Method and system for generating interest point recommendation model and storage medium

Abstract

The invention provides a method and system for generating a point-of-interest recommendation model and a storage medium, and is used for improving the accuracy of point-of-interest recommendation, and the method comprises the steps: obtaining an order sample of a user; based on the order sample, determining a to-be-input feature; the to-be-input features comprise a user behavior sequence; the user behavior sequence comprises latitude and longitude hash values of the positions of the interest points in the order sample and weights of the interest points; and training a preset model based on the to-be-input features, and generating an interest point recommendation model.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, system and storage medium for generating a point of interest recommendation model. Background Art

[0002] Targeted content recommendation is a service that uses data analysis and algorithm technology to provide users with recommendations on products, points of interest or information that meet their unique needs based on their historical behavior, interest preferences, personal information, etc.

[0003] In order to achieve target content recommendation, one possible implementation method is to obtain implicit feedback of user interaction from the user's historical behavior data, sort the interaction data by timestamp to construct the user's related behavior sequence, and model based on the related behavior sequence to achieve target content recommendation for the user. However, the above method is only applicable to scenarios where user behavior is relatively frequent, such as the recommendation of goods or audio. In the freight scenario where the target content is points of interest, the points of interest recommended by the recommendation model established by the above method are not the same as the points of interest that the user actually wants to choose, and the recommendation effect is poor. Therefore, how to improve the accuracy of recommending points of interest has become a technical problem that needs to be solved at present. Summary of the invention

[0004] In view of the above problems, the purpose of this application is to provide a method, system and storage medium for generating a POI recommendation model to improve the accuracy of recommending POIs and enhance the recommendation effect of POIs. The specific scheme is as follows:

[0005] In a first aspect, an embodiment of the present application provides a method for generating a point of interest recommendation model, the method comprising:

[0006] Get the user's order sample;

[0007] Based on the order sample, determining features to be input; the features to be input include a user behavior sequence; the user behavior sequence includes a hash value of the longitude and latitude of each point of interest in the order sample, and a weight of each point of interest;

[0008] A preset model is trained based on the features to be input to generate a point of interest recommendation model.

[0009] Optionally, obtaining the user's order sample includes:

[0010] Positive samples and negative samples are respectively extracted from the order log of the user to obtain the order samples; the positive samples are the orders of the user on that day; and the negative samples are the orders of the user before that day.

[0011] Optionally, the method further includes:

[0012] Construct a preset model; the preset model includes an input layer, an embedding layer, an output layer, a factor decomposition machine layer, and a deep neural network layer.

[0013] Optionally, based on the order sample, determining the features to be input includes:

[0014] Setting the weight of the point of interest based on the frequency of occurrence of the point of interest in the order sample;

[0015] Extracting point of interest features and context features from the order sample;

[0016] User characteristics are determined based on the weights of the interest points and the order samples.

[0017] Optionally, the determining the user characteristics based on the weight of the point of interest and the order sample includes:

[0018] Obtaining the longitude and latitude hash values ​​of the locations of each point of interest in the order sample;

[0019] Constructing the user behavior sequence based on the interest point weight and the longitude and latitude hash value;

[0020] extracting user information from the order sample;

[0021] The user characteristics are determined based on the user information and the user behavior sequence.

[0022] Optionally, extracting user information from the order sample includes:

[0023] User identification information, user age information, user gender information and consumption frequency information are extracted from the order sample.

[0024] Optionally, extracting point of interest features and context features from the order sample includes:

[0025] Extracting the point of interest category information, point of interest identifier, point of interest heat information and longitude and latitude hash values ​​of the point of interest from the order sample to obtain the point of interest feature;

[0026] The time information and the location point information are extracted from the order sample to obtain the context feature.

