Interest point recommendation method and system based on subjective and objective factor fusion

Abstract

The invention belongs to the technical field of smart city calculation and recommendation systems. According to the interest point recommendation method and system based on subjective and objective factor fusion, Gaussian distribution and two-hop information are used for capturing a motion mode and simulating a user travel decision, personal subjective preference factors and objective factors are considered at the same time, and when personal subjective preference is considered, interest points are embedded into low-dimensional vectors, so that the interest points are recommended. Mahalanobis distance is used as a distance measurement standard, so that the embedding space is non-flat and stable, and the asymmetric relationship of the interest points is expressed; when objective factors are considered, the two-hop information is mined through the global trajectory graph, the association between the interest points and the user movement is reflected, and the user behavior pattern can be mined more accurately by combining personal subjective preference factors and the objective factors.

Description

Technical Field

[0001] The present invention relates to the technical field of smart city computing and recommendation systems, and in particular to a method and system for recommending points of interest based on the fusion of subjective and objective factors. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] The popularity of smartphones and the support of global positioning systems have promoted the use of location-based social networking platforms. These platforms have formed a large amount of check-in data by collecting geographic location data and integrating social networking functions, which has greatly enriched the user's interactive experience and provided convenient services for their daily lives, thus realizing the digitization, intelligence and intelligence of cities. The next point of interest recommendation is one of the main research contents of smart city computing. By analyzing and modeling the large amount of user and point of interest information contained in the historical check-in data, user preferences and behavior patterns can be mined, which is convenient for itinerary planning and helps users explore attractive points of interest. It can also help point of interest service providers to fully understand user preferences, discover more accurate market target groups, and realize intelligent advertising push.

[0004] Early studies mainly used traditional methods such as Markov chains and matrix decomposition for recommendation. With the development of machine learning and artificial intelligence, many researchers use deep learning methods to mine the spatiotemporal information in check-in data for recommendation. However, there are still some problems with the current next point of interest recommendation method: (1) Traditional fixed point embedding cannot represent the asymmetric characteristics between points of interest. Models based on user trajectory and location information usually embed points of interest as fixed-point vectors, which causes some attributes of points of interest to be ignored. It cannot well represent the asymmetric characteristics between points of interest, resulting in biased recommendation results. (2) Existing models based on matrix decomposition and Markov chains add global information by constructing a transfer matrix, focusing only on the association between consecutive POIs in the trajectory, but ignoring the two-hop transfer characteristics of the second-to-last POI for recommendation. For example, when recommending the next POI, it is necessary to consider not only the impact of the previous location on the POI, but also the user behavior pattern constructed by the previous two locations and the next POI. (3) People often choose their next point of interest based on subjective preferences or objective needs. For example, after dining in a restaurant, they go to a shopping mall based on subjective preferences, or after leaving a subway station, they go to the office based on objective needs. Previous work has always modeled these two factors together without paying attention to the advantages of modeling them independently. It is difficult to accurately capture this complementary pattern of user behavior. Summary of the invention

[0005] In order to address the deficiencies of the prior art, the present invention provides a method and system for recommending points of interest based on the fusion of subjective and objective factors, which utilize Gaussian distribution and two-hop information to capture motion patterns and simulate user travel decisions, while considering personal subjective preference factors and objective factors. When considering personal subjective preferences, points of interest are embedded as low-dimensional vectors, and the Mahalanobis distance is used as a distance measurement standard to make the embedding space non-flat and stable, thereby expressing the asymmetric relationship between points of interest; when considering objective factors, two-hop information is mined through a global trajectory graph to reflect the association between points of interest and user movement. By combining personal subjective preference factors and objective factors, user behavior patterns can be more accurately mined.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for recommending points of interest based on the fusion of subjective and objective factors.

