Continuous interest point recommendation method and system based on meta-learning

Abstract

The invention discloses a meta-learning-based continuous interest point recommendation method and system, and the method comprises the steps: obtaining the sign-in data of each user, and obtaining the preference of the user based on a self-attention mechanism; mapping the positions of all the interest points into the corresponding sub-block regions by using a coding algorithm, taking sign-in data in the sub-block regions nearby within a preset range of the current position of the user, and obtaining region preference based on a self-attention mechanism; splicing the user preferences and the region preferences to obtain comprehensive preferences, and calculating the probability that each interest point is asked by using a softmax function; and carrying out loss optimization training by utilizing meta-learning to obtain a trained continuous interest point recommendation model to carry out interest point recommendation. Through a coding algorithm, the interest point is mapped into the region, and the sign-in data of the region is used as the training data of preference modeling, so that the data sparsity is slowed down to a certain extent. And user preferences and regional preferences are combined, and user features and regional features are comprehensively considered, so that the accuracy of point-of-interest recommendation can be further improved.

Description

Technical Field

[0001] The present invention relates to the field of recommendation systems, and in particular to a continuous point of interest recommendation method and system based on meta-learning. Background Art

[0002] As the user base of location-based social networking software grows, users record and share their locations (i.e., points of interest) on social networks. Personalized recommendation systems learn from a large number of users' check-in records, accurately capture user preferences, and recommend points of interest to users to improve their experience. Depending on whether the most recent check-in data is paid more attention to, the point of interest recommendation task can be divided into two types: traditional point of interest recommendation and continuous point of interest recommendation. Compared with traditional point of interest recommendation, continuous point of interest recommendation not only needs to model user preferences, but also needs to pay attention to the user's recent location and provide user-satisfied recommendations in a timely manner.

[0003] Although POI recommendation has developed rapidly in recent years, it still faces the following challenges:

[0004] (1) Compared with low-cost recommendations such as music and movie recommendations, which can be achieved by "checking in" with just a click, checking in at points of interest requires higher costs in terms of time and money, resulting in greater sparsity in users' historical check-in data.

[0005] (2) A large number of new users and new POIs lack historical interaction data, and the POI recommendation task faces the cold start problem.

[0006] (3) Users have different preferences in different scenarios. Their choices are greatly affected by time, space, etc., and the preferences of different users vary greatly.

[0007] Therefore, the existing POI recommendation methods cannot accurately recommend POIs to users. Summary of the invention

[0008] Purpose of the invention: In view of the above shortcomings, the present invention provides a continuous POI recommendation method and system based on meta-learning to improve the accuracy of POI recommendation.

[0009] Technical solution: To solve the above problems, the present invention adopts a continuous POI recommendation method based on meta-learning, which includes the following steps:

[0010] Get the check-in data of each user, and obtain the user preferences of each user based on the self-attention mechanism;

[0011] The GeoHash encoding algorithm is used to map the locations of all points of interest to the corresponding sub-block areas. For each user, the check-in data in the sub-block areas near the preset range of the user's current location is obtained, and the regional preference of each user is obtained based on the self-attention mechanism;

[0012] The user preference and the area preference are concatenated to obtain the comprehensive preference of each user, and the softmax function is used to calculate the probability of each point of interest being visited by the user next.

[0013] Meta-learning is used to perform loss optimization training to obtain a trained continuous POI recommendation model.

[0014] Recommend points of interest to users through the trained continuous point of interest recommendation model.

[0015] Furthermore, the user u in the check-in data i The sign-in record is represented as:

[0016]

[0017] Among them, S′(u i ) represents user u i Check-in record, u i Indicates the user ID, The data representing the point of interest at the time of the user's nth check-in, t n Indicates the user's nth check-in time. The data of the point of interest includes the point of interest ID and the latitude and longitude of the point of interest.

[0018] The user's check-in records are mapped into trainable vector representations through a multimodal embedding layer, and the trainable vector representations mapped by the check-in records are concatenated to obtain the user's check-in trajectory matrix. The self-attention mechanism is used to reallocate weights for the user's check-in trajectory matrix to obtain user preferences.

