KR102327029B1 - Method of recommending poi and recommendation system performing therof - Google Patents

Abstract

A POI (Point of Interest) recommendation method according to the present invention is provided. The POI recommendation method includes a POI information receiving step of receiving POI information associated with a POI transition pattern for each user, and a POI matrix generating a POI-POI matrix (Q) indicating a user's transition preference based on the POI information. generating, optimizing a target function associated with a preference for a POI using information associated with the POI-POI matrix Q and the user-POI matrix P, and calculating according to the optimized target function The method may include providing a POI candidate providing a POI candidate to each user terminal based on the selected preference for the POI.

Description

POI recommendation method and recommendation providing system performing the same

The present invention relates to a POI recommendation method and a recommendation providing system for performing the same, and more particularly, to a POI recommendation method in conjunction with a user terminal, a recommendation server for performing the same, and a recommendation providing system.

A recommendation system enables a user or consumer to find items of interest and recommended additional items.

In order to recommend an item, the recommendation provider must collect a “rating” for the item from each user. In addition, the recommendation engine of the recommendation provider needs to filter products based on the user's previous history.

However, the recommendation provider must infer personal information, ie, gender, age, disposition and preference, etc. according to the submitted grade. However, it can be cumbersome to collect ratings for each of these items. In addition, there is a problem in that it is impossible to accurately infer personal information, ie, gender, age, inclination and preference, etc. according to the collected grades.

Accordingly, an object of the present invention is to provide a recommendation method and a recommendation providing system capable of recommending an item without collecting ratings or ratings for each item, in order to solve the above problem.

Another object of the present invention is to provide a recommendation method and a recommendation providing system capable of providing a POI accurately and quickly by optimizing a target function related to a preference for a POI.

A POI (Point of Interest) recommendation method according to the present invention for solving the above problems is provided. The POI (Point of Interest) recommendation method includes a POI information receiving step of receiving POI information related to a POI transition pattern for each user, and a POI-POI matrix (Q) representing a user's transition preference based on the POI information. POI matrix generation step of generating; a target function optimization step of optimizing a target function related to a preference for a POI using information related to the POI-POI matrix Q and the user-POI matrix P, and the POI calculated according to the optimized target function The POI candidate providing step of providing a POI candidate to each user terminal based on the preference for .

A recommendation providing system for performing a POI (Point of Interest) recommendation method according to another aspect of the present invention is provided. The recommendation providing system includes: a recommendation server configured to receive POI information associated with a POI transition pattern for each user, and generate a POI-POI matrix (Q) indicating a user's transition preference based on the POI information; and a user terminal configured to transmit information associated with the user-POI matrix P to the recommendation server.

According to an embodiment, the recommendation server optimizes a target function related to a preference for a POI using information related to the POI-POI matrix Q and the user-POI matrix P, and the optimized target function It may be further configured to provide a POI candidate to the user terminal based on the calculated preference for the POI.

According to the present invention, it is possible to provide a recommendation method and a recommendation providing system capable of recommending an item by using only POI information associated with a POI transition pattern for each user without collecting a rating or rating for each item.

In addition, according to the present invention, by using information associated with the POI-POI matrix (Q) and the user-POI matrix (P), the objective function associated with the preference for the POI is optimized to provide a recommendation that can provide POI accurately and quickly A method and a system for providing a recommendation for performing the same may be provided.

The features and effects of the present invention described above will become more apparent through the following detailed description in relation to the accompanying drawings, and accordingly, those of ordinary skill in the art to which the present invention pertains can easily implement the technical idea of the present invention. will be able

1 shows a conceptual diagram related to a continuous (successive) POI (Point of Interest) recommendation related to the present invention.
2 is a conceptual diagram illustrating a user-POI matrix and a POI-POI matrix decomposition process according to the present invention.
3 is a flowchart of a POI recommendation method according to the present invention.
4 is a diagram illustrating a process in which a recommendation server and a user terminal perform the POI recommendation providing method of FIG. 3 using each user's transition pattern and number of visits according to the present invention.
5 shows a step of generating a POI matrix according to the present invention.
6 shows a transition pattern perturbation algorithm.
7 shows a gradient perturbation process using a profile vector between a user and a server according to the present invention.
8 shows a gradient perturbation algorithm using a piecewise mechanism.

