Product recommendation method and device, computer readable storage medium, terminal

Abstract

A product recommendation method and device, a computer readable storage medium and a terminal, the method comprising: obtaining one or more historical product viewing information of a user within a preset time length; using a preset kernel function, forming a cumulative distribution function for each historical purchase influence feature, and using the preset kernel function, forming a result cumulative distribution function for the historical purchase result; selecting one or more cumulative distribution functions and the result cumulative distribution function to construct a Gaussian coupled joint cumulative distribution function, and constructing a Gaussian coupled joint density function; constructing an objective function, maximizing the objective function to determine the probability of the target product being purchased with the optimal fitting effect; and determining whether to recommend the target product to the user. The present application has the opportunity to determine whether to recommend to the user according to the viewing willingness of the user to the target product.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a product recommendation method and device, a computer readable storage medium and a terminal. BACKGROUND

[0002] With the rapid development of Internet technology and the rapid popularization of intelligent terminals, recommending products to users through a network platform has become a common product recommendation method. For example, a user can view various products recommended by a business party on his / her mobile device through a product recommendation page.

[0003] In the prior art, a single dimension is usually used for sorting, and then the products are recommended to the user. Taking the purchase of goods as an example, in response to the user opening a product recommendation page, it is determined that the user has issued a product recommendation request, and various products recommended to the user are automatically displayed in a shopping page. For example, the products can be recommended to the user according to the dimension of product discount ranking, or the length of time from the discount to expiration.

[0004] However, the above scheme does not consider the matching between the product and the user, and often recommends a large number of products that the user is not interested in and has a low purchase willingness to the user, resulting in that the user lacks the willingness to view the recommended products, and even misses the product that should have a purchase willingness.

[0005] Therefore, there is an urgent need for a product recommendation method that can pre-judge a target product to be recommended, so as to have the opportunity to decide whether to recommend the target product to the user according to the user's viewing willingness. SUMMARY

[0006] The present application solves the technical problem of providing a product recommendation method and device, a computer readable storage medium and a terminal, which have the opportunity to decide whether to recommend a target product to a user according to the user's viewing willingness, effectively improve the recommendation-viewing conversion rate, and even improve the probability of being purchased.

[0007] To solve the above technical problems, the embodiment of the present application provides a product recommendation method, comprising: obtaining one or more historical product viewing information of a user within a preset time length, wherein the historical product viewing information comprises a historical purchase result and one or more historical purchase influence characteristics; using a preset kernel function, forming a cumulative distribution function for each historical purchase influence characteristic, and using the preset kernel function, forming a result cumulative distribution function for the historical purchase result; selecting one or more cumulative distribution functions and the result cumulative distribution function, constructing a Gaussian coupled joint cumulative distribution function, and constructing a Gaussian coupled joint density function; constructing a target function according to a purchase influence characteristic of a target product and the Gaussian coupled joint density function, maximizing the target function to determine the probability that the target product with the optimal fitting effect is purchased; and determining whether to recommend the target product to the user according to the probability that the target product with the optimal fitting effect is purchased.

[0008] Optionally, the historical product to which the historical product viewing information belongs and the target product are iterative products, the target product is an iterative product selected in response to a product recommendation request of the user after receiving the product recommendation request of the user; and determining whether to recommend the target product to the user according to the probability that the target product with the optimal fitting effect is purchased comprises: sorting the probabilities that a plurality of target products are purchased; and determining whether to recommend the target product corresponding to each probability to the user according to the sorting result.

[0009] Optionally, the historical purchase influence characteristic is selected from one or more of the following: product attribute information of a historical product, individual characteristic information of the user, and a time factor of viewing the historical product.

[0010] Optionally, the forming of the cumulative distribution function comprises: using the preset kernel function to form a smooth density function for each historical purchase influence characteristic; and forming the cumulative distribution function according to the smooth density function.

[0011] Optionally, the historical purchase influence characteristic is product attribute information of a historical product, and the preset kernel function is a Gaussian kernel function; the smooth density function is formed using the following formula:

[0012]

[0013] And / or, the cumulative distribution function is formed according to the smooth density function using the following formula:

[0014]

[0015] wherein, is used to represent product attribute information d viewed by user i, is used to represent a smooth density function based on , and the function is a Gaussian function, is used to represent product attribute information d of product j viewed by user i at time t, is used to represent a cumulative distribution function based on , h is a smoothing parameter, i.e., a bandwidth; T is used to represent the preset time length, and M it is used to represent the number of products viewed by user i within the preset time length.

[0016] Optionally, the historical purchase influence feature is individual feature information of the user, and the preset kernel function is a Gaussian kernel function; the following formula is used to form the smooth density function:

[0017]

[0018] and / or the following formula is used to form the cumulative distribution function according to the smooth density function:

[0019]

[0020] , wherein, is used to represent individual feature information c of user i, is used to represent a smooth density function based on , and the function is a Gaussian function, is used to represent individual feature information c of user i when viewing product j at time t, is used to represent a cumulative distribution function based on , h is a smoothing parameter, i.e., a bandwidth; T is used to represent the preset time length, and M it is used to represent the number of products viewed by user i within the preset time length.

[0021] Optionally, the historical purchase influence feature is a time factor of viewing the historical product, and the preset kernel function is a Gaussian kernel function; the following formula is used to form the smooth density function:

[0022]

[0023] and / or the following formula is used to form the cumulative distribution function according to the smooth density function:

[0024]

[0025] , wherein, li is used to represent a time factor of user i, is used to represent a smooth density function based on li , and the function is a Gaussian function, lijta time factor for indicating a product j viewed by a user i at a time t, a result cumulative distribution function based on y li , h is a smoothing parameter, i.e., a bandwidth; T is used to indicate the preset time length, M it is used to indicate the number of products viewed by the user i within the preset time length.

[0026] Optionally, the forming the result cumulative distribution function comprises: adopting the preset kernel function, forming a result smoothing density function for the historical purchase result; and forming the result cumulative distribution function according to the result smoothing density function.

