App preference determination method and device, computer readable storage medium, and terminal

Abstract

An APP preference determination method and device, a computer readable storage medium and a terminal, the method comprising: obtaining access attribute information of one or more historical APPs of a user accessed within a preset time length; determining an expression of a utility function of a recommendation model, the expression of the utility function containing the access attribute information, an implicit attribute parameter and an APP association relationship parameter; estimating a plurality of sets of parameter vectors of the recommendation model, wherein each set of parameters includes the access attribute information, the implicit attribute parameter and the APP association relationship parameter; using a preset objective function to obtain a set of parameter vectors with the optimal fitting effect; and using the set of parameter vectors with the optimal fitting effect to determine the preference of the user for a target APP. The application has the opportunity to determine whether to recommend to the user according to the installation willingness of the user for the target APP, thereby effectively improving the recommendation-download conversion rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to an APP preference determining method and device, computer readable storage medium and terminal. BACKGROUND

[0002] With the rapid development of Internet technology and the rapid popularization of intelligent terminals, application program (APP) related technologies are booming, and various APPs that can meet the application needs of users in different fields and different problems are emerging in an endless stream, and the demand of users for APP download and installation is also increasing.

[0003] In the prior art, a background server can recommend APPs to users and provide download channels via platforms such as application stores, however, the existing APP recommendation method is too single, which may recommend APPs that users are not interested in and have low installation willingness to users, resulting in a low conversion rate of APP recommendation-download.

[0004] There is an urgent need for an APP preference determining method, which can judge target APPs in an APP database in advance, so as to have the opportunity to decide whether to recommend to users according to the installation willingness of users to the target APPs. SUMMARY

[0005] The technical problem solved by the present application is to provide an APP preference determining method and device, computer readable storage medium and terminal, which have the opportunity to decide whether to recommend to users according to the installation willingness of users to target APPs, and effectively improve the conversion rate of recommendation-download.

[0006] To solve the above technical problem, an application program (APP) preference determining method is provided, comprising: obtaining access attribute information of one or more historical APPs of a user accessed within a preset time length; determining an expression of a utility function of a recommendation model, the expression of the utility function containing the access attribute information, implicit attribute parameters and APP association relationship parameters; estimating a plurality of groups of parameter vectors of the recommendation model, wherein each group of parameters includes the access attribute information, implicit attribute parameters and APP association relationship parameters; using a preset target function to obtain a group of parameter vectors with the optimal fitting effect; and using the group of parameter vectors with the optimal fitting effect to determine the preference of the user to a target APP.

[0007] Optionally, the target APP is one or more APPs selected based on the request APP after receiving the user's input; determining the user's preference for the target APP using the set of parameter vectors with the best fitting effect includes: if, in the parameter vector with the best fitting effect, the transpose vector ρ of the latent attribute parameter α of the target APP and the APP association parameter of the request APP... T If the product of the parameters is greater than zero, then the target app is determined to be the app to be recommended based on the requested app; if, in the parameter vector with the best fitting effect, the transpose vector ρ of the latent attribute parameter α of the target app and the app association parameter of the requested app is greater than zero, then the target app is determined to be the app to be recommended based on the requested app; T If the product of the two is less than zero, then the target APP is determined to be the APP to be avoided based on the requesting APP.

[0008] Optionally, the access attribute information is selected from one or more of the following: APP download volume parameter λ, user preference parameter θ, potential time coefficient δ, access time parameter μ, user data traffic sensitivity parameter γ, APP data traffic parameter β, data traffic usage m, APP category parameter c, and the period parameter d to which the access time belongs.

[0009] Optionally, for each access to a historical app, the expression of the utility function of the recommendation model includes a first utility sub-function ψ. its and the sum of the second utility subfunction; wherein, the first utility subfunction ψ its The second utility function is determined based on the access attribute information and implicit attribute parameters of the historical APP; the second utility function is determined based on the implicit attribute parameters of other historical APPs accessed before the historical APP within a preset time period and the association parameters between the historical APP and each other historical APP.

[0010] Optionally, the first utility subfunction ψ its The expressions include The sum of one or more of the following: the download count parameter λ of the historical apps. s ; Transpose of the latent time coefficient With access time parameter μ s product The transpose of the user's data traffic sensitivity within a preset time period The historical APP's data traffic parameter β s The logarithm of the historical APP's data traffic usage per minute within the preset time period. ts negative of the product in, Transpose vector representing the user's preference parameters a product of the implicit attribute parameter a of the historical APP s i represents the i-th user, s represents the APP s, and t represents the time of the current visit of the historical APP.

[0011] Optionally, the expression of the second utility sub-function is wherein, a transpose vector of the APP association parameter of the historical APP, an implicit attribute parameter of the k-th other historical APP accessed before the historical APP within the preset time length; j represents the order of the current visit of the historical APP within the preset time length, k is a positive integer, and k≤j-1; i represents the i-th user, s represents the APP s, and t represents the time of the current visit of the historical APP.

[0012] Optionally, for each visit of each historical APP, the expression of the utility function of the recommendation model further includes a sum of a third utility sub-function; wherein the third utility sub-function is a maximum value obtained by calculating a sum value of the first utility sub-function and the second utility sub-function for one or more APPs accessed subsequently to the historical APP.