[0007] A method for recommending points of interest based on the fusion of subjective and objective factors includes the following processes: Get the user's historical check-in data and location information of points of interest; According to the historical check-in data and the location information, construct a corresponding trajectory space relationship matrix, a trajectory time relationship matrix and a global trajectory graph; According to the trajectory spatial relationship matrix and the trajectory temporal relationship matrix, the multimodal trajectory information of the user is embedded and represented to obtain a trajectory spatiotemporal embedding matrix; According to the user's embedded trajectory matrix and the trajectory spatiotemporal embedding matrix, a self-attention layer and a recurrent neural network are used to obtain the sequential embedding points of the user's historical trajectory, and the Mahalanobis distance is used as a measure of the embedding of the sequential embedding points and the points of interest to obtain the subjective preference probability distribution of the next point of interest; According to the global trajectory graph, a basic transfer matrix is ​​generated in combination with a self-attention mechanism, and the objective probability distribution is learned by aggregation with the two-hop transfer matrix; The subjective preference probability distribution is merged with the objective probability distribution to obtain a final probability distribution, an interest ranking list is determined according to the final probability distribution, and an optimal interest point is determined according to the interest ranking list.

[0008] In a second aspect, the present invention provides a point of interest recommendation system based on the fusion of subjective and objective factors.

[0009] A point of interest recommendation system based on the fusion of subjective and objective factors includes the following processes: A data acquisition unit is configured to: acquire the user's historical check-in data and location information of points of interest; A data processing unit is configured to: construct a corresponding trajectory space relationship matrix, a trajectory time relationship matrix and a global trajectory graph according to the historical check-in data and the location information; The multimodal spatiotemporal embedding unit is configured to: embed the multimodal trajectory information of the user according to the trajectory spatial relationship matrix and the trajectory temporal relationship matrix to obtain a trajectory spatiotemporal embedding matrix; The subjective preference probability calculation unit is configured to: obtain the sequential embedding points of the user's historical trajectory according to the user's embedding trajectory matrix and the trajectory spatiotemporal embedding matrix, using the self-attention layer and the recurrent neural network, and use the Mahalanobis distance as a measure of the embedding of the sequential embedding points and the points of interest to obtain the subjective preference probability distribution of the next point of interest; The objective probability generation unit is configured to: generate a basic transfer matrix according to the global trajectory graph in combination with a self-attention mechanism, and aggregate and learn the objective probability distribution with the two-hop transfer matrix; The optimal interest point recommendation unit is configured to: merge the subjective preference probability distribution with the objective probability distribution to obtain a final probability distribution, determine an interest ranking list according to the final probability distribution, and determine the optimal interest point according to the interest ranking list.

[0010] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention innovatively proposes a method for recommending points of interest based on the fusion of subjective and objective factors. It uses Gaussian distribution and two-hop information to capture movement patterns and simulate user travel decisions, while considering personal subjective preference factors and objective factors. When considering personal subjective preference, the points of interest are embedded as low-dimensional vectors, and the Mahalanobis distance is used as the distance measurement standard to make the embedding space non-flat and stable, thereby expressing the asymmetric relationship between the points of interest; when considering objective factors, the two-hop information is mined through the global trajectory graph to reflect the association between the points of interest and the user's movement. By combining personal subjective preference factors and objective factors, the user behavior pattern can be mined more accurately.

[0011] 2. The present invention innovatively proposes a method for recommending points of interest based on the fusion of subjective and objective factors. Compared with the existing next point of interest recommendation method, the present invention models the subjective preference factors and objective factors separately, and considers the influence of two-hop points of interest on the recommendation of the next point of interest. At the same time, it fully expresses the asymmetric relationship of the points of interest and predicts the probability distribution of the next point of interest, which can provide users with more accurate point of interest recommendations.