[0027] In a second aspect, an embodiment of the present application provides a system for generating a point of interest recommendation model, the system comprising:

[0028] The acquisition module is used to obtain the user's order samples;

[0029] A determination module, configured to determine features to be input based on the order sample; the features to be input include a user behavior sequence; the user behavior sequence includes a hash value of the longitude and latitude of each point of interest in the order sample, and a weight of each point of interest;

[0030] A generation module is used to train a preset model based on the features to be input to generate a point of interest recommendation model.

[0031] Optionally, the determining module is specifically used to:

[0032] Setting the weight of the point of interest based on the frequency of occurrence of the point of interest in the order sample;

[0033] Extracting point of interest features and context features from the order sample;

[0034] User characteristics are determined based on the weights of the interest points and the order samples.

[0035] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the above-described methods for generating a point of interest recommendation model.

[0036] Compared with the prior art, this application has the following beneficial effects:

[0037] By obtaining the user's order samples, the features to be input are determined based on the order samples. The preset model is trained based on the features to be input to generate a point of interest recommendation model. Since the features to be input include the user behavior sequence, the user behavior sequence includes the longitude and latitude hash values ​​of the locations of each point of interest in the order sample, and the weight of each point of interest, the longitude and latitude hash values ​​are set in the user behavior sequence to avoid the long tail problem of the user behavior sequence due to the extremely sparse user behavior on the point of interest, and the amount of information obtained from the order sample is increased by setting weights for the points of interest, so that the accuracy of the recommended points of interest generated by the point of interest recommendation model trained based on the user behavior sequence is improved, thereby improving the recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0039] Figure 1 A flowchart of a method for generating a point of interest recommendation model provided in an embodiment of the present application;

[0040] Figure 2 A schematic diagram of the structure of the preset model provided in the embodiment of the present application;

[0041] Figure 3 A schematic diagram of the structure of a system for generating a point of interest recommendation model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0043] The terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0044] This application takes the freight scenario as an example. The current modeling of POI recommendation in the freight scenario follows the modeling method of related behavior sequence of commodity or audio categories. However, there is a large difference between the POI recommendation results and the POI actually selected by the user, and the recommendation accuracy is low and the effect is poor. Data analysis found that due to the extremely sparse behavior of users on POIs in the freight scenario, the recommendation accuracy is low. To solve this technical problem, this application provides a method, system and storage medium for generating a POI recommendation model, as follows.

[0045] like Figure 1 As shown, the embodiment of the present application provides a method for generating a point of interest recommendation model, the method comprising:

[0046] S101: Obtaining a user's order sample.

[0047] Extract the user's order samples from the user's order log. The order samples are the user's historical orders. The historical orders include user information and order-related information, such as the time when the order was generated, points of interest, the user's location when the order was placed, the user's age, and other information.

[0048] In an optional embodiment, obtaining the user's order sample includes:

[0049] Positive samples and negative samples are extracted from the user's order log to obtain order samples; the positive samples are the user's orders on the day; the negative samples are the user's orders before the day.

[0050] All order samples of the user are obtained from the user's order log, and the order information placed before the current date is used as a negative sample, and the order information placed on the current date is used as a positive sample, so as to divide the order information and obtain order samples including positive samples and negative samples.

[0051] For example, if the current date is May 2, 2022, and the historical order information includes Order A placed on September 12, 2020, Order B placed on July 1, 2021, and Order C placed on May 2, 2022, since the order dates of Order A and Order B are earlier than the current time, and the order date of Order C is the same as the current date, Order A and Order B are negative samples, and Order C is a positive sample.

[0052] S102: Based on the order sample, determine the features to be input; the features to be input include a user behavior sequence; the user behavior sequence includes the longitude and latitude hash values ​​of the locations of each point of interest in the order sample, and the weight of each point of interest.

[0053] Extract the input features from the order samples, which are the features of the order samples, including user features, point of interest features, and context features. User features include user behavior sequences.

[0054] It should be noted that, when the order samples include positive samples and negative samples, the features to be input include the features to be input of the positive samples and the features to be input of the negative samples.