[0027] Optionally, the preset kernel function is a Gaussian kernel function; and the result smoothing density function is formed by using the following formula:

[0028]

[0029] and / or the result cumulative distribution function is formed according to the result smoothing density function by using the following formula:

[0030]

[0031] wherein, y i is used to indicate that the historical purchase result of the user i is purchase or non-purchase, is used to indicate a result smoothing density function based on y i , and the function is a Gaussian function, y ijt is used to indicate that the historical purchase result of the product j viewed by the user i at the time t is purchase or non-purchase, is used to indicate a result cumulative distribution function based on y i , h is a smoothing parameter, i.e., a bandwidth; T is used to indicate the preset time length, M it is used to indicate the number of products viewed by the user i within the preset time length.

[0032] Optionally, the preset kernel function is selected from one or more of the following: a Gaussian kernel function, an exponential kernel function, a polynomial kernel function and a box kernel function.

[0033] Optionally, the cumulative distribution function comprises a cumulative distribution function of product attribute information of a historical product, individual feature information of the user and a time factor for viewing the historical product; and one or more of the cumulative distribution functions and the result cumulative distribution function are selected to construct a Gaussian coupled joint cumulative distribution function by using the following formula:

[0034]

[0035] wherein, R(i) represents the Gaussian coupled joint cumulative distribution function, and R(i) represents the variance-covariance matrix of the Gaussian distribution function. Φ R(i) The joint cumulative distribution function used to represent a multivariate Gaussian distribution with zero mean and variance-covariance matrix R(i), Used to represent probability integral transformation functions; y i Used to represent historical purchase results Used to represent the cumulative distribution function of the result. A probability integral transformation function used to represent the cumulative distribution function of the outcome; d is used to represent the various product attribute information viewed by user i. Used to represent the cumulative distribution function based on product attribute information d. Used to indicate based on The inverse cumulative distribution function, Used to indicate based on The probability integral transformation function of the cumulative distribution function; c is used to represent the individual characteristic information of user i. Used to represent the cumulative distribution function based on individual characteristic information c. Used to indicate based on The inverse cumulative distribution function, Used to indicate based on The probability integral transformation function of the cumulative distribution function; Used to represent the time factor of user i Used to represent the cumulative distribution function based on time factors. Used to indicate based on The inverse cumulative distribution function, Used to indicate based on The probability integral transformation function of the cumulative distribution function.

[0036] Optionally, the variance-covariance matrix R(i) of the Gaussian distribution function can be determined using the following formula:

[0037] R(i)=Λ(i) -1 / 2 Σ(i)Λ(i) -1 / 2

[0038]

[0039] in, Used to represent The observed value, M it The term Λ(i) represents the number of products viewed by user i within the preset time period; Λ(i) represents a diagonal matrix containing the diagonal elements of Σ(i), and Σ(i) represents the covariance matrix.

[0040] Optionally, the Gaussian coupling joint density function is constructed by using the following formula:

[0041]

[0042] wherein, is used to represent the Gaussian coupling joint density function, c(x) is used to represent the coupling function, y i is used to represent the historical purchase result, is used to represent the result cumulative distribution function, is used to represent each product attribute information d viewed by the user i, is used to represent the cumulative distribution function based on the product attribute information d, is used to represent the individual feature information c of the user i, is used to represent the cumulative distribution function based on the individual feature information c, is used to represent the time factor of the user i, is used to represent the cumulative distribution function based on the time factor, is used to represent the result smoothing density function based on y i , is used to represent the smoothing density function based on , is used to represent the smoothing density function based on , is used to represent the smoothing density function based on ω li .

[0043] Optionally, the target function is constructed by using the following formula:

[0044]

[0045] wherein,

[0046] is used to represent the target product, the y' i is used to represent the probability that the target product is purchased, the is used to represent the product attribute information of the target product, the is used to represent the individual feature information of the user i, and the ω i is used to represent the time factor in which the user i views the target product.

[0047] Optionally, maximizing the target function to determine the probability that the target product is purchased with the optimal fitting effect comprises: selecting multiple values in [0, 1] for y' i , and respectively calculating is determined by using the following formula: the y' i, the probability that the target product is purchased as the target product with the optimal fitting effect:

[0048]

[0049] wherein y' i = Φ -1 (R(i)).

[0050] To solve the above technical problems, an embodiment of the present application provides a product recommendation device, comprising: an acquisition module configured to acquire one or more historical product viewing information of a user within a preset time length, wherein the historical product viewing information comprises a historical purchase result and one or more historical purchase influence features; a first function determination module configured to form a cumulative distribution function for each historical purchase influence feature by using a preset kernel function, and form a result cumulative distribution function for the historical purchase result by using the preset kernel function; a second function determination module configured to select one or more of the cumulative distribution functions and the result cumulative distribution function, construct a Gaussian coupled joint cumulative distribution function, and construct a Gaussian coupled joint density function; a fitting module configured to construct a target function according to a purchase influence feature of a target product and the Gaussian coupled joint density function, maximize the target function to determine the probability that the target product is purchased with the optimal fitting effect; and a recommendation determination module configured to determine whether to recommend the target product to the user according to the probability that the target product is purchased with the optimal fitting effect.

[0051] To solve the above technical problems, an embodiment of the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is run by a processor to perform the steps of the product recommendation method.

[0052] To solve the above technical problems, an embodiment of the present application provides a terminal comprising a memory and a processor, wherein the memory has a computer program stored thereon, the computer program is capable of being run on the processor, and the processor is run to perform the steps of the product recommendation method.

[0053] Compared with the prior art, the technical scheme of the embodiment of the present application has the following beneficial effects:

[0054] In the embodiment of the present application, by obtaining the historical product viewing information of the user, and forming the cumulative distribution function and the result cumulative distribution function based on the historical viewing information, and then constructing the Gaussian coupling joint cumulative distribution function and the Gaussian coupling joint density function, and constructing the target function according to the purchase influence characteristics of the target product, the probability of the target product being purchased with the optimal fitting effect is determined, and the willingness of the user to the target product is predicted, so that through the big data processing technology, after judging the target product to be recommended in advance, there is an opportunity to decide whether to recommend the target product to the user according to the viewing willingness of the user to the target product, the recommendation-viewing conversion rate is effectively improved, and even the probability of being purchased is improved.

