommendation score of each product through the inner product: Among them, s2 j and Both represent the second recommendation score of recommended product j.

6. The fusion ranking recommendation method based on multi-way recall according to claim 5 is characterized in that: The S304 specifically includes: S3041, construct the confidence matrix through nonlinear function: c ij =1+α·log(1+cl ij )+β·su ij Among them, c ij represents the confidence matrix, α represents the influencing factor of click behavior, log represents the logarithmic function, β represents the influencing factor of subscription behavior and β>α, cl ij represents the number of clicks of user i on product j, su ij Indicates whether user i has subscribed to product j, su ij =0 means no subscription, or su ij =1 means subscription; S3042, by introducing a time decay factor, reflecting the current preference of the user: Among them, t represents the current time, t ij represents the time when the interaction occurs between user i and product j, e represents the base of the natural logarithm, and τ represents the time decay parameter used to control the rate of time decay; S3043: Update the confidence matrix in combination with the time decay factor: in, Indicates the time decay factor of the click behavior, Indicates the time decay factor of the subscription behavior; represents the time when a click occurs between user i and product j. Indicates the moment when the subscription behavior occurs between user i and product j; S3044, introducing the updated confidence matrix into the error term of the explicit feedback score to construct the objective function: Among them, R represents the set of user-item pairs with ratings, min represents the minimum value, and e ij represents the adjustment of explicit scoring error according to the weight of implicit feedback, λ represents the regularization parameter, || || 2 Represents the square of the norm.

7. The fusion ranking recommendation method based on multi-way recall according to claim 5 is characterized in that: The S305 specifically includes: S3051, respectively initializing the potential feature vectors of the user and the product; S3052, by setting the learning rate, controls the step size of each update; S3053, randomly selecting a small batch of users and corresponding product pairs, calculating the gradient of the objective function, and updating the parameters of the objective function: p i ←p i +η(e ij ·q j -λp i ) q j ←q j +η(e ij ·p i -λq j ) in, represents the symbol for partial derivative, η represents the learning rate, and ← represents the update parameter; S3054, repeat the iteration until convergence or the set number of iterations is reached, and then stop the optimization.

8. The fusion ranking recommendation method based on multi-way recall according to claim 1 is characterized in that: The S4 specifically includes: S401, using the click rates of different recall methods as their respective weight factors: Among them, CTR1 represents the weight factor of recalling products based on content, and CTR2 represents the weight factor of recalling products based on collaborative filtering; S402, calculating the weighted average score of the products recommended by the content recall method and the products recommended by the collaborative filtering recall method: Among them, s j ′ represents the weighted average score of product j, s1 j and s2 j They represent the first recommendation score and the second recommendation score generated for product j by the content recall method and the collaborative filtering recall method respectively, and I1 and I2 represent the product sets recalled by the content recall method and the collaborative filtering recall method respectively.

9. The fusion ranking recommendation method based on multi-way recall according to claim 1 is characterized in that: The preprocessing includes filtering, deduplication, paging and supplementation.

10. A fusion ranking recommendation system based on multi-way recall, characterized in that: include: Memory and processor; An application is stored in the memory, and the application is suitable for being executed by the processor to implement the multi-way recall-based fusion ranking recommendation method according to any one of claims 1 to 9.

Images