[0019] Further, the check-in record of the check-in data in the sub-block area near the preset range of the user's current location is expressed as:

[0020] H′(u i )={(u α ,a α ,t α ),(u β ,a β ,t β ),...,(u k ,a k ,t k )}

[0021] Among them, H′(u i ) represents user u i The check-in record closest to the current location in the sub-block area; u α represents the user ID in the check-in record of the αth sub-block area that is closest to the present time; a αrepresents the code of the interest point in the αth sub-block area; t α The time in the sign-in record of the αth sub-block area that is closest to the present time;

[0022] The sign-in record H′(u i ) is mapped into a trainable vector representation through a multimodal embedding layer, and the trainable vector representation mapped by the check-in record is concatenated to obtain the user's regional trajectory matrix. The self-attention mechanism is used to reallocate the weights of the user's regional trajectory matrix to obtain the regional preference.

[0023] Furthermore, the probability calculation formula of each point of interest being visited by the user next time is as follows:

[0024]

[0025] Among them, E(v) is the trainable projection matrix of all interest points, P(u) is the user preference, R(u) is the region preference, Operator for concatenating user preferences and locale preferences.

[0026] Furthermore, the use of meta-learning for loss optimization training is specifically as follows: the check-in data of each user is divided into a support set and a query set, the data of the support set is used for local training, and its optimization goal is to minimize the recommendation loss of a single user; after local training on the support set, the global parameters are updated according to the loss on the query set, and the optimization goal of the global parameter update is to minimize the user's training loss on the query set.

[0027] The present invention also adopts a continuous point of interest recommendation system based on meta-learning, comprising:

[0028] The user preference determination module is used to obtain the check-in data of each user and obtain the user preference of each user based on the self-attention mechanism;

[0029] The regional preference determination module is used to map the locations of all points of interest to the corresponding sub-block areas using the GeoHash encoding algorithm. For each user, the check-in data in the sub-block areas near the preset range of the user's current location is obtained, and the regional preference of each user is obtained based on the self-attention mechanism;

[0030] The POI prediction module is used to combine user preferences and region preferences to obtain the comprehensive preferences of each user, and use the softmax function to calculate the probability of each POI being visited by the user next.

[0031] The meta-learning parameter update module is used to perform loss optimization training using meta-learning to obtain a trained continuous POI recommendation model;

[0032] The execution module is used to recommend points of interest to users through the trained continuous point of interest recommendation model.

[0033] The present invention also adopts a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0034] The present invention also adopts a computer-readable storage medium on which a computer program is stored, and the computer program implements the steps of the above method when executed by a processor.

[0035] Beneficial effects: Compared with the prior art, the present invention has the significant advantage of mapping points of interest to regions through the GeoHash encoding algorithm, and using the check-in data of the region as training data for preference modeling, which alleviates data sparsity to a certain extent. Combining user preferences with regional preferences, comprehensively considering user characteristics and regional characteristics, can further improve the accuracy of point of interest recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flowchart of the continuous point of interest recommendation method in the present invention.

[0037] Figure 2 A schematic diagram of the process of user preference modeling in the present invention.

[0038] Figure 3 A schematic diagram of the process of regional preference modeling in the present invention.

[0039] Figure 4 It is a schematic diagram of the process of predicting points of interest in the present invention.

[0040] Figure 5 Schematic diagram of the process of updating meta-learning parameters in the present invention. DETAILED DESCRIPTION

[0041] Example 1

[0042] like Figure 1 As shown, in this embodiment, a continuous interest point recommendation method based on meta-learning includes the following steps:

[0043] Step 1: Get the check-in data of each user, and use the self-attention mechanism to learn the user preferences for each user based on the check-in data. Point of interest: Point of interest, POI, is usually a geographical location, which can be a library, shopping mall, company, etc. The user check-in data includes the point of interest ID and the latitude and longitude information of the point of interest.