[0055] Further, the historical product and the target product are iterative products, for example, the product is automatically updated and displayed, and if the display is too late, the opportunity of being viewed by the user may be lost, in such a case, the probabilities of the target products being purchased are sorted, and whether to recommend the target products corresponding to the probabilities to the user is determined according to the sorting result, which can recommend the products more in line with the willingness of the user to the user, and is beneficial to improve the recommendation-viewing conversion rate, and even improve the probability of being purchased.

[0056] Further, the historical purchase influence characteristics are selected from the product attribute information of the historical product, the individual characteristic information of the user, and the time factor of viewing the historical product, so that the characteristics of the historical product and the characteristics of the user can be used to determine the selection condition of the historical product, so as to more effectively judge the potential purchase habit of the user, form a target function more customized and more accurate for the user, and further improve the accuracy of judging the purchase willingness of the user.

[0057] Further, by using the Gaussian kernel function to form the smooth density function and then forming the cumulative distribution function, the effectiveness of the obtained cumulative distribution function can be improved by using the symmetry and smoothness of the Gaussian kernel function, and the accuracy of the subsequent formed target function is further improved.

[0058] Further, the variance-covariance matrix R(i) of the Gaussian distribution function is determined by using Σ(i) and the diagonal matrix containing the diagonal elements of Σ(i), so that the variance-covariance matrix can be more effectively and accurately determined, and then the Gaussian coupling joint cumulative distribution function is constructed based on the variance-covariance matrix R(i) in the subsequent step.

[0059] Further, by constructing a proper target function, the target function can be maximized to determine the probability of the target product being purchased with the optimal fitting effect, and the accuracy of judging the purchase willingness of the user is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1is a flowchart of a product recommendation method in an embodiment of the present application;

[0061] Figure 2 is Figure 1 is a flowchart of a specific implementation of step S15 in the method;

[0062] Figure 3 is a structural schematic diagram of a product recommendation device in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In the prior art, a background server can recommend products to a user and provide a purchase channel through a network platform, however, the existing product recommendation method is usually based on a single dimension for sorting, and the matching between the product and the user is not considered, and a large number of products that the user is not interested in and has low purchase willingness are often recommended to the user, resulting in that the user lacks the willingness to view the recommended products, and even misses the product that should have the purchase willingness.

[0064] The inventors of the present application have found through research that in the existing product recommendation scheme, the target product to be recommended is usually sorted according to the product recommendation request after receiving the product recommendation request input by the user, and is recommended to the user.

[0065] However, various problems may exist in the recommendation process, for example, the first recommended product may not be as suitable for the user as the later recommended product, and the user is difficult to make a decision while being uncertain about the future choice; for example, each target product may have timeliness, such as expiration of discount, limited quantity, etc. Specifically, taking an online dating website as an example, the user will receive one or more recommended "matching files", and once the user slides over (i.e. gives up), the user may not be able to return to choose the same file; taking a daily transaction website, a flash sale website, an online streaming media service sales website as an example, there is a time-limited promotion activity, and the user needs to decide whether to purchase the product (or service) immediately or wait for the next product (or service), wherein the abandoned product (or service) may no longer be displayed due to expiration, resulting in that the user misses the matching product, and the user experience is poor.

[0066] The inventors of the present application have also found through research that the existing product recommendation scheme only focuses on the target product to be recommended itself, sorts using a single dimension of the target product, and does not consider the individual characteristics of the user itself, nor the potential purchase habits of the user, resulting in that customization is insufficient in the recommendation process, and effective recommendation is difficult to be made.

[0067] In the embodiment of the present application, by acquiring the historical product viewing information of the user, and forming a cumulative distribution function and a result cumulative distribution function based on the historical viewing information, a Gaussian coupling joint cumulative distribution function and a Gaussian coupling joint density function are further constructed, and then a target function is constructed according to the purchase influence characteristics of the target product, the probability of the target product being purchased with the optimal fitting effect is determined, the willingness of the user to the target product is predicted, so that through the big data processing technology, after judging the target product to be recommended in advance, there is an opportunity to decide whether to recommend the target product to the user according to the viewing willingness of the user to the target product, the recommendation-viewing conversion rate is effectively improved, and even the probability of being purchased is improved.

[0068] In order to make the above-mentioned purpose, characteristics and beneficial effects of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0069] Reference Figure 1 , Figure 1 is a flowchart of a product recommendation method in the embodiment of the present application. The product recommendation method can include steps S11 to S15:

[0070] Step S11: acquiring one or more historical product viewing information of a user within a preset time length, the historical product viewing information including historical purchase results and one or more historical purchase influence characteristics;

[0071] Step S12: forming a cumulative distribution function for each historical purchase influence characteristic by using a preset kernel function, and forming a result cumulative distribution function for the historical purchase results by using the preset kernel function;

[0072] Step S13: selecting one or more cumulative distribution functions and the result cumulative distribution function to construct a Gaussian coupling joint cumulative distribution function, and to construct a Gaussian coupling joint density function;

[0073] Step S14: constructing a target function according to the purchase influence characteristics of the target product and the Gaussian coupling joint density function, and maximizing the target function to determine the probability of the target product being purchased with the optimal fitting effect;

[0074] Step S15: determining whether to recommend the target product to the user according to the probability of the target product being purchased with the optimal fitting effect.

[0075] It can be understood that in the specific implementation, the method can be realized in the form of a software program running in a processor integrated in a chip or a chip module.

[0076] In the implementation of step S11, for the user, a certain number of products can be viewed in each time period, and the related information of the viewed historical products can be analyzed as historical data.

[0077] It can be understood that the preset time period should not be set too short, otherwise the amount of information obtained will be too small, and it will be difficult to determine the viewing influence factors or purchase influence factors of the user; the preset time period should not be set too long, otherwise the calculation amount will be too large, affecting the determination accuracy. As a non-limiting example, the preset time period can be selected from 1 hour to 1 month, for example, 1 day.

[0078] In a specific application, the user i can view M it historical products one by one at time t (for example, the tth day), 1≤j≤M it , j can be used to represent the order in which the historical products are viewed within the preset time period, and each historical product has historical product viewing information.

[0079] The historical product viewing information can include historical purchase results and one or more historical purchase influence features. Specifically, the historical purchase results can include purchased and not purchased.