[0055] In an optional embodiment, determining the features to be input based on the order sample includes:

[0056] Set the weight of the points of interest based on the frequency of occurrence of the points of interest in the order samples;

[0057] Extracting point of interest features and context features from order samples;

[0058] Based on the weights of points of interest and order samples, user characteristics are determined.

[0059] Specifically, the unique identifier (IdentityDocument, ID) of each point of interest that appears in the user's order sample, the number of times each point of interest appears, and the total number of points of interest that appear. The total number of points of interest that appear is the sum of the number of times each point of interest appears. The ratio of the number of times each point of interest appears to the total number of points of interest is calculated to obtain the frequency of occurrence of each point of interest. The frequency of occurrence of each point of interest is used as the weight corresponding to the ID of the point of interest. Thus, by setting weights for points of interest, the amount of information extracted from the user's behavior is increased.

[0060] For example, when the points of interest appearing in the historical order information include point of interest 1 with ID 1, point of interest 2 with ID 2, and point of interest 3 with ID 3, if point of interest 1 appears 2 times, and points of interest 2 and 3 each appear once, the total number of points of interest appearing is 4 times. The frequencies of occurrence of points of interest 1, 2, and 3 are 0.5, 0.25, and 0.25, respectively, so the weight of point of interest 1 is 0.5, and the weights of points of interest 2 and 3 are both 0.25. At this time, the IDs of the points of interest and the weights of the points of interest can be expressed as [1:0.5, 2:0.25, 3:0.25].

[0061] Furthermore, based on the weights of the points of interest and the order samples, the user characteristics are determined to include:

[0062] Get the longitude and latitude hash values ​​of each point of interest in the order sample;

[0063] Construct user behavior sequence based on interest point weights and longitude and latitude hash values;

[0064] Extract user information from order samples;

[0065] Determine user characteristics based on user information and user behavior sequences.

[0066] Obtain the location of each point of interest in the order sample, and convert the location of each point of interest into longitude and latitude information based on a map or other application with location and longitude and latitude conversion functions to obtain the longitude and latitude information of each point of interest. Based on the longitude and latitude information of each point of interest, calculate the longitude and latitude hash value of each point of interest, for example, convert the longitude and latitude information of each point of interest into a longitude and latitude hash value through geohash.

[0067] The weight of each point of interest in the order sample and the longitude and latitude hash value of each point of interest are input into the user behavior sequence template to obtain the user behavior sequence. For example, if the user behavior sequence template is [longitude and latitude hash value of the first point of interest: weight of the first point of interest, longitude and latitude hash value of the second point of interest: weight of the second point of interest…longitude and latitude hash value of the nth point of interest: weight of the nth point of interest], then when the order sample includes points of interest 4, 5 and 6, and the longitude and latitude hash value of point of interest 4 is geohash1, the longitude and latitude hash value of point of interest 5 is geohash2, the longitude and latitude hash value of point of interest 3 is geohash3, the weight of point of interest 4 is 0.6, the weight of point of interest 5 is 0.3, and the weight of point of interest 6 is 0.1, the user behavior sequence is [geohash1: 0.6, geohash2: 0.3, geohash3: 0.1]. Thus, by constructing a user behavior sequence based on the longitude and latitude hash values ​​of the points of interest and the weights of the points of interest, the situation in which user behavior is sparse and the long tail problem is serious in the freight scenario is improved.

[0068] It should be noted that the order of arrangement of each interest point in the user behavior sequence can be selected based on actual needs, such as arranging the interest points from early to late according to their appearance time, or arranging them from large to small according to their weights.

[0069] Specifically, extracting user information from order samples includes:

[0070] User identification information, user age information, user gender information and consumption frequency information are extracted from the order samples.

[0071] The user identification information is the user ID, which is used to indicate a unique user. The consumption frequency information is the frequency of the user's order placement within a preset time period, that is, the consumption frequency information is the number of orders placed by the user within the preset time period. The preset time period can be set based on actual needs, such as 1 month, half a year, or from the creation of the user account to the current time.