The features and effects of the present invention described above will become more apparent through the following detailed description in relation to the accompanying drawings, and accordingly, those of ordinary skill in the art to which the present invention pertains can easily implement the technical idea of the present invention. will be able

Since the present invention can have various changes and can have various embodiments, specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present invention.

In describing each figure, like reference numerals are used for like elements.

Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. shouldn't

The suffix module, block, and part for the components used in the following description are given or mixed in consideration of only the ease of writing the specification, and do not have distinct meanings or roles by themselves.

Hereinafter, a preferred embodiment of the present invention will be described with reference to the accompanying drawings so that a person skilled in the art can easily implement it. In the following description of embodiments of the present invention, if it is determined that a detailed description of a related known function or a known configuration may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted.

Hereinafter, a method for providing a POI recommendation and a system for providing a recommendation for performing the method according to the present invention will be described.

가 주어진 경우, POI 추천 시스템은 사용자의 최근 체크 인 정보에 기반하여 다음 POI 후보들을 제공할 수 있다. 여기서, U는 사용자들의 집합(set)을 나타내고, L은 POI들의 집합을 나타내고,

는 사용자 u의 체크 인 히스토리를 나타낸다.In this regard, FIG. 1 shows a conceptual diagram related to a successive POI (Point of Interest) recommendation related to the present invention. 1 , check-in history

If , the POI recommendation system may provide the following POI candidates based on the user's recent check-in information. Here, U represents a set of users, L represents a set of POIs,

represents the check-in history of user u.

An assumption in the context of the present invention is that static user interest and transition preference share the same POI latent factor. Static user interest means personal preference for POI based on the number of visits. On the other hand, the transition preference can be approximated with a first-order Markov assumption about the user's sequential actions.

On the other hand, it is an object of the present invention to integrate static user interest and transition preference. In this regard, POI profile vectors are shared in a matrix factorization process.

Meanwhile, the transfer learning approach may be expressed by a user-POI matrix and a POI-POI matrix. In this regard, FIG. 2 shows a conceptual diagram of a user-POI matrix and a POI-POI matrix decomposition process according to the present invention.

Referring to FIG. 2 , the user-POI matrix P means user preference for POIs, and may be decomposed and approximated as ^{P = UV T .} Here, P denotes a static preference for all users, that is, the number of visits to each POI. In this case, the user-POI matrix (P) may be decomposed into ^{P = UV T} using the user latent matrix (U) and the POI latent matrix (V).

On the other hand, the POI-POI matrix Q means a transition preference, and ^{can be approximated by decomposing Q = VV T} . Here, Q means a transition preference for all users, that is, the number of preference transitions. In this case, the POI-POI matrix (Q) may be decomposed into ^{Q = VV T using two POI latent matrices (V).}

As described above, a POI (Point of Interest) recommendation method using the user-POI matrix (P) and the POI-POI matrix (Q) will be described as follows. In this regard, FIG. 3 shows a flowchart of a POI recommendation method according to the present invention. The POI recommendation method may be mainly performed by a recommendation server. However, in some processes of the POI recommendation method, the recommendation server may perform an optimization process by interworking and cooperation with the user terminal.

Referring to FIG. 3 , the POI recommendation method may include receiving POI information ( S100 ), generating a POI matrix ( S200 ), optimizing a target function ( S300 ), and providing a POI candidate ( S400 ).