[0012] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0014] Figure 1 A schematic diagram of a flow chart of a model training method provided in Example 1 of the present invention; Figure 2 A schematic diagram of the flow of a method for recommending points of interest based on the fusion of subjective and objective factors provided in Example 1 of the present invention; Figure 3 A schematic diagram of a point of interest recommendation system based on the fusion of subjective and objective factors provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0016] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0017] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0018] Embodiment 1: This implementation proposes a method for recommending points of interest based on the fusion of subjective and objective factors, and constructs a next point of interest recommendation model that considers the complex information of users and points of interest, providing more comprehensive information for user decision-making planning. Figure 1 As shown, the following process is included: S1: Check-in data collection.

[0019] Collect check-in data and location information of location-based applications to construct a check-in dataset.

[0020] More specifically, user check-in data and location information of location-based applications are collected, wherein the user check-in data includes a user name, a point of interest number, and a timestamp; the location information includes a point of interest number and latitude and longitude, and a check-in dataset is constructed.

[0021] S2: Construct trajectory spatial relationship matrix, temporal relationship matrix and global trajectory graph.

[0022] According to the location information and timestamp, the trajectory spatial relationship matrix and temporal relationship matrix are constructed respectively using the time interval and geographic distance, and the global trajectory graph is constructed according to the transfer frequency between every two points of interest.

[0023] More specifically, they include: For each two access data a and b, the time interval between the ath access and the bth access is expressed as: (1); The spatial distance between the ath visit and the bth visit is expressed as: (2); in, is the radius of the Earth, are the latitudes of points a and b respectively, are the longitudes of points a and b respectively, is the inverse sine function, is a sine function, is the cosine function, thus the trajectory space relationship matrix is ​​obtained And the trajectory time relationship matrix ,in, n is the length of the largest trajectory among all trajectories, Represents the set of real numbers.

[0024] Global trajectory map of points of interest is a weighted graph, where and are the sets of nodes and edges respectively, is a weighted adjacency matrix, is the number of all points of interest, Describes and The direct transformation relationship between Represents all users and The switching frequency between Point of interest Number.

[0025] S3: Multimodal spatiotemporal embedding representations.

[0026] The user trajectory information and location information are embedded and fused in a multimodal manner. The location information is also used to embed the interest points into two low-dimensional vectors to obtain their Gaussian distribution representation.

[0027] The user name, point of interest number and timestamp information of each track in the user check-in dataset are embedded into low-dimensional vectors, which are represented as , d is the embedding dimension, forming the embedding of the user trajectory: (3); For each user sequence ,in, For users uAll trajectories of It is expressed as: (4); For every two check-in data and , and its spatial interval and time interval embedding are expressed as , traverse all check-in data and obtain the time embedding and space embedding matrices , and finally add the time-space matrix embeddings together to create the trajectory time-space embedding matrix : (5); in, is the summation function.

[0028] In order to express the asymmetric information of interest point embedding, the interest points are embedded in Gaussian space, where all interest points are represented as Gaussian distributions. Gaussian embedding is used. Assuming that the embedding dimensions are irrelevant, for an interest point For example, is considered as a diagonal covariance matrix, is the Gaussian mean, which can be further expressed as a vector , Embedded with interest point number Shared parameters. Therefore, the Gaussian mean vector of the interest point is obtained and variance vector : (6); (7); in, is an array filled with 1. is the activation function.

[0029] S4: User subjective preference learning.

[0030] Self-attention layer and GRU are used to obtain the sequential embedding points of the user's historical trajectory, which are placed in the same space as the Gaussian embedding of the points of interest to represent the user's preference for points of interest and the Gaussian distribution of the points of interest. The Mahalanobis distance is used as a metric to obtain the subjective preference probability distribution of the next point of interest.

[0031] More specifically, they include: According to The embedding trajectory matrix of a user and the trajectory spatiotemporal embedding matrix , first construct the mask matrix , where the upper left element is 1, and the remaining elements are 0. In order to explicitly add the time-space interval factor to the interaction, the present invention uses different parameter matrices Linear projection of the user trajectory to calculate the internal aggregation representation : (8); The specific attention mechanism of the self-attention layer is: (9); in, Represents the trajectory spatiotemporal relationship matrix.