[0080] In one specific embodiment of the embodiment of the application, the historical products involved in the historical product viewing information can be historical products that the user has browsed the product details, for example, the user has once clicked on the link page of a product and entered the detail page of the product.

[0081] Specifically, the user specifically browses the product among a large number of recommended products, which can indicate that the user has a certain interest in the product. Taking such products as historical products and obtaining historical product viewing information can make the subsequent recommendation step more targeted.

[0082] In another specific embodiment of the embodiment of the application, the historical products involved in the historical product viewing information can be products recommended to the user, whether the user has ever clicked on the link page of the product or not, all of which are historical products and historical product viewing information is obtained.

[0083] Specifically, taking the products recommended to the user in the past as historical products can effectively expand the breadth and width of collecting user information, so as to determine the part that the user is really interested in among more historical products, which can make the subsequent recommendation step more representative.

[0084] Further, the historical purchase influence features can be selected from one or more of the following: product attribute information of the historical product, individual feature information of the user, and time factor of viewing the historical product.

[0085] The product attribute information of the historical product can be used to represent the objective attribute of the historical product.

[0086] The individual feature information of the user can be used to represent the personal selection tendency of the user.

[0087] The time factor of viewing the historical product can be used to represent the time effect (also known as seasonality) which will affect the selection of the product by the user, for example, during Christmas, the user will more frequently view and purchase products with Christmas elements.

[0088] In the embodiments of the present application, the characteristics of the historical product and the characteristics of the user can be used to determine the selection of the historical product, so as to more effectively judge the potential purchase habit of the user, form a target function which is more customized and more accurate for the user, and further improve the accuracy of judging the purchase willingness of the user.

[0089] In the specific implementation of step S12, a preset kernel function is used to form a cumulative distribution function for each historical purchase influencing feature, and the preset kernel function is used to form a result cumulative distribution function for the historical purchase result.

[0090] Further, the preset kernel function can be selected from one or more of the following: a Gaussian kernel function, an exponential kernel function, a polynomial kernel function, and a box kernel function.

[0091] As a non-limiting specific embodiment, the preset kernel function can be a Gaussian kernel function.

[0092] In the embodiments of the present application, by using the Gaussian kernel function to form a smooth density function and then forming a cumulative distribution function, the effectiveness of the obtained cumulative distribution function can be improved by using the symmetry and smoothness of the Gaussian kernel function, and the accuracy of the subsequently formed target function can be further improved.

[0093] Further, the step of forming a cumulative distribution function can include: using the preset kernel function to form a smooth density function for each historical purchase influencing feature; and forming the cumulative distribution function according to the smooth density function.

[0094] The following can be described for the historical purchase influencing feature being the product attribute information of the historical product, the individual feature information of the user, or the time factor of viewing the historical product.

[0095] Further, the historical purchase influencing feature is the product attribute information of the historical product, and the preset kernel function is a Gaussian kernel function; the smooth density function can be formed using the following formula:

[0096]

[0097] and / or,

[0098] The cumulative distribution function can be formed according to the smooth density function by using the following formula:

[0099]

[0100] wherein, is used to represent the product attribute information d viewed by the user i, is used to represent the smooth density function based on , and the function is a Gaussian function, is used to represent the product attribute information d of the product j viewed by the user i at the time t, is used to represent the cumulative distribution function based on , h is a smooth parameter, i.e., a bandwidth; T is used to represent the preset time length, and M it is used to represent the number of products viewed by the user i within the preset time length.

[0101] Further, the historical purchase influence feature is the individual feature information of the user, and the preset kernel function is a Gaussian kernel function; the smooth density function can be formed by using the following formula:

[0102]

[0103] and / or,

[0104] The cumulative distribution function can be formed according to the smooth density function by using the following formula:

[0105]

[0106] wherein, is used to represent the individual feature information c of the user i, is used to represent the smooth density function based on , and the function is a Gaussian function, is used to represent the individual feature information c of the user i when viewing the product j at the time t, is used to represent the cumulative distribution function based on , h is a smooth parameter, i.e., a bandwidth; T is used to represent the preset time length, and M it is used to represent the number of products viewed by the user i within the preset time length.

[0107] Further, the historical purchase influence feature is the time factor of viewing the historical product, and the preset kernel function is a Gaussian kernel function; the smooth density function can be formed by using the following formula:

[0108]

[0109] and / or,

[0110] The cumulative distribution function is formed according to the smooth density function by using the following formula:

[0111]

[0112] wherein ω li is used to represent the time factor of the user i, is used to represent the cumulative distribution function based on ω li , and the function is a Gaussian function, ω lijt is used to represent the time factor of the product j viewed by the user i at the time t, is used to represent the cumulative distribution function based on ω li , and h is a smooth parameter, i.e., a bandwidth; T is used to represent the preset time length, M it is used to represent the number of products viewed by the user i within the preset time length.

[0113] Further, the step of forming the result cumulative distribution function can include: forming a result smooth density function for the historical purchase result by using the preset kernel function; and forming the result cumulative distribution function according to the result smooth density function.

[0114] Further, the preset kernel function is a Gaussian kernel function; and the result smooth density function is formed by using the following formula:

[0115]

[0116] and / or,

[0117] The result cumulative distribution function is formed according to the result smooth density function by using the following formula:

[0118]

[0119] wherein y i is used to represent whether the historical purchase result of the user i is purchase or non-purchase, is used to represent the result smooth density function based on y i , and the function is a Gaussian function, y ijt is used to represent whether the historical purchase result of the product j viewed by the user i at the time t is purchase or non-purchase, is used to represent the result cumulative distribution function based on y i , and h is a smooth parameter, i.e., a bandwidth; T is used to represent the preset time length, M itThis indicates the number of products viewed by user i within the preset time period.

[0120] Continue to refer to Figure 1 In the specific implementation of step S13, a Gaussian coupled joint cumulative distribution function and a Gaussian coupled joint density function can be constructed.

[0121] When selecting the cumulative distribution function, one or more of the following can be selected: product attribute information of historical products, individual characteristic information of the user, and time factor of viewing the historical products.