In the POI information receiving step S100 , POI information associated with a POI transition pattern may be received for each user. In the POI matrix generation step S200 , a POI-POI matrix Q indicating a user's transition preference may be generated based on the POI information. Meanwhile, in the target function optimization step S300 , the target function related to the preference for the POI may be optimized using information related to the POI-POI matrix Q and the user-POI matrix P. In this case, the user-POI matrix (P) may be decomposed into ^{P = UV T} using the user latent matrix (U) and the POI latent matrix (V). Meanwhile, the POI-POI matrix (Q) may be decomposed into ^{Q = VV T using the POI latent matrix (V).}

Accordingly, in the POI candidate providing step S400 , the POI candidate may be provided to each user terminal based on the preference for the POI calculated according to the optimized target function. Meanwhile, the target function optimization step S300 may be repeated a predetermined number of times or may be repeatedly performed until a reference condition is satisfied. As an example, in the step of optimizing the target function ( S300 ), the optimization of the target function may be iteratively performed by Equation 1 below. That is, the user latent matrix (U) and the POI latent matrix (V) can be jointly optimized by using the loss function of Equation (1).

[Equation 1]

Here, preference may be calculated using Table 1 as follows using specific components of the user-POI matrix (P) and the POI-POI matrix (Q).

Meanwhile, FIG. 4 shows a process in which a recommendation server and a user terminal perform the POI recommendation providing method of FIG. 3 using each user's transition pattern and visit number according to the present invention. 3 and 4 , the POI matrix generation step S200 may be performed by the recommendation server 100 , and the objective function optimization step S300 may be performed by the user terminal 200 . Alternatively, some processes of the target function optimization step S300 may be performed by the recommendation server 100 , and the remaining processes may be performed by the user terminal 200 . Meanwhile, FIG. 5 shows a step of generating a POI matrix according to the present invention. In this regard, Fig. 6 shows a transition pattern perturbation algorithm. 5 and 6 , 1) a bit representation of a sampled transition pattern may be generated using 1) a check-in history of a user. In addition, 3) using the perturbed bit string, 4) a POI-POI matrix may be generated.

In this regard, each element of the POI-POI matrix reflects a global preference transition. The build process may be performed as follows.

1) Each user terminal samples a transition pattern from the check-in history.

2) Each user terminal ^{initializes n 2} bit-strings. In this regard, each bit represents a corresponding transition pattern, and only 1 bit can be set to 1.

3) Each bit can be perturbated through an optimized random response.

As described above, the POI information associated with the POI transition pattern may be a bit representation in which a sampled transition pattern based on a check-in history associated with a place visited by each user is expressed as a bit string. Also, the POI information associated with the POI transition pattern may be perturbation applied to the bit string. In this regard, in the POI matrix generation step S200 , the POI-POI matrix Q may be generated by aggregating POI information for each current POI.

According to an embodiment, after obtaining the noisy POI-POI matrix (Q), the next step is to combine the two matrices and decompose them as in Equation (2). In this regard, the decomposition process of Equation 2 corresponds to the joint optimization process of Equation 1.

[Equation 2]

Meanwhile, in the target function optimization step ( S300 ), the user profile vector (P) associated with the user latent matrix (U) may be optimized using Equation (3).

[Equation 3]

That is, in order to optimize the user profile vector P, the first optimization method, for example, an Alternative Least Square (ALS) method may be applied.

On the other hand, in the objective function optimization step S300 , the POI profile vector may be optimized using a second optimization method different from the first optimization method, for example, a stochastic gradient descent (SGD) method. In this regard, the POI latent matrix (V) may be updated using gradient perturbation. Equation 4 expresses the user profile vector and the POI profile vector through gradient perturbation for the target function.

[Equation 4]

In this regard, FIG. 7 shows a gradient perturbation process using a profile vector between a user and a server according to the present invention. 8 shows a gradient perturbation algorithm using a piecewise mechanism. Meanwhile, Table 2 shows the gradient perturbation process using the profile vector according to the present invention.

Referring to Table 2, FIGS. 7 and 8 , the recommendation server 100 may provide the POI latent matrix (V) to the user terminal 200 . Accordingly, the user terminal 200 may update the user latent matrix (U) by using the POI latent matrix (V). Also, the user terminal 200 may provide the i-th element of the perturbed gradient for each POIj to the recommendation server 100 . Accordingly, the recommendation server 100 may update the POI latent matrix (V).

Meanwhile, in the POI candidate providing step S400 , when the user ui is in the POI(vj), the preference for the next POI(vk) may be calculated as in Equation 5. Here, vj represents the profile vector of the current POI.