[0032] After obtaining a new sequence containing internal trajectory and spatiotemporal information After that, GRU is used to convert the length of Embedded as a fixed point vector in latent space: (10); Hypothetical Point of Interest The embedding satisfies Dimensional Gaussian distribution , so the conditional probability distribution can be approximated by the Mahalanobis distance : (11); Here, the probability of selecting an interest point is negatively correlated with the distance, and the next interest point is pulled closer in space so that interest points with similar concepts can be close to each other. The probability distribution of the next interest point is expressed as , called dynamic transfer distribution: (12).

[0033] S5: Two-hop transfer objective learning.

[0034] The constructed global trajectory graph and self-attention mechanism are used to generate the basic transfer matrix of the interest point. Matrix calculation is performed on the basic transfer matrix to obtain the two-hop distribution, and these matrices are adaptively aggregated to obtain the two-hop transfer distribution, which is combined with the basic transfer matrix to obtain the objective learning probability distribution.

[0035] More specifically, they include: The present invention uses graph information to construct static transfer distribution. According to the relevant knowledge of matrix decomposition, it can be obtained: (13); in, Indicates on the track User No. From formula (13), we can see that the static transfer distribution Correlation with the points of interest visited Directly proportional.

[0036] According to a given global trajectory graph We hope to understand the transferability information relationship between interest points through the topological information structure of the graph. To this end, we first use interest point feature embedding to obtain interest point node embedding , embed the interest points in a separate graph representation space and do not share parameters with the interest point embedding of the multimodal embedding layer, obtaining objective intrinsic connections between interest points. Use the self-attention mechanism to explicitly capture the dynamic transfer distribution between interest points: (14); in, Reflects the active transfer query of interest points to other interest points, It reflects the passive key value of the interest point itself. The attention mechanism can model the dynamic transfer of interest points to each other, that is, the transfer attention matrix composed of the edge attention scores between all interest points.

[0037] The obtained transfer attention matrix reflects the objective internal connections between interest points, and can reflect the general pattern of transfer within interest points to a certain extent, providing users with an objective view of the recommendation results obtained from the interest points themselves.

[0038] The present invention aggregates static information in the global trajectory graph in a simple and effective way. Since spectral GCN performs well in characterizing the spatial relationship of the overall graph, the present invention performs the following steps to calculate the Laplacian matrix: First calculate the normalized matrix: (15); in, express The adjacency matrix of express The degree matrix of express The identity matrix of .

[0039] In order to extract more in-depth graph information, the obtained interest point attention scores are combined with the graph information, and the calculated basic transmission matrix of the interest point is used to weight the normalized Laplacian matrix , get the first hop transfer matrix as: (16); in, is the length of the largest trajectory among all trajectories, Represents element-wise multiplication, normalizing the matrix The range of is changed from [0,1] to [1,e], following the principle of enlarging larger original elements while avoiding the value 0, where e refers to the base of the natural logarithm of the mathematical constant.

[0040] Considering the need to integrate the two-hop information into the two-hop transfer matrix, the two-hop transfer matrix is ​​defined as follows: (17); in, is the first-hop transfer matrix.

[0041] Get the two-hop transmission matrix Finally, the information of the two hops is aggregated to obtain the transfer distribution: (18); in, Indicates the length of the current trajectory. represent No. OK, represent No. OK.

[0042] The two-hop convergence phenomenon is explicitly described and the critical reverse access information is incorporated to reflect the interactions between points of interest.

[0043] Considering that different points of interest have different visit frequencies, the parameter To adjust the weight of each interest point in the two-hop transfer, the adaptive aggregation of the transfer distribution can be achieved by learning the applicability attributes of the two-hop of different interest points: (19); in, The attention scores of different points of interest in the two-hop transfer are adaptively adjusted.