[0013] Optionally, the expression of the third utility sub-function is

[0014] wherein, ψ its′ a first utility sub-function of the APP accessed subsequently to the historical APP, a transpose vector of the APP association parameter of the APP accessed subsequently to the historical APP, an implicit attribute parameter of the k-th other historical APP accessed before the historical APP within the preset time length, a s an implicit attribute parameter of the historical APP; j represents the order of the current visit of the historical APP within the preset time length, k is a positive integer, and k≤j-1; i represents the i-th user, s represents the APP s, and t represents the time of the current visit of the historical APP.

[0015] Optionally, the following posterior distribution formula is used to estimate the multiple groups of parameter vectors of the recommendation model:

[0016]

[0017] Ω is used to represent the space set of the access attribute information, the implicit attribute parameter and the APP association relationship parameter, is used to represent the set of historical APPs used by the plurality of users within the plurality of preset time lengths, x is used to represent other influence factors, is used to represent the given and x, the predicted access probability of the APP based on the Ω space; z it is used to represent the historical APP accessed by the i-th user at the t-th time, x it is used to represent other influence factors when the i-th user accesses the historical APP at the t-th time, p(z it |Ω, x it is used to represent the predicted access probability of the historical APP accessed by the i-th user at the t-th time under the condition of the given Ω space and x, and p(Ω) is used to represent the prior distribution based on the Ω space; is used to represent the predicted access probability of the set of historical APPs accessed by the i-th user at the t-th time under the condition of x.

[0018] Optionally, the probability parameters are determined as follows:

[0019]

[0020]

[0021]

[0022] wherein s, s' and s" are respectively used to represent the historical APPs APP s, APP s' and APP s", Ψ(s, z it,j-1 ) is used to represent the j-1-th APP accessed by the i-th user at the t-th time as APP s, Ψ(s', z it,j-1 ) is used to represent the j-1-th APP accessed by the i-th user at the t-th time as APP s', p(z it,j =s|z it,j-1 ) is used to represent the probability that the j-1-th APP accessed by the i-th user at the t-th time is APP s.

[0023] Optionally, the prior distribution p(Ω) is selected from a Gaussian prior distribution and a GAMMA prior distribution.

[0024] Optionally, the other influence factors are selected from one or more of the following: data traffic usage m, APP category parameter c and period parameter d to which the access time belongs; wherein, in the process of estimating a plurality of groups of parameter vectors of the recommendation model at the same time, the other influence factors do not repeat the parameters contained in the access attribute information.

[0025] Optionally, the preset target function is:

[0026]

[0027] Wherein, L(v) is used for representing the preset target function, For representing the joint density function after given x, q(Ω;v) is used for representing the APP access probability based on Ω space of the parameter randomly extracted from v complex parameter space, E q(Ω；v) For representing the maximum lower bound function of evidence.

[0028] To solve the above technical problems, the embodiment of the present application provides an APP preference determination device, comprising: an acquisition module, configured to acquire access attribute information of one or more historical APPs of a user in a preset time length; an expression determination module, configured to determine an expression of a utility function of a recommendation model, the expression of the utility function containing the access attribute information, an implicit attribute parameter and an APP association relationship parameter; an estimation module, configured to estimate a plurality of parameter vectors of the recommendation model, wherein each parameter includes the access attribute information, the implicit attribute parameter and the APP association relationship parameter; a fitting module, configured to obtain a parameter vector with optimal fitting effect by using a preset target function; and a preference determination module, configured to determine a preference of the user for a target APP by using the parameter vector with optimal fitting effect.

[0029] To solve the above technical problems, the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is run on a processor to execute the steps of the APP preference determination method.

[0030] To solve the above technical problems, the embodiment of the present application provides a terminal, comprising a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor runs the computer program to execute the steps of the APP preference determination method.

[0031] Compared with the prior art, the technical scheme of the embodiment of the present application has the following beneficial effects:

[0032] In the embodiment of the present application, by constructing a utility function, estimating a parameter vector by using a plurality of parameters including the access attribute information, the implicit attribute parameter and the APP association relationship parameter, and then obtaining a parameter vector with optimal effect through fitting, the preference of the user for a target APP can be predicted, so that through big data processing technology, after judging the target APP in the APP database in advance, there is an opportunity to decide whether to recommend the user according to the installation willingness of the user for the target APP, and the recommendation-download conversion rate is effectively improved.

[0033] Further, after receiving the request APP of the user input, the target APP selected by the request APP is determined, and the product of the implicit attribute parameter a and the transpose vector p of the APP association relationship parameter of the request APP is determined T The precious recommendation resources can be implemented on the APP with higher installation willingness, the recommendation-download conversion rate is improved, and the recommendation quota is avoided from being wasted.

[0034] Further, the expression of the utility function can be constructed based on the historical APP and other historical APPs accessed earlier, so that the influence of the other historical APPs on the selection of the historical APP is determined by using the implicit attribute parameters of the two and the association relationship parameter therebetween, the potential download habit of the user is more effectively judged, the utility function customized for the user is more accurate, and the accuracy of judging the installation willingness of the user is further improved.

[0035] Further, a third utility sub-function is constructed based on the APP accessed after the historical APP, so that the influence of the other historical APPs on the selection of the historical APP and the influence of the APP accessed subsequently on the access of the historical APP are determined by using the implicit attribute parameters of the historical APP, the other historical APPs accessed previously and the APP accessed subsequently and the association relationship parameters between the latter two and the historical APP, the potential download habit of the user is more effectively judged, the utility function customized for the user is more accurate, and the accuracy of judging the installation willingness of the user is further improved.