[0122] Furthermore, the cumulative distribution function includes product attribute information of historical products, individual characteristic information of the user, and a cumulative distribution function of the time factor of viewing the historical products; using the following formula, one or more of the cumulative distribution functions and the resulting cumulative distribution function are selected to construct a Gaussian coupled joint cumulative distribution function:

[0123]

[0124] in, R(i) represents the Gaussian coupled joint cumulative distribution function, and R(i) represents the variance-covariance matrix of the Gaussian distribution function. Φ R(i) The joint cumulative distribution function used to represent a multivariate Gaussian distribution with zero mean and variance-covariance matrix R(i), Used to represent probability integral transformation functions; y i Used to represent historical purchase results Used to represent the cumulative distribution function of the result. A probability integral transformation function used to represent the cumulative distribution function of the outcome; d is used to represent the various product attribute information viewed by user i. Used to represent the cumulative distribution function based on product attribute information d. Used to indicate based on The inverse cumulative distribution function, Used to indicate based on The probability integral transformation function of the cumulative distribution function; c is used to represent the individual characteristic information of user i. Used to represent the cumulative distribution function based on individual characteristic information c. Used to indicate based on The inverse cumulative distribution function, Used to indicate based on The probability integral transformation function of the cumulative distribution function; Used to represent the time factor of user i Used to represent the cumulative distribution function based on time factors. inverse cumulative distribution function based on probability integral transform function for representing a cumulative distribution function based on

[0125] The variance-covariance matrix R(i) of the Gaussian distribution function can be a preset value, and can also be obtained by calculation.

[0126] Further, the variance-covariance matrix R(i) of the Gaussian distribution function can be determined by the following formula:

[0127] R(i) = Λ(i) -1 / 2 Σ(i)Λ(i) -1 / 2

[0128]

[0129] wherein, the observation value of M it represents the number of products viewed by the user i in the preset time length; Λ(i) represents a diagonal matrix containing diagonal elements of Σ(i), and Σ(i) represents a covariance matrix.

[0130] In the embodiment of the present application, the variance-covariance matrix R(i) of the Gaussian distribution function is determined by using Σ(i) and the diagonal matrix containing the diagonal elements of Σ(i), which can more effectively and accurately determine the variance-covariance matrix, and then construct the Gaussian coupled joint cumulative distribution function based on the variance-covariance matrix R(i) in the subsequent step.

[0131] Further, the Gaussian coupled joint density function can be constructed by the following formula:

[0132]

[0133] wherein, represents the Gaussian coupled joint density function, c(x) represents a coupling function, and y i represents the historical purchase result, represents the result cumulative distribution function, represents each product attribute information d viewed by the user i, represents a cumulative distribution function based on the product attribute information d, represents individual feature information c of the user i, represents a cumulative distribution function based on the individual feature information c, represents a time factor of the user i, ​​Used to represent the cumulative distribution function based on time factors. Used to represent based on y i The result is a smooth density function. Used to indicate based on The smooth density function, Used to indicate based on The smooth density function, Used to represent based on ω li The smooth density function.

[0134] In this embodiment of the invention, by taking advantage of the fact that coupling functions do not require any assumptions about the marginal distribution of variables, they are more suitable for handling imbalanced data.

[0135] It is understandable that, given the variance-covariance matrix R(i), the joint cumulative distribution of the variables can be determined by comparison. as well as We can determine whether user i will purchase product j on day t. As a non-restrictive example, if the probability of purchase is greater than 0.5, we can conclude that user i will purchase the product.

[0136] In the specific implementation of step S14, an objective function is constructed based on the purchase impact characteristics of the target product and the Gaussian coupling joint density function. Then, the objective function is maximized to determine the probability that the target product with the best fitting effect is purchased.

[0137] Furthermore, the objective function can be constructed using the following formula:

[0138]

[0139] in,

[0140] The y' i Used to represent the probability that a target product will be purchased, the The product attribute information used to represent the target product, the ω is used to represent the individual characteristic information of user i. i This is used to indicate the time factor by which the user i views the target product.

[0141] Furthermore, the step of maximizing the objective function to determine the probability that the target product with the best fit is purchased may include: selecting multiple values ​​in [0,1] and assigning them to y' i and calculate respectively The following formula is used to determine y' at its maximum iThe probability that the target product with the best fitting effect will be purchased:

[0142]

[0143] Among them, y' i =Φ -1 (R(i)).

[0144] It should be noted that the obtained The inverse cumulative distribution function of the maximum value can be used to predict behavior.

[0145] For example, select and assign the values ​​0.1, 0.3, and 0.5 to y' in the range [0,1]. i And calculate y' i When the value is 0.3 If the probability of purchase is maximized, then the probability of the target product with the best fitting effect being purchased is 0.3. If it is pre-set that a purchase probability greater than 0.5 indicates that user i will purchase the product, then a purchase probability of 0.3 for the target product is used to indicate that user i will not purchase the target product.

[0146] In this embodiment of the invention, by constructing an appropriate objective function, the objective function can be maximized to determine the probability of the target product with the best fit being purchased, thereby further improving the accuracy of judging the user's purchase intention.

[0147] In the specific implementation of step S15, based on the probability that the target product with the best fitting effect is purchased, it is determined whether to recommend the target product to the user.

[0148] Reference Figure 2 , Figure 2 yes Figure 1 A flowchart of a specific implementation of step S15.

[0149] The historical product viewing information and the target product are both iterative products. The target product can be an iterative product selected in response to the user's product recommendation request. The step of determining whether to recommend the target product to the user based on the probability that the target product with the best fitting effect is purchased can include steps S21 to S22. Each step is described below.

[0150] In step S21, the probabilities of purchasing the multiple target products are sorted.

[0151] Specifically, the aforementioned steps can be used to determine y' at its maximum iand the probability of each target product is ranked.

[0152] As a non-limiting example, the probabilities of 3 target products A, B, C being purchased are 0.2, 0.4, 0.9 respectively, then the 3 target products can be ranked from large to small, and the target product C, B, A is obtained.

[0153] In step S22, it is determined whether to recommend the target product corresponding to each probability to the user according to the ranking result.

[0154] In the above example, the probability of the target product C being purchased is 0.9, which means that the user has a high probability of purchasing, and the target product C can be recommended; the probability of the target product B being purchased is 0.4, which means that the user has a low probability of purchasing, and the target product B can be recommended when there are more recommendation positions, and can not be recommended when there are fewer recommendation positions; the probability of the target product B being purchased is 0.2, which means that the user has a very small probability of purchasing, and the target product B can not be recommended.