[0044] S5: Fusion.

[0045] The results of the two modules of user subjective preference learning and two-hop transfer objective learning are integrated.

[0046] More specifically, they include: Select the Mahalanobis distance in the potential Gaussian space to obtain the recommendation result , which reflects the user's subjective preference for points of interest; the two-hop transfer and attention mechanism are used to learn the intrinsic objective correlation between points of interest and obtain the recommendation results In order to obtain the final recommendation results , add these two modules together: (20); in, It is a fusion hyperparameter used to control the distance selection and the recommendation ratio of objective interest points.

[0047] S6: Network training.

[0048] The network is trained using the cross entropy loss function and training is stopped when the validation set loss function of the network no longer decreases.

[0049] More specifically, they include: Using cross entropy as the loss function for the next POI recommendation, the final POI recommendation result is: (twenty one); in, is the user's current trajectory length, and For users The current length is The trajectory of Recommended results for points of interest, is an indicator if the point of interest It is equal to 1 if it is true, otherwise it is 0.

[0050] The present invention trains the model, and when the loss function of the validation set no longer decreases, or reaches the set maximum number of training rounds, stops training and saves the parameters of the model.

[0051] Given the trained recommendation model, such as Figure 2 As shown, this implementation provides a specific method for recommending points of interest based on the fusion of subjective and objective factors, including the following process: Step 1: Input data and construct trajectory spatial relationship matrix, trajectory temporal relationship matrix and global trajectory graph.

[0052] The user's check-in data and location information are collected as historical check-in data. The trajectory spatial relationship matrix, trajectory temporal relationship matrix and global trajectory graph of points of interest are constructed using the known check-in data.

[0053] Step 2: Multimodal spatiotemporal embedding representation.

[0054] The user's multimodal trajectory information is embedded and represented, and the location information embedding vector is a Gaussian distribution representation of the interest point.

[0055] Step 3: Integrate subjective preferences with objective learning.

[0056] The self-attention mechanism and GRU are used to embed the user trajectory, and the distance to the interest point is measured by the Mahalanobis distance to obtain the subjective preference probability distribution; the basic transfer matrix is ​​generated by the global trajectory graph and the self-attention mechanism, and the objective probability distribution is aggregated with the two-hop transfer matrix to learn the objective probability distribution; the two probability distributions are fused to obtain the final probability distribution, and the interest points are sorted from large to small according to the probability to obtain an interest sorting list.

[0057] Step 4: The system displays the next point of interest recommendation result.

[0058] In the system, the top points of interest are recommended to users in a recommended sorted list, and the information of the points of interest is displayed to users, including category, specific location, street name, business hours and other information; users can clearly understand the next possible location and make reasonable itinerary plans; operators can also use the displayed points of interest information to push more accurate advertisements to target groups.

[0059] Embodiment 2: like Figure 3 As shown, this implementation provides a point of interest recommendation system based on the fusion of subjective and objective factors, including the following processes: A data acquisition unit is configured to: acquire the user's historical check-in data and location information of points of interest; A data processing unit is configured to: construct a corresponding trajectory space relationship matrix, a trajectory time relationship matrix and a global trajectory graph according to the historical check-in data and the location information; The multimodal spatiotemporal embedding unit is configured to: embed the multimodal trajectory information of the user according to the trajectory spatial relationship matrix and the trajectory temporal relationship matrix to obtain a trajectory spatiotemporal embedding matrix; The subjective preference probability calculation unit is configured to: obtain the sequential embedding points of the user's historical trajectory according to the user's embedding trajectory matrix and the trajectory spatiotemporal embedding matrix, using the self-attention layer and the recurrent neural network, and use the Mahalanobis distance as a measure of the embedding of the sequential embedding points and the points of interest to obtain the subjective preference probability distribution of the next point of interest; The objective probability generation unit is configured to: generate a basic transfer matrix according to the global trajectory graph in combination with a self-attention mechanism, and aggregate and learn the objective probability distribution with the two-hop transfer matrix; The optimal interest point recommendation unit is configured to: merge the subjective preference probability distribution with the objective probability distribution to obtain a final probability distribution, determine an interest ranking list according to the final probability distribution, and determine the optimal interest point according to the interest ranking list.