[0072] It should be noted that, since the user feature includes user information and user behavior sequence, after the user information and the user behavior sequence are respectively acquired, the user information and the user behavior sequence may be combined to obtain the user feature.

[0073] Specifically, extracting point of interest features and context features from order samples includes:

[0074] Extracting the category information, identification, heat information and longitude and latitude hash values ​​of the points of interest from the order samples to obtain the features of the points of interest;

[0075] Extract time information and location information from order samples to obtain context features.

[0076] The POI features include POI category information, POI identifiers, POI heat information, and latitude and longitude hash values ​​of each POI appearing in the order sample. The POI heat information is used to indicate the search heat of the POI at the current time.

[0077] The context features include time information and location information. The time information is used to indicate the time when the user places an order, and the location information is used to indicate the user's location when placing an order.

[0078] Extracting the latitude and longitude hash values ​​of the points of interest from the order samples includes: extracting the location of each point of interest from the order samples, and converting the location of each point of interest into longitude and latitude, generating the latitude and longitude hash values ​​of the points of interest based on the longitude and latitude of each point of interest. The specific method of obtaining the latitude and longitude hash values ​​of the points of interest can refer to the description related to the latitude and longitude hash values ​​of the points of interest in the above step S102.

[0079] S103: Training a preset model based on the features to be input to generate a POI recommendation model.

[0080] The features to be input determined based on the order samples are input into a preset model for training to obtain a POI recommendation model. When the order samples include positive samples and negative samples, the features to be input include positive sample features to be input and negative sample input features.

[0081] For example, when the order sample includes positive samples and negative samples, the weights of each point of interest in the order sample are extracted from the positive sample and the negative sample as a whole. Then, the latitude and longitude hash values ​​of each point of interest in the positive sample are determined based on the positive sample, and the user behavior sequence of the positive sample is generated by combining the weights of each point of interest in the positive sample. The user behavior sequence of the positive sample, the user identification information, the user age information, the user gender information and the consumption frequency information extracted from the positive sample are used as the user features of the positive sample. The point of interest features and the context features are extracted from the positive sample respectively, and the point of interest features and the context features extracted from the positive sample and the user features of the positive sample are used as the features to be input for the positive sample.

[0082] Based on the negative samples, the longitude and latitude hash values ​​of each point of interest in the negative samples are determined to generate the user behavior sequence of the negative samples in combination with the weights of each point of interest in the negative samples. The user behavior sequence of the negative samples, the user identification information, the user age information, the user gender information and the consumption frequency information extracted from the negative samples are used as the user features of the negative samples. The point of interest features and the context features extracted from the negative samples are respectively extracted, and the point of interest features and the context features extracted from the negative samples and the user features of the negative samples are used as the features to be input for the negative samples.

[0083] By setting longitude and latitude hash values ​​in the user behavior sequence, the long-tail problem of the user behavior sequence can be avoided due to the extremely sparse user behavior on points of interest. By setting weights for points of interest, the amount of information obtained from order samples is increased, so that the accuracy of the recommended points of interest generated by the point of interest recommendation model trained based on the user behavior sequence is improved, thereby improving the recommendation effect.

[0084] In an optional embodiment, before training the preset model based on the features to be input, the method further includes:

[0085] Build a preset model; the preset model includes input layer, embedding layer, output layer, factor decomposition machine layer, and deep neural network layer.

[0086] The structure of the preset model is as follows Figure 2As shown, the input layer is linked to the embedding layer, the embedding layer is connected to the deep factorization machine layer (Deep Factorization Machine, DeepFM) and the deep neural network layer (Deep Neural Network, DNN) respectively, and the DeepFM layer and the DNN layer are both connected to the output layer.