[0155] Specifically, the historical product and the target product are iterative products, for example, the product is automatically updated and displayed, and if the display is too late, the product may lose the opportunity to be viewed by the user.

[0156] In the scenario of iterative products, because of the uncertainty of the products that will appear in the future, the user needs to consider more complex factors in the decision-making process, and in the embodiment of the present application, all purchase and non-purchase data, transaction characteristics, individual characteristics, time factors and purchase labels are used to construct a joint distribution describing the entire probability space of purchase behavior, which can more accurately predict user behavior and effectively handle imbalance data problems.

[0157] More specifically, when the observation results are disproportionately distributed in different decision categories, imbalance data is prone to occur, and it is difficult to combine many variables from different distributions to predict user behavior. In the embodiment of the present application, by using the semi-parametric Gaussian continuous (SGC) method, the advantage of not requiring any assumption on the marginal distribution of the variable can be utilized by using the coupling function, and the imbalance data can be effectively handled. And because a semi-parametric model is used, the prediction is based on the selection likelihood, so it has strong prediction performance when applied to imbalance data.

[0158] In the embodiments of the present application, in the case that the historical products and the target products are iterative products, the probabilities of the target products being purchased are sorted, and it is determined whether to recommend the target products corresponding to the probabilities to the user according to the sorting result, so that the products more in line with the user's intention can be recommended to the user, and the recommendation-view conversion rate is improved, and even the probability of being purchased is improved.

[0159] Further, the iterative products can be fast iterative products.

[0160] The fast iterative products can be displayed with an update frequency greater than or equal to a preset frequency, or the next update time is difficult to determine. For example, a special offer product in a shopping page has a time limit, and will be invalid after a certain period of time; for example, a limited quantity product will be invalid after being purchased; for example, a recommendation page of online dating, after the user scrolls through the page, the user may not be able to return to view again. In such cases, it is more necessary to recommend products more in line with the user's intention to the user, so as not to miss the matching products in the process of viewing other products.

[0161] In a non-limiting specific embodiment, unbalanced data with different purchase rates is generated, including severe imbalance (5%) and moderate imbalance (20%) purchase rates. During data generation, 12 covariates are simulated: four are sampled from a normal distribution, four are sampled from a mixed distribution, and four are generated from a skewed gamma distribution, so as to reflect the complexity of the data obtained in the real scene.

[0162] Further, since factors are usually nonlinearly related to the final decision, square, exponential and cosine transformations can also be applied to each group of covariates.

[0163] Next, the SGC method in the embodiments of the present application can be used to predict the purchase amount and check the hit rate. In the following description, 5000 training records are simulated, and the prediction performance of SGC is tested on another 2500 reserved records, wherein the purchase proportion of the training data and the test data is the same (i.e. 5% or 20%).

[0164] After 500 rounds of simulation, the average hit rate of the data with severe imbalance is 91.88% (6.16%), and the average hit rate of the data with moderate imbalance is 96.03% (3.96%). In both cases, the SGC method in the embodiments of the present application can obtain more than 90% of the future purchase.

[0165] Specifically, the SGC method uses a non-parametric kernel density estimation (KDE) to smooth the feature values to the same scale and estimate the density of each feature, which can effectively describe the distribution of the data. Taking the selection of a covariate in a normal distribution, the selection of a covariate in a mixture distribution, and the selection of a covariate in a skew distribution as examples, the covariate distribution of the purchase transaction is first predicted, and then compared with the covariate distribution of the real (simulated) purchase transaction. No matter whether the distribution is unimodal or multimodal, the SGC method can effectively capture the distribution of the covariate. This shows that the SGC method can accurately reflect the information embedded in the complex distribution of different covariates and combine them for prediction. Therefore, even for unbalanced data, SGC can achieve excellent hit rate.

[0166] In the embodiment of the present application, by obtaining the historical product viewing information of the user, and forming a cumulative distribution function and a result cumulative distribution function based on the historical viewing information, a Gaussian-Gaussian coupling joint cumulative distribution function and a Gaussian coupling joint density function are further constructed, and then a target function is constructed according to the purchase influence characteristics of the target product, the probability of the target product being purchased with the optimal fitting effect is determined, and the willingness of the user to the target product is predicted, so that through big data processing technology, after judging the target product to be recommended in advance, there is an opportunity to decide whether to recommend the target product to the user according to the viewing willingness of the user to the target product, effectively improving the recommendation-viewing conversion rate, and even improving the probability of being purchased.

[0167] Reference Figure 3 , Figure 3 is a structural schematic diagram of a product recommendation device in an embodiment of the present application. The product recommendation device can include:

[0168] The acquisition module 31 is configured to acquire one or more historical product viewing information of a user within a preset time length, and the historical product viewing information includes historical purchase results and one or more historical purchase influence characteristics.

[0169] The first function determination module 32 is configured to form a cumulative distribution function for each historical purchase influence characteristic by using a preset kernel function, and form a result cumulative distribution function for the historical purchase results by using the preset kernel function.

[0170] The second function determination module 33 is configured to select one or more cumulative distribution functions and the result cumulative distribution function, construct a Gaussian coupling joint cumulative distribution function, and construct a Gaussian coupling joint density function.

[0171] The fitting module 34 is configured to construct a target function according to the purchase influence feature of the target product and the Gaussian coupling joint density function, and maximize the target function to determine a probability that the target product is purchased with the best fitting effect.

[0172] The recommendation determination module 35 is configured to determine whether to recommend the target product to the user according to the probability that the target product is purchased with the best fitting effect.

[0173] In specific implementations, the apparatus can correspond to a chip with a data processing function in a terminal, or a chip module including a chip with a data processing function in a terminal, or a terminal.

[0174] As to Figure 3 The working principle, working mode and beneficial effects of the product recommendation apparatus shown above can be referred to the foregoing description and Figures 1 to 2 The foregoing description, and will not be repeated here.

[0175] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is run by a processor to execute the steps of the product recommendation method. The storage medium can include ROM, RAM, magnetic disk or optical disk, etc. The storage medium can further include non-volatile memory or non-transitory memory, etc.