[0044] Self-attention mechanism: Self-attention mechanism is a technology widely used in fields such as natural language processing and computer vision. It allows the model to automatically learn the correlation between different positions when processing sequence data, thereby capturing richer contextual information.

[0045] Specifically, the check-in data contains the user set U = {u1,u2,...,u |U|}, the set of interest points V = {v1,v2,...,v |V|}, user u i The check-in record is represented as S′(u i )={(u i ,v1,t1),(u i ,v2,t2),...,(u i ,v n ,t n )},u i Indicates the user ID, v n The data representing the point of interest at the time of the user's nth check-in, t n Indicates the nth check-in time of the user. The dimensions of user and point of interest are |U| and |V| respectively. Due to the continuity and infinity of time, it cannot be directly mapped, and the user's check-in has a certain time periodicity, so the time is divided into 7*24=168 time dimensions according to the week (7 days a week) and time period (1 time period every hour), expressed as |T|.

[0046] Using a multimodal embedding layer E W Mapped to a trainable vector representation of a check-in record:

[0047]

[0048] W∈{W u ,W v ,W t}

[0049] in The weight matrices represent users, points of interest, and time respectively. Input the embedding matrix of each user's historical check-in records in Using the self-attention mechanism, we learn the user’s stable preferences, and the formula is defined as:

[0050] P(u)=Attention(S(u)W Q ,S(u)W K ,S(u)W V )

[0051]

[0052] in is the weight matrix of the self-attention layer, d is the embedding dimension, P(u)∈ is the user trajectory matrix The updated representation after assigning different weights is used as the user preference representation.

[0053] Step 2: Use the GeoHash encoding algorithm to map the locations of all points of interest to the corresponding sub-block areas. For each user, take the check-in data in the sub-block areas near the preset range of the user's current location, and obtain the regional preference of each user based on the self-attention mechanism.

[0054] First, after all points of interest are mapped to different sub-block areas through the GeoHash algorithm, different points of interest in the same sub-block area have the same code. The sub-block area set A = {a1, a2, ..., a |A|}, the dimension of the sub-block area is |A|. Taking the area where the user is currently located as the center point, take the user check-in data of the surrounding 25 areas, and the m check-in records closest to the present time in the nearby area are represented as H′(u i )={(u α ,a α ,t α ),(u β ,a β ,t β ),...,(u k ,a k ,t k )},u α represents the user ID in the check-in record of the αth sub-block area that is closest to the present time; a α represents the code of the interest point in the αth sub-block area; t α The time in the check-in record that is closest to the present time in the αth sub-block area.

[0055] Using a multimodal embedding layer E W′ Mapped to the check-in record H′(u i ) is represented by a trainable vector:

[0056]

[0057] W′∈{W′ u ,W′ a ,W′ t}

[0058] in The weight matrices represent users, points of interest, and time respectively. The embedding matrix of the check-in records in the input nearby area in Using the self-attention mechanism, we learn the user’s regional preference, and its formula is defined as:

[0059] R(u)=Attention(H(u)W′ Q ,H(u)W′ K ,H(u)W′ V )

[0060]

[0061] in is the weight matrix of the self-attention layer, d is the embedding dimension, is the user area trajectory matrix The updated representation after assigning different weights is used as the user's regional preference representation.

[0062] By desensitizing and exporting the power data, we can obtain the source data in CSV format to be analyzed. We use the Pandas library in Python to read the source data in CSV format. For the data set to be predicted, we perform data preprocessing.

[0063] Step 3: Concatenate user preferences and region preferences to obtain the comprehensive preferences of each user, and use the softmax function to calculate the probability of each point of interest being visited by the user next.

[0064] After obtaining the user preference P(u) and the region preference R(u), they are concatenated as the user's comprehensive preference representation, and the probability of each point of interest in the point of interest set V being visited is calculated using the softmax function. The formula is as follows:

[0065]

[0066] Where E(v) is the trainable projection matrix of all interest points, is the concatenation operation of user preference and region preference. The point of interest that the target user is most likely to visit at the next time step t is the point of interest with the maximum probability.

[0067] Step 4: Use meta-learning to perform loss optimization training to obtain a trained continuous POI recommendation model.