[0036] Further, based on more users and a longer preset time length, a posterior distribution formula is used to estimate a plurality of parameter vectors of the recommendation model, so that the commonness of the potential download habits of more users can be determined by using big data processing technology, the application range of the recommendation model in the embodiment of the application is more extensive and accurate.

[0037] Further, by selecting a proper target function, v that maximizes L(v) can be found by iterative optimization after the distribution family of q(Ω; v) is specified, that is, a set of parameter vectors with the optimal fitting effect is obtained, and the accuracy of determining the preference of the user for the target APP is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is a flowchart of an application program APP preference determination method in the embodiment of the application;

[0039] Figure 2 is Figure 1 is a flowchart of a specific implementation of step S15 in the embodiment of the application.

[0040] Figure 3 is a structural schematic diagram of an application program APP preference determination device in an embodiment of the present application. DETAILED DESCRIPTION

[0041] In the prior art, a background server can recommend an APP to a user and provide a download channel via a platform such as an application store, however, the existing APP recommendation method is too single, and a user can be recommended an APP which he / she is not interested in and has a low installation willingness, resulting in a low recommendation-download conversion rate of the APP.

[0042] The inventor of the present application has found through research that in the existing application program recommendation scheme, a user input request APP is usually obtained first, and APPs of the same type are filtered out from an APP database according to the type of the request APP, and are recommended to the user. However, since the APPs of the same type are often replaceable, the user can meet the demand by downloading one of them, and therefore has a low installation willingness for the recommended APP and a poor user experience.

[0043] In an embodiment of the present application, a utility function is constructed, a plurality of parameters including the access attribute information, the implicit attribute parameter and the APP correlation relationship parameter are used to estimate a parameter vector, and then a set of parameter vectors with the best effect is obtained through fitting, so as to predict the preference of a user for a target APP, and thus through big data processing technology, the target APP in the APP database is judged in advance, and there is an opportunity to decide whether to recommend the user according to the installation willingness of the user for the target APP, thereby effectively improving the recommendation-download conversion rate.

[0044] In order to make the above-mentioned purposes, characteristics and advantages 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.

[0045] Reference Figure 1 , Figure 1 is a flowchart of an application program APP preference determination method in an embodiment of the present application. The application program APP preference determination method can include steps S11 to S15:

[0046] Step S11: obtaining access attribute information of one or more historical APPs of a user accessed within a preset time length;

[0047] Step S12: determining an expression of a utility function of a recommendation model, the expression of the utility function containing the access attribute information, an implicit attribute parameter and an APP correlation relationship parameter;

[0048] Step S13: Estimate multiple sets of parameter vectors for the recommendation model, wherein each set of parameters includes the access attribute information, latent attribute parameters, and APP association parameters;

[0049] Step S14: Using a preset objective function, obtain a set of parameter vectors with the best fitting effect;

[0050] Step S15: Using the set of parameter vectors with the best fitting effect, determine the user's preference for the target APP.

[0051] In the specific implementation of step S11, for the user, there may be a certain number of apps in each time period. The relevant information of these apps can be used as historical data to build a recommendation model.

[0052] It is understandable that the preset duration should not be set too short, otherwise the amount of information obtained will be too small, making it difficult to determine the factors influencing the user's download; the preset duration should not be set too long, otherwise the computational load will be too large, affecting the stability of the recommendation model. As a non-limiting example, the preset duration can be set to be selected from 1 hour to 1 month, for example, 1 day.

[0053] In practical applications, user i can access m one by one on day t. t A history app can be represented as Among them, z it,j This can represent the j-th historical app visited by user i, where 1 ≤ j ≤ m. t ,j can be used to represent the order in which the historical APP is accessed within the preset time period.

[0054] Each time user i faces a choice, she selects the app that maximizes her utility from all available apps. Her utility from choosing the app in the j-th order can be expressed by the following formula: where,

[0055] U its (z it,j-1 )=Ψ(s,z it,j-1 )+ε its

[0056] Among them, U its (z it,j-1 ) is z it,j-1 A function Ψ(s,z) is used to represent the utility of user i choosing the app in the j-th order. it,j-1 ) is used to represent deterministic utility, ε its Used to represent the independent distribution of the Gumbel distribution.

[0057] Further, the access attribute information can be selected from one or more of the following: an APP download amount parameter λ, a user preference parameter θ, a potential time coefficient δ, an access time parameter μ, a user data flow sensitivity parameter γ, an APP data flow parameter β, a data flow usage amount m, an APP category parameter c, and a period parameter d to which the access time belongs.

[0058] In particular, the machine learning method is usually based on the similarity of the application and the user, and aims to maximize the prediction rate. However, the user often determines the APP usage decision based on a combination of various reasons, such as considering the application attributes of the APP, the personal preferences of the user, the time effect of the APP, etc. These factors should also be extracted as influencing factors of the user's preference for the APP for analysis.

[0059] The APP download amount parameter λ can be used to represent the overall popularity of each application, and a higher λ value will increase the utility of the user when accessing the APP.

[0060] The user preference parameter θ can be used to represent the personal selection tendency of the user.

[0061] Regarding the potential time coefficient δ and the access time parameter μ, the time effect (also known as seasonality) also affects the user's selection of the application. For example, on weekdays, users more frequently use work-related APPs, and on weekends, users more frequently use game APPs and audio / video APPs.