[Equation 5]

In order to recommend the next POI (vk), there is an advantage that the user does not need to provide his/her current location as a recommendation server in the present invention. That is, there is an advantage that each user (terminal) can locally calculate a preference for the next POI (vk).

Meanwhile, in relation to the above-described objective function optimization step S300, performance can be improved by 1) learning using user groups, 2) normalizing implicit feedback, and 3) accelerating the learning process as follows.

1) Learning with user group

- In the learning process, the update requires k iterations.

- Depending on the composition theorem, each user can perturb the gradient using the privacy budget at each iteration.

- In this regard, we divide all users into k groups, and each user can submit a perturbed gradient only once using the entire privacy budget.

2) normalizing implicit feedback

- The estimated frequency of the transition pattern is expressed as Equation 6 according to the number of users.

[Equation 6]

- In the present invention, the tracking frequency can be normalized to accurately reflect the transition pattern.

- You can use the sigmoid function to normalize the estimation frequency.

3) Accelerating learning process

- The user profile vector may be updated by ALS, and the POI profile vector may be updated by SGD.

- SGD may require more iterations to reach the optimum compared to ALS.

- Meanwhile, the Adam optimizer can be used to update the POI profile vector. In this regard, the Adam optimizer corresponds to a variant of the gradient descent method.

- Adam optimizer can maintain exponentially decaying averages of previous gradient m and squared gradient v as shown in Equation 7.

[Equation 7]

In the above, a method for providing a POI recommendation according to the present invention has been described. Hereinafter, a recommendation providing system for performing the POI recommendation providing method according to the present invention will be described. Meanwhile, the description of the POI recommendation providing method described above may be applied to the following recommendation providing system.

In this regard, referring to FIGS. 5 and 7 , a recommendation providing system for performing a POI (Point of Interest) recommendation method may include a recommendation server 100 and a plurality of user terminals 200 .

The recommendation server 100 may be configured to receive POI information associated with a POI transition pattern for each user, and generate a POI-POI matrix Q indicating a user's transition preference based on the POI information. Meanwhile, the user terminal 200 may be configured to transmit information related to the user-POI matrix P to the recommendation server 100 .

Meanwhile, the recommendation server 100 may be further configured to optimize the target function associated with the preference for the POI by using the information associated with the POI-POI matrix Q and the user-POI matrix P. Accordingly, the recommendation server 100 may be further configured to provide a POI candidate to the user terminal based on the preference for the POI calculated according to the optimized target function.

In this regard, optimizing the objective function associated with the preference for the POI using the information associated with the POI-POI matrix (Q) and the user-POI matrix (P) is performed by the recommendation server 100 and the user terminal 200 by the recommendation server 100 and the user terminal 200 . may be performed.

Meanwhile, the user-POI matrix (P) may be decomposed into ^{P = UV T} using the user latent matrix (U) and the POI latent matrix (V). In addition, the POI-POI matrix (Q) may be decomposed into ^{Q = VV T using the POI latent matrix (V).} In this case, the recommendation server 100 may repeatedly perform optimization on the target function by Equation (1).

On the other hand, the user terminal 200 may perform optimization of the user profile vector (P) associated with the user latent matrix (U) using Equation (3). On the other hand, the recommendation server 100 may optimize the POI profile vector using a stochastic gradient descent (SGD) method. Also, when the user ui is in the POI(vj), the recommendation server 100 may calculate the preference for the next POI(vk) by Equation 5.

In the above, a method for providing a POI recommendation according to the present invention and a recommendation system for providing the same have been described. Looking at the technical effects of the present invention are as follows.

According to the present invention, it is possible to provide a recommendation method and a recommendation providing system capable of recommending an item by using only POI information associated with a POI transition pattern for each user without collecting a rating or rating for each item.

In addition, according to the present invention, by using information associated with the POI-POI matrix (Q) and the user-POI matrix (P), the objective function associated with the preference for the POI is optimized to provide a recommendation that can provide POI accurately and quickly A method and a system for providing a recommendation for performing the same may be provided.