[0068] Meta-learning is used to optimize the user preference and region preference modeling modules, and local and global parameters are updated on the support set and query set respectively.

[0069] At the beginning of training, the global parameters are randomly initialized. For each user, there is a support set and a query set. The user check-in data on the support set has time steps up to t n-2, the user check-in data time step on the query set is t n-2 The data of the support set is used for local training, and its optimization goal is to minimize the recommendation loss of a single user. After local training on the support set, the global parameters are updated according to the loss on the query set. The optimization goal of the global parameter update is to minimize the training loss of the user on the query set.

[0070] Step 5: Recommend POIs to users through the trained continuous POI recommendation model, and use POIs with high probability as recommendation results.

[0071] The following is a specific description of the continuous interest point recommendation method using an embodiment. Figure 2-Figure 5 As shown:

[0072] Step 101: Analyze the user's historical check-in dataset to determine the embedding dimensions of user, point of interest, and time;

[0073] Step 102: Mapping the user's check-in data into a trainable vector representation using a multimodal embedding layer;

[0074] Step 103: Use the embedded user check-in data to model user preferences through a self-attention layer;

[0075] Step 201: Encode all points of interest using the GeoHash algorithm, and map all points of interest to different sub-block areas according to longitude and latitude;

[0076] Step 202: Taking the user's current location as the center, obtain the interaction data between all users and the points of interest in the surrounding 25 sub-block areas;

[0077] Step 203: Map the check-in data of the nearby sub-block areas into a trainable vector representation using a multimodal embedding layer;

[0078] Step 204: Modeling the user's regional preference through a self-attention mechanism using the user's interest point interaction data in the nearby area after embedding the representation;

[0079] Step 301: concatenate the user preference representation and the region preference representation, and perform matrix multiplication with the trainable matrix of all interest points;

[0080] Step 302: Calculate the probability of each point of interest being visited using the softmax function;

[0081] Step 303: taking the N points of interest with the highest probability as the recommendation results;

[0082] Step 401: Randomly initialize global parameters;

[0083] Step 402: Divide each user's check-in data into a support set and a query set according to the time step;

[0084] Step 403: Perform local training on the support set, and its optimization goal is to minimize the recommendation loss of a single user;

[0085] Step 404: Use the query set to update global parameters, and the optimization goal is to minimize the loss of the user query set.

[0086] In summary, in the process of recommending continuous points of interest to users, the present invention uses the self-attention mechanism to mine user preferences based on the check-in records of a single user; according to the user's current location, the user-point of interest interaction data in the nearby area is obtained, and the user's regional preferences are mined using the self-attention mechanism; after splicing the user preference and the regional preference representation, the probability of each point of interest being visited is calculated through softmax. During the training process, meta-learning is used to update local parameters and global parameters on the support set and query set respectively to alleviate data sparsity.

[0087] Example 2

[0088] In this embodiment, a continuous point of interest recommendation system based on meta-learning includes:

[0089] The user preference determination module is used to obtain the check-in data of each user and obtain the user preference of each user based on the self-attention mechanism.

[0090] The recommendation of each user is regarded as a learning task, and the user preferences are captured through each user's own check-in data. First, the user, interest point and time in each user's check-in record are mapped to their trainable vector representations using a multimodal embedding layer. Due to the continuity and infinity of time, it cannot be directly mapped, and the user's check-in has a certain time periodicity, so the time is divided into 7*24=168 time dimensions by week (7 days a week) and time period (1 time period every 1 hour). Using the self-attention mechanism, the above embedded representations of users, interest points and time are aggregated, and the output result is used as the user's preference.

[0091] The regional preference determination module is used to map the locations of all points of interest to the corresponding sub-block areas using the GeoHash encoding algorithm. For each user, the check-in data in the sub-block areas near the preset range of the user's current location is obtained, and the regional preference of each user is obtained based on the self-attention mechanism.