[0062] More specifically, the access time parameter μ can be used to model the seasonality of the application. As a non-limiting example, two comparable applications can have similar seasonal effects, so we can assume that APPs of the same category share the same access time parameter μ. The potential time coefficient δ can be used to represent the change between different preset time periods (e.g., daily changes).

[0063] Regarding the user data flow sensitivity parameter γ, the APP data flow parameter β, and the data flow usage amount m, considering that users are often sensitive to the data flow consumption of the APP, they may deliberately reduce or even not use a certain APP due to the consumption of data flow. Therefore, the user's sensitivity to data flow can be decomposed into a potential vector for each user (i.e., the user data flow sensitivity parameter γ), a potential vector for each application (i.e., the APP data flow parameter β), and the user data flow usage amount m (e.g., which can be refined to the user's data flow usage amount per minute for the APP on the tth day).

[0064] It should be noted that, in order to avoid the influence of scale difference, and facilitate other parameters to capture the average result distribution, the average value of each APP can be used to standardize the data traffic usage m.

[0065] As a non-limiting example, The personal sensitivity of user i to APP s can be represented. Since the probability of user i selecting APP s decreases when the data traffic usage m of APP s per minute is large, a negative expression (for example ) can be used to better characterize the actual situation.

[0066] The APP category parameter c can be used to represent the category of the APP, such as shopping APP, maternal and infant APP, game APP, etc.

[0067] The period parameter d to which the access time belongs can be used to represent one or more of the following: which day of the week the access time belongs to and holiday information at that time, such as whether it belongs to a holiday, what kind of holiday, etc.

[0068] In the embodiments of the present application, by setting multiple access attribute information, the APP usage decision of the user can be analyzed based on the comprehensive results of multiple reasons, which helps to improve the accuracy of subsequent judgment and prediction.

[0069] In the specific implementation of step S12, the expression of the utility function of the recommendation model can include the access attribute information, the implicit attribute parameter, and the APP association relationship parameter.

[0070] When the user determines the APP usage decision based on the comprehensive results of multiple reasons, in addition to considering various explicit factors (such as the application attribute of the APP, the personal preference of the user himself, the time effect of the APP, etc. mentioned above), the user may also be influenced by environmental factors, other people's influence, etc. when selecting the APP. The implicit factors have non-obviousness and are difficult to be observed by researchers, and can be represented by the implicit attribute parameter α.

[0071] The APP association relationship parameter ρ can be used to represent the association relationship between APPs, such as the competition or complementarity between the jth APP accessed by the user and the 1st, 2nd, …, j-1th historical APPs accessed before.

[0072] In other words, APPs' are complementary to APPs, meaning that after a user visits APPs', the likelihood of visiting APPs increases; for example, after visiting a maternal and infant APP, the likelihood of visiting a shopping APP increases. APPs' are competitive to APPs, meaning that after a user visits APPs', the likelihood of visiting APPs decreases; for example, after visiting the Baidu Maps APP, the likelihood of visiting the Gaode Maps APP decreases.

[0073] In this embodiment of the invention, by additionally setting the implicit attribute parameter α and the APP association parameter ρ, the user's access decisions for each historical APP can be analyzed more comprehensively, which helps to improve the accuracy of prediction when predicting the user's preference for the target APP in the future.

[0074] Furthermore, for each access to a historical app, the expression of the utility function of the recommendation model may include a first utility sub-function ψ. its and the sum of the second utility subfunction; wherein, the first utility subfunction ψ its The second utility function is determined based on the access attribute information and implicit attribute parameters of the historical APP; the second utility function is determined based on the implicit attribute parameters of other historical APPs accessed before the historical APP within a preset time period and the association parameters between the historical APP and each other historical APP.

[0075] Furthermore, the first utility subfunction ψ its The expression can include The sum of one or more of the following: the download count parameter λ of the historical apps. s ; Transpose of the potential time coefficient With access time parameter μ s product The transpose of the user's data traffic sensitivity within a preset time period The historical APP's data traffic parameter β s The logarithm of the historical APP's data traffic usage per minute within the preset time period. ts negative of the product in, Transpose vector representing the user's preference parameters The implicit attribute parameter α of the historical APP s The product of; i is used to indicate that the user is the i-th user, s is used to indicate that the historical APP is APP s, and t is used to indicate the time when the historical APP was accessed for the current time.

[0076] As a non-limiting example, when the above three are selected, the first utility sub-function ψ its is expressed as:

[0077]

[0078] As can be seen from the foregoing, t can also be simplified to represent that the current visit of the historical APP is made on the tth day, and in embodiments of the present application, the precision of t is not limited.

[0079] Further, the expression of the second utility sub-function is wherein, is a transpose vector used to represent the APP association relationship parameter of the historical APP, is an implicit attribute parameter used to represent the kth other historical APP accessed before the historical APP within the preset time period; j is used to represent the order of the current visit of the historical APP within the preset time period, k is a positive integer, and k≤j-1; i is used to represent the ith user, s is used to represent the APP s, and t is used to represent the time of the current visit of the historical APP.

[0080] Specifically, APP s' and APP s can have a complementary relationship or a competitive relationship. In embodiments of the present application, the product of the implicit attribute parameter a and the transpose vector p T of the APP association relationship parameter can be used to represent the competitive level between the two APPs.