[0087] The embedding layer is used to convert the input features to the preset model into a 32-dimensional vector, so that the embedding layer outputs a vector of m*128 dimensions, where m is the total number of features contained in the input features. Taking the user behavior sequence as an example, when the user behavior sequence is [1:0.66, 2:0.33], the longitude and latitude hash values ​​of the points of interest are the mapping indexes of the points of interest, that is, the number of mapping points is 1 and 2 respectively. Construct a weight matrix x of a*128 dimensions, where a is the total number of all determined longitude and latitude hash values. When the mapping index of the point of interest is 1, the data in the first row of the weight matrix x is calculated, and the obtained value is determined as the converted 1*128 dimensional vector. That is, the value of x[1]*0.66+x[2]*0.33 is used as a 1*128 dimensional vector.

[0088] The input of the DeepFM layer is the user features and POI features converted to 128-dimensional vectors. The output layer is used to output the model score, model score = Yfm + Ydnn. Among them, Ydnn is the structured output of the three-layer DNN layer, for example [128, 64, 1]. Yfm is the target variable value predicted by the DeepFM layer. Yfm is determined based on the first-order cross and second-order cross features, and both Yfm and Ydnn are real numbers. Yfm = <ω, V j >xj1×xj2, where ω is the weight vector of the linear term, d is the number of input features, V is the factor vector, V i is the hidden vector of the i-th dimension feature, V j is the latent vector of the j-th dimension feature, <, > is the vector dot product, xj1 is the j1-th element in the weight matrix, and xj2 is the j2-th element in the weight matrix. The loss function value to be minimized is calculated as the cross entropy.

[0089] By obtaining the user's order samples, the features to be input are determined based on the order samples. The preset model is trained based on the features to be input to generate a point of interest recommendation model. Since the features to be input include the user behavior sequence, the user behavior sequence includes the longitude and latitude hash values ​​of the locations of each point of interest in the order sample, and the weight of each point of interest, the longitude and latitude hash values ​​are set in the user behavior sequence to avoid the long tail problem of the user behavior sequence due to the extremely sparse user behavior on the point of interest, and the amount of information obtained from the order sample is increased by setting weights for the points of interest, so that the accuracy of the recommended points of interest generated by the point of interest recommendation model trained based on the user behavior sequence is improved, thereby improving the recommendation effect.

[0090] like Figure 3 As shown, an embodiment of the present application provides a system for generating a point of interest recommendation model, the system comprising:

[0091] The acquisition module 301 is used to acquire the order sample of the user;

[0092] The determination module 302 is used to determine the features to be input based on the order sample; the features to be input include a user behavior sequence; the user behavior sequence includes the longitude and latitude hash values ​​of the locations of each point of interest in the order sample, and the weight of each point of interest;

[0093] The generation module 303 is used to train a preset model based on the features to be input and generate a point of interest recommendation model.

[0094] In an optional embodiment, the acquisition module 301 is specifically used for:

[0095] Positive samples and negative samples are extracted from the user's order log to obtain order samples; the positive samples are the user's orders on the day; the negative samples are the user's orders before the day.

[0096] In an optional embodiment, the above system further includes:

[0097] The building module is used to build a preset model; the preset model includes an input layer, an embedding layer, an output layer, a factor decomposition machine layer, and a deep neural network layer.

[0098] In an optional embodiment, the determination module 302 is specifically configured to:

[0099] Set the weight of the points of interest based on the frequency of occurrence of the points of interest in the order samples;

[0100] Extracting point of interest features and context features from order samples;

[0101] Based on the weights of points of interest and order samples, user characteristics are determined.

[0102] In an optional embodiment, the determination module 302 is specifically configured to:

[0103] Get the longitude and latitude hash values ​​of each point of interest in the order sample;

[0104] Construct user behavior sequence based on interest point weights and longitude and latitude hash values;

[0105] Extract user information from order samples;

[0106] Determine user characteristics based on user information and user behavior sequences.

[0107] In an optional embodiment, the determination module 302 is specifically configured to:

[0108] User identification information, user age information, user gender information and consumption frequency information are extracted from the order samples.