[0092] According to the longitude and latitude of the POI, all POIs are mapped using the GeoHash algorithm, with a GeoHash length of 5 and a region size of 4.9km in length and width. Since the geographical distance between two consecutive check-ins by a user is usually no more than 20km, the user's current location is taken as the center point, and the user check-in data of the surrounding 25 areas is used to capture regional preferences. Similar to user preference modeling, a multimodal embedding layer is first used to map the trainable vector representation of users, POI areas, and time. Then, the self-attention mechanism is used to assign different weights to different user check-in trajectories, and regional preferences are mined from the check-in data of nearby areas.

[0093] The POI prediction module is used to combine user preferences and region preferences to obtain the comprehensive preferences of each user, and use the softmax function to calculate the probability of each POI being visited by the user next.

[0094] After obtaining the user preference representation and the region preference representation, they are connected and combined as the user's comprehensive preference representation, and the probability of all points of interest being visited is calculated using the softmax function. The point of interest that the target user is most likely to check in next time has the highest calculated probability.

[0095] The meta-learning parameter updating module is used to perform loss optimization training using meta-learning to obtain a trained continuous POI recommendation model.

[0096] At the beginning of training, global parameters are first randomly initialized. For each user, there is a support set and a query set. The data of the support set is used for local training, and its optimization goal is to minimize the recommendation loss of a single user. After local training on the support set, the global parameters are updated according to the loss on the query set. The optimization goal of the global parameter update is to minimize the training loss of the user on the query set.

[0097] The execution module is used to recommend points of interest to users through the trained continuous point of interest recommendation model.

[0098] Example 3

[0099] An embodiment of the present invention discloses a continuous POI recommendation device based on meta-learning. The device can be applied to a POI recommendation system, such as for continuous POI recommendation based on meta-learning, etc., which is not limited by the embodiment of the present invention.

[0100] The device may include a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the method described in the above embodiment and can achieve the technical effect consistent with the above method.

[0101] The memory may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the memory may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as a "hard drive"). A program / utility having a set (at least one) of program modules may be stored in, for example, the memory, such program modules including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include the implementation of a network environment. The program modules typically perform the functions and / or methods in the embodiments described herein.

[0102] The processor executes various functional applications and data processing by running the program stored in the memory, such as implementing the method provided in the first embodiment of the present invention.

[0103] Example 4

[0104] Embodiment 4 of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method described in the above embodiment are implemented and the technical effect consistent with the above method can be achieved.

[0105] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, a device or a device or used in combination with it.

[0106] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0107] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0108] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or combinations thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0109] Of course, the storage medium containing computer executable instructions provided by an embodiment of the present invention, whose computer executable instructions are not limited to the above method operations, can also execute related operations in the method provided by any embodiment of the present invention.

[0110] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A continuous point of interest recommendation method based on meta-learning, characterized in that: The following steps are involved: Get the check-in data of each user, and obtain the user preferences of each user based on the self-attention mechanism; The GeoHash encoding algorithm is used to map the locations of all points of interest to the corresponding sub-block areas. For each user, the check-in data in the sub-block areas near the preset range of the user's current location is obtained, and the regional preference of each user is obtained based on the self-attention mechanism; The user preference and the area preference are concatenated to obtain the comprehensive preference of each user, and the softmax function is used to calculate the probability of each point of interest being visited by the user next. Meta-learning is used to perform loss optimization training to obtain a trained continuous POI recommendation model. Recommend points of interest to users through the trained continuous point of interest recommendation model.

2. The method for recommending continuous points of interest according to claim 1, characterized in that: The user u in the check-in data i The sign-in record is represented as: S′(u i )={(u i ,v1,t1),(u i ,v2,t2),...,(u i ,v n ,t n )} Among them, S′(u i ) represents user u i Check-in record, u i Indicates the user ID, v n The data representing the point of interest at the time of the user's nth check-in, t n Indicates the user's nth check-in time. The data of the point of interest includes the point of interest ID and the latitude and longitude of the point of interest. The user's check-in records are mapped into trainable vector representations through a multimodal embedding layer, and the trainable vector representations mapped by the check-in records are concatenated to obtain the user's check-in trajectory matrix. The self-attention mechanism is used to reallocate weights for the user's check-in trajectory matrix to obtain user preferences.