[0081] As can be seen from the above, for each visit of each historical APP, the expression of the utility function of the recommendation model can be:

[0082]

[0083] In embodiments of the present application, the expression of the utility function can be constructed based on the historical APP and other historical APPs accessed earlier, so as to determine the influence of the other historical APPs on the selection of the historical APP by using the implicit attribute parameters of the two and the association relationship parameter therebetween, thereby more effectively judging the potential download habits of the user, forming a utility function more customized and more accurate for the user, and further improving the accuracy of judging the installation willingness of the user.

[0084] Further, for each visit of each historical APP, the expression of the utility function of the recommendation model can also include a sum of a third utility sub-function; wherein the third utility sub-function is the maximum value obtained by calculating the sum value of the first utility sub-function and the second utility sub-function for one or more APPs subsequently visited after the historical APP.

[0085] Further, the expression of the third utility sub-function can be:

[0086]

[0087] wherein ψ its' a first utility sub-function for representing the subsequent accessed APP of the historical APP, a transposed vector of the APP association relationship parameter for representing the subsequent accessed APP of the historical APP, an implicit attribute parameter for representing the kth other historical APP accessed before the historical APP within the preset time length, α s an implicit attribute parameter for representing the historical APP; j is used for representing the order of the current accessed time of the historical APP within the preset time length, k is a positive integer, and k≤j-1; i is used for representing the i th user, s is used for representing the APP s, s' is used for representing the APP s' which is the subsequent accessed APP of the historical APP, and t is used for representing the time of the current accessed time of the historical APP.

[0088] Specifically, the user can control to reduce or even cancel the use of the current APP due to the prediction of the use of the subsequent APP, for example, the average time length of each access to a game APP is 1 hour, and the access time length of the game APP is shortened to 15 minutes due to the prediction of the subsequent existence of the data flow and / or the screen time length limitation (for example, video call needs to be performed).

[0089] As can be seen from the above, for each access of each historical APP, the expression of the utility function of the recommendation model can be:

[0090]

[0091] In the embodiments of the present application, the third utility sub-function can also be constructed based on the APP accessed after the historical APP, so as to determine the influence of the other historical APP on the selection of the historical APP and the influence of the subsequent accessed APP on the access of the historical APP by using the implicit attribute parameters of the historical APP, the other historical APP accessed before and the subsequent accessed APP, and the association relationship parameters between the latter two and the historical APP, thereby more effectively judging the potential download habit of the user, forming a utility function more customized and more accurate for the user, and further improving the accuracy of judging the installation willingness of the user.

[0092] In the implementation of step S13, more users and longer preset time length can be used to estimate the plurality of parameter vectors of the recommendation model based on the posterior distribution formula.

[0093] Specifically, the plurality of parameter vectors of the recommendation model can be estimated based on the posterior distribution formula as follows:

[0094]

[0095] Ω represents a space set of the access attribute information, the implicit attribute parameters and the APP association relationship parameters, represents a set of historical APPs used by the plurality of users in the plurality of preset time length, i.e. x represents other influence factors, represents a given and x, the predicted access probability of the APP based on the Ω space; z it represents the historical APP accessed by the i-th user at the t time, x it represents other influence factors when the i-th user accesses the historical APP at the t time, p(z it |Ω,x it represents the predicted access probability of the historical APP accessed by the i-th user at the t time given the Ω space and x, and p(Ω) represents the prior distribution based on the Ω space. represents the predicted access probability of the set of historical APPs accessed by the i-th user at the t time based on the x condition.

[0096] In the implementation, a set of latent variables Ω={ρ,α,λ,θ,δ,μ,γ,β} can be inferred by the Bayesian method, and the APPs accessed by n users in T days can be observed, Other influence factors x can also be observed.

[0097] Further, the other influence factors can be selected from one or more of the following: data traffic usage m, APP category parameter c and period parameter d to which the access time belongs, wherein the other influence factors do not repeat the parameters contained in the access attribute information in the same estimation of the plurality of parameter vectors of the recommendation model.

[0098] In the embodiments of the application, by setting other influence factors, the number of parameters for analysis can be further increased, so that the access decision of the user to each historical APP can be more comprehensively analyzed, which helps to improve the prediction accuracy when predicting the preference of the user to the target APP subsequently.

[0099] Furthermore, the prior distribution p(Ω) can be determined by selecting from the Gaussian prior distribution and the GAMMA prior distribution.

[0100] In practice, an independent prior distribution can be set for each parameter.

[0101] As a non-restrictive example, for real-valued parameters, such as the latent attribute parameter α, the APP association parameter ρ, the APP download volume parameter λ, the user's preference parameter θ, the latent time coefficient δ, and the access time parameter μ, a Gaussian prior can be used; for positive-valued parameters, such as the user data traffic sensitivity parameter γ and the APP data traffic parameter β, a Gamma prior can be used, thereby effectively improving the prior effect.

[0102] Furthermore, the following formula can be used to determine each probability parameter:

[0103]

[0104]

[0105]

[0106] Where s, s', and s" are used to represent historical apps as APP s, APP s', and APP s", respectively, and Ψ(s,z) it,j-1 ) is used to represent the (j-1)th APP accessed by the i-th user at time t as APP s, Ψ(s',z) it,j-1 ) is used to represent the (j-1)th APP accessed by the i-th user at time t as APP s', p(z it,j =s|z it,j-1 ) is used to represent the probability that the (j-1)th APP accessed by the i-th user at time t is APP s.