The features and effects of the present invention described above will become more apparent through the following detailed description in relation to the accompanying drawings, and accordingly, those of ordinary skill in the art to which the present invention pertains can easily implement the technical idea of the present invention. will be able

Since the present invention can have various changes and can have various embodiments, specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present invention.

According to the software implementation, not only the procedures and functions described in this specification but also the design and parameter optimization for each component may be implemented as a separate software module. The software code may be implemented as a software application written in a suitable programming language. The software code may be stored in a memory and executed by a controller or a processor.

Claims (10)

delete In the POI (Point of Interest) recommendation method, the method is performed by a recommendation server, the method comprising:
a POI information receiving step of receiving POI information associated with a POI transition pattern for each user;
a POI matrix generation step of generating a POI-POI matrix (Q) indicating a user's transition preference based on the POI information;
a target function optimization step of optimizing a target function related to a preference for a POI using information related to the POI-POI matrix (Q) and the user-POI matrix (P); and
a POI candidate providing step of providing a POI candidate to each user terminal based on the preference for the POI calculated according to the optimized objective function;
The user-POI matrix (P) is decomposed into ^{P = UV T} using a user latent matrix (U) and a POI latent matrix (V),
The POI-POI matrix (Q) is ^{decomposed into Q = VV T} using the POI latent matrix (V), POI recommendation method.
3. The method of claim 2,
In the objective function optimization step, the optimization of the objective function is

POI recommendation method, which is iteratively performed by
In the POI (Point of Interest) recommendation method, the method is performed by a recommendation server, the method comprising:
a POI information receiving step of receiving POI information associated with a POI transition pattern for each user;
a POI matrix generation step of generating a POI-POI matrix (Q) indicating a user's transition preference based on the POI information;
a target function optimization step of optimizing a target function related to a preference for a POI using information related to the POI-POI matrix (Q) and the user-POI matrix (P); and
a POI candidate providing step of providing a POI candidate to each user terminal based on the preference for the POI calculated according to the optimized objective function;
The POI information associated with the POI transition pattern includes:
A sampled transition pattern sampled based on a check-in history associated with a place visited by each user is a bit representation expressed as a bit string, characterized in that perturbation is applied to the bit string,
In the generating of the POI matrix, the POI-POI matrix (Q) is generated by aggregating the POI information for each current POI.
3. The method of claim 2,
In the objective function optimization step,

POI recommendation method, characterized in that the optimization of the user profile vector (P) associated with the user latent matrix (U) using
3. The method of claim 2,
The POI recommendation method, characterized in that in the objective function optimization step, a POI profile vector is optimized using a stochastic gradient descent (SGD) method.
3. The method of claim 2,
In the POI candidate providing step, if the user (ui) is in the POI (vj), the preference for the next POI (vk) is

POI recommendation method, characterized in that calculated as.
In the recommendation providing system for performing the POI (Point of Interest) recommendation method,
a recommendation server configured to receive POI information associated with a POI transition pattern for each user, and generate a POI-POI matrix (Q) indicating a user's transition preference based on the POI information; and
a user terminal configured to transmit information associated with a user-POI matrix (P) to the recommendation server;
The recommendation server is
optimizing a target function associated with a preference for a POI by using the information associated with the POI-POI matrix (Q) and the user-POI matrix (P);
and to provide a POI candidate to the user terminal based on the preference for the POI calculated according to the optimized target function,
The user-POI matrix (P) is decomposed into ^{P = UV T} using a user latent matrix (U) and a POI latent matrix (V),
The POI-POI matrix (Q) is ^{decomposed into Q = VV T} using the POI latent matrix (V), a recommendation providing system.
9. The method of claim 8,
The recommendation server is

Iteratively performs the optimization on the objective function by a recommendation providing system.
10. The method of claim 9,
The user terminal is

Optimizing the user profile vector (P) associated with the user latent matrix (U) using
The recommendation server is
Optimizing the POI profile vector using the stochastic gradient descent (SGD) method,
If the user (ui) is in POI (vj), the preference for the next POI (vk)

, which is a recommendation providing system.

Images