3. The method for recommending continuous points of interest according to claim 2, characterized in that: The check-in record of the check-in data in the sub-block area near the preset range of the user's current location is represented as: H′(u i )={(u α ,a α ,t α ),(u β ,a β ,t β ),...,(u k ,a k ,t k )} Among them, H′(u i ) represents user u i The check-in record closest to the current location in the sub-block area; u α represents the user ID in the check-in record of the αth sub-block area that is closest to the present time; a α represents the code of the interest point in the αth sub-block area; t α The time in the sign-in record of the αth sub-block area that is closest to the present time; The sign-in record H′(u i ) is mapped into a trainable vector representation through a multimodal embedding layer, and the trainable vector representation mapped by the check-in record is concatenated to obtain the user's regional trajectory matrix. The self-attention mechanism is used to reallocate the weights of the user's regional trajectory matrix to obtain the regional preference.

4. The method for recommending continuous points of interest according to claim 3, characterized in that: The probability calculation formula of each point of interest being visited by the user next time is as follows: Among them, E(v) is the trainable projection matrix of all interest points, P(u) is the user preference, R(u) is the region preference, Operator for concatenating user preferences and locale preferences.

5. The method for recommending continuous points of interest according to claim 4, characterized in that: The loss optimization training using meta-learning is specifically as follows: the check-in data of each user is divided into a support set and a query set, the data of the support set is used for local training, and its optimization goal is to minimize the recommendation loss of a single user; After local training on the support set, the global parameters are updated according to the loss on the query set. The optimization goal of the global parameter update is to minimize the user's training loss on the query set.

6. A continuous point of interest recommendation system based on meta-learning, characterized in that: include: The user preference determination module is used to obtain the check-in data of each user and obtain the user preference of each user based on the self-attention mechanism; The regional preference determination module is used to map the locations of all points of interest to the corresponding sub-block areas using the GeoHash encoding algorithm. For each user, the check-in data in the sub-block areas near the preset range of the user's current location is obtained, and the regional preference of each user is obtained based on the self-attention mechanism; The POI prediction module is used to combine user preferences and region preferences to obtain the comprehensive preferences of each user, and use the softmax function to calculate the probability of each POI being visited by the user next. The meta-learning parameter update module is used to perform loss optimization training using meta-learning to obtain a trained continuous POI recommendation model; The execution module is used to recommend points of interest to users through the trained continuous point of interest recommendation model.

7. The continuous point of interest recommendation system according to claim 6, characterized in that: The user u in the check-in data i The sign-in record is represented as: Among them, S′(u i ) represents user u i Check-in record, u i Indicates the user ID, v tn The data representing the point of interest at the time of the user's nth check-in, t n Indicates the user's nth check-in time. The data of the point of interest includes the point of interest ID and the latitude and longitude of the point of interest. The user's check-in records are mapped into trainable vector representations through a multimodal embedding layer, and the trainable vector representations mapped by the check-in records are concatenated to obtain the user's check-in trajectory matrix. The self-attention mechanism is used to reallocate weights for the user's check-in trajectory matrix to obtain user preferences.

8. The continuous point of interest recommendation system according to claim 6, characterized in that: The check-in record of the check-in data in the sub-block area near the preset range of the user's current location is represented as: H′(u i )={(u α ,a α ,t α ),(u β ,a β ,t β ),...,(u k ,a k ,t k )} Among them, H′(u i ) represents user u i The check-in record closest to the current location in the sub-block area; u α represents the user ID in the check-in record of the αth sub-block area that is closest to the present time; a α represents the code of the interest point in the αth sub-block area; t α The time in the sign-in record of the αth sub-block area that is closest to the present time; The sign-in record H′(u i ) is mapped into a trainable vector representation through a multimodal embedding layer, and the trainable vector representation mapped by the check-in record is concatenated to obtain the user's regional trajectory matrix. The self-attention mechanism is used to reallocate the weights of the user's regional trajectory matrix to obtain the regional preference.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images