[0107] This is used to represent a multiplication operation on the data of n users, including the i-th user. Used to represent a multiplication operation on data at T time points (e.g., T days), including time t.

[0108] For more information on the parameters in the above formula, please refer to the previous text and the expression of the utility function of the recommendation model; it will not be repeated here.

[0109] In the embodiment of the present application, more users and longer preset time length can be used to estimate the plurality of parameter vectors of the recommendation model based on the posterior distribution formula, so that the commonness of the potential downloading habits of more users can be determined by using big data processing technology, and the application range of the recommendation model in the embodiment of the present application is more extensive and more accurate.

[0110] In the implementation of step S14, a preset objective function is used to obtain a set of parameter vectors with the optimal fitting effect.

[0111] Further, the preset objective function can be:

[0112]

[0113] Wherein, L(v) is used to represent the preset objective function, is used to represent the joint density function after x is given, q(Ω; v) is used to represent the APP access probability based on the Ω space of the parameters randomly extracted from the v complex parameter space, E q(Ω；v) is used to represent the maximum evidence lower bound function.

[0114] It should be noted that, The product can be calculated by the above-determined and Other conventional methods can also be used to determine.

[0115] In the implementation, the mean-field family of q(Ω; v) can be specified, so that each latent variable is independent of each other and changes with the variational parameter. However, the latent variable can not be assumed to be the same distribution, so as to improve the universality of the recommendation model.

[0116] The inventor of the present application has found that it is difficult to directly derive the analytical form of the posterior due to the difficult-to-handle marginal distribution, denominator integral and softmax function of the large selection set. Therefore, the Bayesian approximation method and variational inference (VI) are combined. This is because VI is more efficient in big data processing.

[0117] Specifically, VI does not iteratively sample from the candidate distribution, but searches for the closest distribution in the same family of the expected distribution, and converts the sampling problem into an optimization problem, so that after a proper objective function is established, the Ω with the optimal fitting effect can be obtained according to the maximum evidence lower bound (ELBO).

[0118] In one non-limiting embodiment, after a distribution family of q(Ω; v) is specified, v that maximizes ELBOL(v) can be found by stochastic optimization, denoted as v * Then the posterior distribution can be approximated by optimizing q(Ω; v * ).

[0119] In embodiments of the present application, by selecting an appropriate objective function, after a distribution family of q(Ω; v) is specified, v that maximizes L(v) can be found by iterative optimization, i.e., a set of parameter vectors with the best fitting effect is obtained, further improving the accuracy of determining the user's preference for the target APP.

[0120] In the implementation of step S15, the set of parameter vectors with the best fitting effect can be used to determine the user's preference for the target APP.

[0121] Referring to Figure 2 , Figure 2 is Figure 1 a flowchart of one implementation of step S15 in FIG. 1. The target APP can be one or more APPs selected according to the request APP after receiving the user input request APP; the step of using the set of parameter vectors with the best fitting effect to determine the user's preference for the target APP can include steps S21 to S22, which are described below.

[0122] In step S21, if the product of the hidden attribute parameter of the target APP and the transposed vector of the APP association parameter of the request APP in the parameter vector with the best fitting effect is greater than zero, the target APP is determined to be a to-be-recommended APP based on the request APP.

[0123] In step S22, if the product of the hidden attribute parameter of the target APP and the transposed vector of the APP association parameter of the request APP in the parameter vector with the best fitting effect is less than zero, the target APP is determined to be a to-be-avoided APP based on the request APP.

[0124] Specifically, the product of the hidden attribute parameter a and the transposed vector p T of the APP association parameter can be used to represent the competition level between two APPs, and the formula is as follows:

[0125]

[0126] Wherein, S1 is used to represent the request APP, and S2 is used to represent the target APP, to represent the complementary relationship and the competitive relationship between S1 and S2.

[0127] It should be noted that when , the behavior of the user inputting the request APP increases the possibility of using the target APP, when increases, the complementary effect is enhanced; when , the behavior of the user inputting the request APP reduces the possibility of using the target APP, when increases, the substitution effect decreases with the increase of ; when is equal to 0, the behavior of the user inputting the request APP does not affect the use of the target APP.

[0128] In the embodiment of the application, after receiving the request APP of the user input, the target APP selected by the request APP can be determined as a to-be-recommended APP or a to-be-avoided APP according to the product of the implicit attribute parameter α and the transpose vector ρ T of the APP association relationship parameter of the request APP, so that the valuable recommendation resources can be implemented on the APP with higher installation willingness, and the recommendation-download conversion rate is improved, and the waste of the recommendation quota is avoided.

[0129] Specifically, the scheme of the embodiment of the application can help the APP owner to identify different levels of competitors of the APP according to the potential attributes of the APP, and the APP owner can review the "moat" of the product line by better understanding the competitors. Correspondingly, clearly understanding the competitors can also help the enterprise to strategically adjust the existing product design, so as to avoid the internal cannibalization of the product line and enhance the competitiveness of the products of other enterprises. In the long run, by better understanding the market structure of the APP industry, the APP owner can obtain more insights in strategic planning of new product design and positioning, and can also improve the pertinence and effectiveness of investment decisions.

[0130] Further, the scheme of the embodiment of the application can also help the platform such as the application store to evaluate the APP according to the potential attributes derived from the user usage data, improve the recommendation efficiency and the total download amount of the APP, and also improve the pertinence and effectiveness of the advertisement placement.