[0109] In an optional embodiment, the determination module 302 is specifically configured to:

[0110] Extracting the category information, identification, heat information and longitude and latitude hash values ​​of the points of interest from the order samples to obtain the features of the points of interest;

[0111] Extract time information and location information from order samples to obtain context features.

[0112] By obtaining the user's order samples, the features to be input are determined based on the order samples. The preset model is trained based on the features to be input to generate a point of interest recommendation model. Since the features to be input include the user behavior sequence, the user behavior sequence includes the longitude and latitude hash values ​​of the locations of each point of interest in the order sample, and the weight of each point of interest, the longitude and latitude hash values ​​are set in the user behavior sequence to avoid the long tail problem of the user behavior sequence due to the extremely sparse user behavior on the point of interest, and the amount of information obtained from the order sample is increased by setting weights for the points of interest, so that the accuracy of the recommended points of interest generated by the point of interest recommendation model trained based on the user behavior sequence is improved, thereby improving the recommendation effect.

[0113] An embodiment of the present application provides a computer storage medium for storing a computer program. When the computer program is executed, it is used to implement any of the above-mentioned methods for generating a point of interest recommendation model.

[0114] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0115] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0116] The technical content provided by the present invention is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for general technical personnel in the field, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for generating a point of interest recommendation model, characterized in that: The method comprises: Get the user's order sample; Based on the order sample, determining features to be input; the features to be input include a user behavior sequence; the user behavior sequence includes a hash value of the longitude and latitude of each point of interest in the order sample, and a weight of each point of interest; A preset model is trained based on the features to be input to generate a point of interest recommendation model.

2. The method according to claim 1, characterized in that The obtaining of the user's order sample comprises: Positive samples and negative samples are respectively extracted from the order log of the user to obtain the order samples; the positive samples are the orders of the user on that day; and the negative samples are the orders of the user before that day.

3. The method according to claim 1, characterized in that The method further comprises: Construct a preset model; the preset model includes an input layer, an embedding layer, an output layer, a factor decomposition machine layer, and a deep neural network layer.

4. The method according to claim 1, characterized in that: Based on the order sample, the features to be input are determined to include: Setting the weight of the point of interest based on the frequency of occurrence of the point of interest in the order sample; Extracting point of interest features and context features from the order sample; User characteristics are determined based on the weights of the interest points and the order samples.

5. The method according to claim 4, characterized in that The determining of user characteristics based on the weight of the point of interest and the order sample comprises: Obtaining the longitude and latitude hash values ​​of the locations of each point of interest in the order sample; Constructing the user behavior sequence based on the interest point weight and the longitude and latitude hash value; extracting user information from the order sample; The user characteristics are determined based on the user information and the user behavior sequence.

6. The method according to claim 5, characterized in that The extracting user information from the order sample comprises: User identification information, user age information, user gender information and consumption frequency information are extracted from the order sample.

7. The method according to claim 4, characterized in that The extracting of interest point features and context features from the order sample comprises: Extracting the point of interest category information, point of interest identifier, point of interest heat information and longitude and latitude hash values ​​of the point of interest from the order sample to obtain the point of interest feature; The time information and the location point information are extracted from the order sample to obtain the context feature.

8. A system for generating a point of interest recommendation model, characterized in that: The system comprises: The acquisition module is used to obtain the user's order samples; A determination module, configured to determine features to be input based on the order sample; the features to be input include a user behavior sequence; the user behavior sequence includes a hash value of the longitude and latitude of each point of interest in the order sample, and a weight of each point of interest; A generation module is used to train a preset model based on the features to be input to generate a point of interest recommendation model.

9. The method according to claim 8, characterized in that The determination module is specifically used for: Setting the weight of the point of interest based on the frequency of occurrence of the point of interest in the order sample; Extracting point of interest features and context features from the order sample; User characteristics are determined based on the weights of the interest points and the order samples.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the method for generating a point of interest recommendation model as described in any one of claims 1 to 7 is implemented.

Images