[0176] The embodiment of the present application further provides a terminal, which includes a memory and a processor, and the memory stores a computer program that can be run on the processor, and the processor executes the steps of the product recommendation method when running the computer program. The terminal includes but is not limited to terminal devices such as mobile phones, computers, tablet computers, servers, cloud platforms, etc.

[0177] It should be understood that, in the embodiments of the present application, the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0178] It should also be understood that the memory in the embodiments of the present application can be a volatile memory or a nonvolatile memory, or can include both volatile and nonvolatile memory. Among them, the nonvolatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or flash memory. The volatile memory can be random access memory (RAM) used as an external cache. By way of example, but not by way of limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchlink dynamic random access memory (SLDRAM) and direct rambus random access memory (DR RAM).

[0179] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer program can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer program can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired or wireless means.

[0180] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically included separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software function unit. For example, for each device or product applied to or integrated in a chip, each module / unit contained therein can be realized in the form of hardware such as circuit, or at least part of the modules / units can be realized in the form of software program running on a processor integrated in the chip, and the remaining (if any) modules / units can be realized in the form of hardware such as circuit; for each device or product applied to or integrated in a chip module, each module / unit contained therein can be realized in the form of hardware such as circuit, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or different components of the chip module, or at least part of the modules / units can be realized in the form of software program running on a processor integrated in the chip module, and the remaining (if any) modules / units can be realized in the form of hardware such as circuit.

[0181] It should be understood that the term "and / or" herein only describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent three cases of A alone, A and B together, and B alone. In addition, the character " / " herein represents an "or" relationship between the associated objects before and after it.

[0182] "Multiple" appearing in the embodiments of the present application means two or more.

[0183] The first, second, and the like appearing in the embodiments of the present application are only for illustrative and distinguishing purposes, and do not have order, nor represent a special limitation on the number of devices in the embodiments of the present application, and cannot constitute any limitation on the embodiments of the present application.

[0184] Although the present application is disclosed as above, the present application is not limited thereto. Any person skilled in the art can make various modifications and changes without departing from the spirit and scope of the present application, and therefore the protection scope of the present application should be subject to the scope defined by the claims.

Claims

1. A product recommendation method characterized by, The method comprises the following steps: obtaining one or more historical product viewing information of a user within a preset time length, wherein the historical product viewing information comprises historical purchase results and one or more historical purchase influence characteristics; using a preset kernel function, forming a cumulative distribution function for each historical purchase influence characteristic, and using the preset kernel function, forming a result cumulative distribution function for the historical purchase results; selecting one or more cumulative distribution functions and the result cumulative distribution function to construct a Gaussian coupled joint cumulative distribution function and a Gaussian coupled joint density function; constructing a target function according to a purchase influence characteristic of a target product and the Gaussian coupled joint density function, and maximizing the target function to determine a probability that the target product is purchased with the best fitting effect; determining whether to recommend the target product to the user according to the probability that the target product is purchased with the best fitting effect; wherein the forming of the cumulative distribution function comprises: using the preset kernel function to form a smooth density function for each historical purchase influence characteristic; forming the cumulative distribution function according to the smooth density function; wherein the forming of the result cumulative distribution function comprises: using the preset kernel function to form a result smooth density function for the historical purchase results; forming the result cumulative distribution function according to the result smooth density function; wherein the cumulative distribution function comprises a cumulative distribution function of product attribute information of a historical product, individual characteristic information of the user, and a time factor of viewing the historical product; selecting one or more cumulative distribution functions and the result cumulative distribution function to construct a Gaussian coupled joint cumulative distribution function using the following formula: ； wherein, R(i) denotes a variance-covariance matrix of a Gaussian distribution function for representing a joint cumulative distribution function of Gaussian coupling, R(i) denotes a variance-covariance matrix of a Gaussian distribution function for representing a joint cumulative distribution function of Gaussian coupling, for representing a probability integral transform function; for representing historical purchase results, for representing a cumulative distribution function of results, a probability integral transform function for representing a cumulative distribution function of results; for representing each product attribute information d viewed by the user i, for representing a cumulative distribution function based on the product attribute information d, for representing a cumulative distribution function based on an inverse cumulative distribution function, for representing a probability integral transform function based on a cumulative distribution function. an individual characteristic information c for representing a user i, a cumulative distribution function for representing based on the individual characteristic information c, an inverse cumulative distribution function for representing based on the individual characteristic information c, a probability integral transform function for representing based on the cumulative distribution function. a time factor for representing a user i, a cumulative distribution function for representing based on the time factor, a cumulative distribution function for representing based on an inverse cumulative distribution function, a probability integral transform function for representing based on a cumulative distribution function.

2. The method of claim 1, wherein, the historical product to which the historical product viewing information belongs and the target product are iterative products, and the target product is an iterative product selected in response to a product recommendation request of the user after receiving the product recommendation request; determining whether to recommend the target product to the user according to the probability that the target product is purchased with the best fitting effect comprises: sorting the probabilities that multiple target products are purchased; determining whether to recommend target products corresponding to each probability to the user according to the sorting result.

3. The method of claim 1, wherein, The historical purchase influence characteristics are selected from one or more of the following: product attribute information of a historical product, individual characteristic information of the user, and a time factor of viewing the historical product.

4. The method of claim 1, wherein, The historical purchase influence characteristics are product attribute information of a historical product, and the preset kernel function is a Gaussian kernel function; using the following formula to form the smooth density function: ； and / or, using the following formula to form the cumulative distribution function according to the smooth density function: ； wherein, for representing the product attribute information d viewed by the user i, for representing the cumulative distribution function based on a smooth density function, and the function is a Gaussian function, for representing the product attribute information d of the product j viewed by the user i at the time t, for representing the cumulative distribution function based on a smooth parameter, i.e., a bandwidth; T is used to represent the preset time length, is used to represent the number of products viewed by user i within the preset time length.