[0131] In the embodiment of the present application, by constructing a utility function, a plurality of parameters including the access attribute information, the implicit attribute parameter and the APP correlation relationship parameter are used to estimate a parameter vector, and then a set of parameter vectors with optimal fitting effect is obtained, so as to predict the preference of the user for the target APP, thereby through the big data processing technology, after judging the target APP in the APP database in advance, it is determined whether to recommend the user according to the installation willingness of the user for the target APP, and the recommendation-download conversion rate is effectively improved.

[0132] With reference to Figure 3 , Figure 3 is a structural schematic diagram of an application program APP preference determination device in the embodiment of the present application. The application program APP preference determination device can include:

[0133] An acquisition module 31 is configured to acquire access attribute information of one or more historical APPs of a user accessed within a preset time length;

[0134] An expression determination module 32 is configured to determine an expression of a utility function of a recommendation model, wherein the expression of the utility function includes the access attribute information, an implicit attribute parameter and an APP correlation relationship parameter;

[0135] An estimation module 33 is configured to estimate a plurality of sets of parameter vectors of the recommendation model, wherein each set of parameters includes the access attribute information, the implicit attribute parameter and the APP correlation relationship parameter;

[0136] A fitting module 34 is configured to obtain a set of parameter vectors with optimal fitting effect by using a preset objective function;

[0137] A preference determination module 35 is configured to determine the preference of the user for a target APP by using the set of parameter vectors with optimal fitting effect.

[0138] With reference to Figure 3 The working principle, working mode and beneficial effects of the APP preference determination device shown in the above description and the related description of the above Figures 1 to 2 , which will not be repeated here.

[0139] The embodiment of the present application further provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is run by a processor to execute the steps of the APP preference determination method described above. The storage medium can include ROM, RAM, magnetic disk or optical disk, etc. The storage medium can also include non-volatile memory or non-transitory memory, etc.

[0140] The embodiment of the present application further provides a terminal comprising a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor executes steps of the APP preference determination method when running the computer program. The terminal includes, but is not limited to, a mobile phone, a computer, a tablet computer, a server, a cloud platform and other terminal devices.

[0141] 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.

[0142] 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).

[0143] 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.

[0144] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can be physically included separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or in the form of hardware plus software functional units. For example, for various devices or products applied to or integrated into a chip, each module / unit contained therein can be implemented in hardware such as circuits, or at least some modules / units can be implemented in software programs that run on the processor integrated inside the chip, and the remaining (if any) modules / units can be implemented in hardware such as circuits; for various devices or products applied to or integrated into a chip module, each module / unit contained therein can be implemented in hardware such as circuits, and different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented in software programs that run on the processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in hardware such as circuits.

[0145] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article indicates that the preceding and following related objects have an "or" relationship.

[0146] In the embodiments of this application, "multiple" refers to two or more.

[0147] The descriptions of "first," "second," etc., appearing in the embodiments of this application are for illustrative purposes and to distinguish the objects being described. They have no order and do not indicate any special limitation on the number of devices in the embodiments of this application, nor do they constitute any limitation on the embodiments of this application.

[0148] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. An application program (APP) preference determination method, characterized by, The method comprises the following steps: obtaining access attribute information of one or more historical APPs of a user within a preset time length; determining an expression of a utility function of a recommendation model, wherein the expression of the utility function comprises the access attribute information, implicit attribute parameters and APP correlation parameters; estimating a plurality of sets of parameter vectors of the recommendation model, wherein each set of parameters comprises the access attribute information, the implicit attribute parameters and the APP correlation parameters; obtaining a set of parameter vectors with the optimal fitting effect by using a preset objective function; determining a preference of the user for a target APP by using the set of parameter vectors with the optimal fitting effect; wherein, for each visit of each historical APP, the expression of the utility function of the recommendation model comprises a sum of a first utility sub-function , a second utility sub-function and a third utility sub-function. The first utility sub-function is determined based on the access attribute information and the implicit attribute parameter of the historical APP. The second utility sub-function is determined based on the implicit attribute parameters of other historical APPs accessed before the historical APP within a preset time length and the association relationship parameters between the historical APP and each of the other historical APPs. The third utility sub-function is a maximum value obtained by calculating the sum of the first utility sub-function and the second utility sub-function for one or more APPs accessed subsequently to the historical APP. wherein the implicit attribute parameters are used to represent implicit factors affecting the selection of APPs by the user, and the APP correlation parameters are used to represent the correlation between APPs.

2. The method of claim 1, wherein, The target APP is one or more APPs selected according to a request APP received by the user; determining the preference of the user for the target APP by using the set of parameter vectors with the optimal fitting effect comprises: If the latent attribute parameters of the target APP are in the parameter vector with the best fitting effect, then... The transpose of the APP association parameter of the requesting APP If the product of the two is greater than zero, then the target APP is determined to be the APP to be recommended based on the requesting APP; If the latent attribute parameters of the target APP are in the parameter vector with the best fitting effect, then... The transpose of the APP association parameter of the requesting APP If the product of the two is less than zero, then the target APP is determined to be the APP to be avoided based on the requesting APP.

3. The method of claim 1, wherein, The access attribute information is selected from one or more of the following: APP download amount parameter , the user's preference parameter , potential time coefficient , access time parameter , user data traffic sensitivity parameter , APP data traffic parameter , data traffic usage amount , APP category parameter c and period parameter d to which the access time belongs.

4. The method of claim 1, wherein, The first utility sub-function The expression of the first utility sub-function includes a sum of one or more of the following: the historical app download amount parameter ; transposed vector of potential time coefficients a product with an access time parameter ;​ a transpose vector of data traffic sensitivity of the user in a preset time length , a data traffic parameter of the historical APP , a logarithm of per-minute data traffic usage of the historical APP in the preset time length of a product of the logarithm of per-minute data traffic usage of the historical APP in the preset time length ; wherein, a transposed vector for representing a preference parameter of the user a product of the explicit attribute parameter and the implicit attribute parameter of the history APP of the history APP. i represents the i-th user, s represents the historical APP s, and t represents the time when the historical APP is currently accessed.

5. The method of claim 1, wherein, The expression of the second utility sub-function is ; wherein, a transposed vector for representing an APP association relationship parameter of the history APP, an implicit attribute parameter for representing a kth other history APP accessed before the history APP within the preset time length; j represents the order of the current access of the historical APP within the preset time length, k is a positive integer, and k≤j-1. i represents the i-th user, s represents the historical APP s, and t represents the time when the historical APP is currently accessed.

6. The method of claim 1, wherein, The expression of the third utility sub-function is ; wherein, a first utility sub-function for representing a subsequent visited APP of the history APP, a transposed vector of an APP association parameter for representing a subsequent visited APP of the history APP, a latent attribute parameter for representing a kth other history APP visited before the history APP within the preset time length, a latent attribute parameter for representing the history APP; j represents the order of the current access of the historical APP within the preset time length, k is a positive integer, and k≤j-1. i represents the i-th user, s represents the historical APP s, and t represents the time when the historical APP is currently accessed.

7. The method of claim 1, wherein, The plurality of sets of parameter vectors of the recommendation model are estimated by using the following posterior distribution formula: ； wherein, a spatial set for representing the access attribute information, the implicit attribute parameters and the APP association parameters, a set for representing the historical APPs used by the plurality of users within a plurality of preset time lengths, i.e. x for representing other influencing factors, for representing the given and x, a predicted access probability of the APP based on the spatial set. a history APP visited by the i-th user at time t, other influencing factors for the i-th user visiting the history APP at time t, a predicted visiting probability of the i-th user visiting the history APP at time t given a space and x conditions, a prior distribution based on a space; Pijt(x) denotes the predicted probability of the ith user accessing the set of applications based on x condition access history at time t.

8. The method of claim 7, wherein, Each probability parameter is determined by using the following formula: ； ； ； wherein s, s', s" are used to represent the historical APPs as APP s, APP s', APP s", respectively, to represent the (j-1)th APP visited by the ith user at time t is APP s, to represent the (j-1)th APP visited by the ith user at time t is APP s', to represent the probability that the (j-1)th APP visited by the ith user at time t is APP s.

9. The method of claim 7, wherein, determining a prior distribution selected from the group consisting of: Gaussian prior distribution and GAMMA prior distribution.

10. The method of claim 7, wherein, The other influence factors are selected from one or more of the following: Data traffic usage , an APP category parameter c, and a period parameter d to which an access time belongs; In the process of estimating the plurality of sets of parameter vectors of the recommendation model at the same time, the other influence factors do not repeat the parameters contained in the access attribute information.

11. The method of claim 1, wherein, The preset objective function is: ； wherein, for representing the preset objective function, for representing the joint density function after given x, for representing the lower bound function of the evidence, the parameters randomly extracted from the complex parameter space are based on the APP access probability of the space, for representing the maximum evidence lower bound function.

12. An APP preference determining apparatus, characterized by comprising: The method comprises the following steps: an obtaining module, configured to obtain access attribute information of one or more historical APPs of a user within a preset time length; an expression determining module, configured to determine an expression of a utility function of a recommendation model, wherein the expression of the utility function comprises the access attribute information, implicit attribute parameters and APP correlation parameters; an estimating module, configured to estimate a plurality of sets of parameter vectors of the recommendation model, wherein each set of parameters comprises the access attribute information, the implicit attribute parameters and the APP correlation parameters; a fitting module, configured to obtain a set of parameter vectors with the optimal fitting effect by using a preset objective function; and a determining module, configured to determine a preference of the user for a target APP by using the set of parameter vectors with the optimal fitting effect. A preference determining module is configured to determine the user's preference for the target APP by using the set of parameter vectors with the optimal fitting effect. wherein, for each visit of each historical APP, the expression of the utility function of the recommendation model comprises a sum of a first utility sub-function , a second utility sub-function and a third utility sub-function. The first utility sub-function is determined based on the access attribute information and the implicit attribute parameter of the historical APP. The second utility sub-function is determined based on the implicit attribute parameters of other historical APPs accessed before the historical APP within a preset time length and the association relationship parameters between the historical APP and each of the other historical APPs. The third utility sub-function is a maximum value obtained by calculating the sum of the first utility sub-function and the second utility sub-function for one or more APPs accessed subsequently to the historical APP. The implicit attribute parameter is used to represent an implicit factor affecting the user's selection of the APP, and the APP association parameter is used to represent an association relationship between APPs.

13. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, performs the steps of the APP preference determining method of any one of claims 1 to 11.

14. 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, The processor, when executing the computer program, performs the steps of the APP preference determining method of any one of claims 1 to 11.

Images