5. The method of claim 1, wherein, The historical purchase influence characteristics are individual characteristic information of the user, and the preset kernel function is a Gaussian kernel function; using the following formula to form the smooth density function: ； and / or, using the following formula to form the cumulative distribution function according to the smooth density function: ； wherein, for representing the individual feature information c of the user i, for representing the cumulative distribution function based on a smooth density function, and the function is a Gaussian function, for representing the individual feature information c of the product j viewed by the user i at the time t, for representing the cumulative distribution function based on a smooth parameter, i.e., a bandwidth; T is used to represent the preset time length, is used to represent the number of products viewed by user i within the preset time length.

6. The method of claim 1, wherein, The historical purchase influence characteristics are a time factor of viewing the historical product, and the preset kernel function is a Gaussian kernel function; The smoothing density function is formed by using the following formula: ； and / or, The cumulative distribution function is formed according to the smoothing density function by using the following formula: ； wherein, a time factor for representing the user i, a cumulative distribution function based on a smoothing density function, and the function is a Gaussian function, a time factor for representing the product j viewed by the user i at time t, a cumulative distribution function based on a smoothing parameter, i.e. bandwidth; T is used to represent the preset time length, is used to represent the number of products viewed by user i within the preset time length.

7. The method of claim 1, wherein, The preset kernel function is a Gaussian kernel function; The result smoothing density function is formed by using the following formula: ； and / or, The result cumulative distribution function is formed according to the result smoothing density function by using the following formula: ； wherein, for indicating the historical purchase result of the user i as purchase or non-purchase, for indicating the result smoothing density function based on , and the function is a Gaussian function, for indicating the historical purchase result of the product j viewed by the user i at the time t as purchase or non-purchase, for indicating the result cumulative distribution function based on , h is a smoothing parameter, i.e. a bandwidth; T is used to represent the preset time length, is used to represent the number of products viewed by user i within the preset time length.

8. The method of claim 1, wherein, The preset kernel function is selected from one or more of the following: a Gaussian kernel function, an exponential kernel function, a polynomial kernel function, and a box kernel function.

9. The method of claim 1, wherein, The variance-covariance matrix R(i) of the Gaussian distribution function is determined by using the following formula: ； ； wherein, for representing an observation value, for representing the number of products viewed by user i in the preset time length; for representing a diagonal matrix containing diagonal elements, for representing a covariance matrix.

10. The method of claim 1, wherein, The Gaussian coupled joint density function is constructed by using the following formula: ； wherein, for representing a Gaussian coupling joint density function, c(x) for representing a coupling function, for representing a historical purchase result, for representing a result cumulative distribution function, for representing each product attribute information d viewed by the user i, for representing a cumulative distribution function based on the product attribute information d, for representing individual characteristic information c of the user i, for representing a cumulative distribution function based on the individual characteristic information c, for representing a time factor of the user i, for representing a cumulative distribution function based on the time factor, for representing a result smoothing density function based on , for representing a smoothing density function based on , for representing a smoothing density function based on , for representing a smoothing density function based on .

11. The method of claim 10, wherein, The objective function is constructed by using the following formula: ； wherein , ， The for representing a probability that a target product is purchased, the for representing product attribute information of the target product, the for representing individual feature information of the user i, the for representing a time factor that the user i views the target product.

12. The method of claim 11, wherein, Maximizing the objective function to determine the probability that the target product is purchased with the best fitting effect includes: Selecting a plurality of values in [0,1] to be assigned to and calculating respectively The following formula is used to determine the maximum probability that the target product is purchased as the best fit for the effect. ； wherein .

13. A product recommendation device characterized by comprising: includes: An acquisition module is configured to acquire one or more historical product viewing information of a user within a preset time length, the historical product viewing information including a historical purchase result and one or more historical purchase influence features; A first function determination module is configured to form a cumulative distribution function for each historical purchase influence feature by using a preset kernel function, and form a result cumulative distribution function for the historical purchase result by using the preset kernel function; A second function determination module is configured to select one or more of the cumulative distribution functions and the result cumulative distribution function, construct a Gaussian coupled joint cumulative distribution function, and construct a Gaussian coupled joint density function; A fitting module is configured to construct an objective function according to a purchase influence feature of a target product and the Gaussian coupled joint density function, and maximize the objective function to determine the probability that the target product is purchased with the best fitting effect; A recommendation determination module is configured to determine whether to recommend the target product to the user according to the probability that the target product is purchased with the best fitting effect; The cumulative distribution function is formed by using the preset kernel function to form a smoothing density function for each historical purchase influence feature; The cumulative distribution function is formed according to the smoothing density function; The result cumulative distribution function is formed by using the preset kernel function to form a result smoothing density function for the historical purchase result; The result cumulative distribution function is formed according to the result smoothing density function; The cumulative distribution function contains the cumulative distribution function of the product attribute information of the historical product, the individual feature information of the user, and the time factor of viewing the historical product; One or more of the cumulative distribution functions and the result cumulative distribution function are selected to construct a Gaussian coupled joint cumulative distribution function by using the following formula: The computer program is executed by the processor to perform the steps of the product recommendation method of any one of claims 1 to 12. The processor executes the computer program to perform the steps of the product recommendation method of any one of claims 1 to 12. ； wherein, R(i) denotes a variance-covariance matrix of a Gaussian distribution function for representing a joint cumulative distribution function of Gaussian coupling, R(i) denotes a variance-covariance matrix of a Gaussian distribution function for representing a joint cumulative distribution function of Gaussian coupling, for representing a probability integral transform function; for representing historical purchase results, for representing a cumulative distribution function of results, a probability integral transform function for representing a cumulative distribution function of results; for representing each product attribute information d viewed by the user i, for representing a cumulative distribution function based on the product attribute information d, for representing a cumulative distribution function based on an inverse cumulative distribution function, for representing a probability integral transform function based on a cumulative distribution function. c is used to represent the individual characteristic information of user i. Used to represent the cumulative distribution function based on individual characteristic information c. Used to indicate based on The inverse cumulative distribution function, Used to indicate based on The probability integral transformation function of the cumulative distribution function; a time factor for representing a user i, a cumulative distribution function for representing based on the time factor, an inverse cumulative distribution function for representing based on a cumulative distribution function for representing based on a probability integral transform function for representing based on a cumulative distribution function.

14. A computer-readable storage medium having stored thereon a computer program, characterized in that, ​ 15. A terminal comprising a memory and a processor, said memory having stored thereon a computer program capable of running on said processor, characterized in that, ​

Images