[0060] The specific working method of the above-mentioned unit is shown in the process of Example 1, which will not be repeated here.

[0061] It is understandable that the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0062] According to another embodiment of the present application, the system described in this embodiment can be constructed, and the method of Example 1 of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general-purpose computing device such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for recommending points of interest based on the fusion of subjective and objective factors, characterized in that: The process includes: Get the user's historical check-in data and location information of points of interest; According to the historical check-in data and the location information, construct a corresponding trajectory space relationship matrix, a trajectory time relationship matrix and a global trajectory graph; According to the trajectory spatial relationship matrix and the trajectory temporal relationship matrix, the multimodal trajectory information of the user is embedded and represented to obtain a trajectory spatiotemporal embedding matrix; According to the user's embedded trajectory matrix and the trajectory spatiotemporal embedding matrix, a self-attention layer and a recurrent neural network are used to obtain the sequential embedding points of the user's historical trajectory, and the Mahalanobis distance is used as a measure of the embedding of the sequential embedding points and the points of interest to obtain the subjective preference probability distribution of the next point of interest; According to the global trajectory graph, a basic transfer matrix is ​​generated in combination with a self-attention mechanism, and the objective probability distribution is learned by aggregation with the two-hop transfer matrix; The subjective preference probability distribution is merged with the objective probability distribution to obtain a final probability distribution, an interest ranking list is determined according to the final probability distribution, and an optimal interest point is determined according to the interest ranking list.

2. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 1, characterized in that: The historical check-in data includes: user name, point of interest number and timestamp; the location information includes: point of interest number and longitude and latitude.

3. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 1 or 2, characterized in that: According to the historical check-in data and the location information, a corresponding trajectory space relationship matrix and a trajectory time relationship matrix are constructed, including: For every two access data, determine the time interval between the two accesses and the spatial distance between the accessed positions of the two accesses; obtain the trajectory spatial relationship matrix and the trajectory temporal relationship matrix according to the time intervals and spatial distances between all the access data.

4. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 1, characterized in that: According to the trajectory spatial relationship matrix and the trajectory temporal relationship matrix, the multimodal trajectory information of the user is embedded and represented to obtain a trajectory spatiotemporal embedding matrix, including: For every two check-in data, determine the spatial interval embedding and time interval embedding, traverse all check-in data, obtain the time embedding matrix and the space embedding matrix, and finally superimpose the time embedding matrix and the space embedding matrix to obtain the trajectory spatiotemporal embedding matrix.

5. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 4, characterized in that: The interest points are embedded in Gaussian space, and all interest points are represented as Gaussian distributions. Assuming that the embedding dimensions are irrelevant, for an interest point , is the diagonal covariance matrix, is the Gaussian mean vector, , ,in, is an array filled with 1. is the activation function, A low-dimensional vector that numbers the interest points, is the variance vector.

6. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 1, characterized in that: Using the self-attention layer and recurrent neural network, we get the sequential embedding points of the user's historical trajectory. We use the Mahalanobis distance as the measure of the embedding between the sequential embedding points and the points of interest, and get the probability distribution of the subjective preference for the next point of interest, including: According to the user's embedded trajectory matrix and the trajectory spatiotemporal embedding matrix, according to the parameter matrix of the self-attention layer, an internal aggregate representation of the user's trajectory is calculated; embedding the internal aggregate representation as a fixed point vector in a latent space according to a recurrent neural network; In combination with the fixed point vector, a conditional probability distribution is calculated according to the Mahalanobis distance; The subjective preference probability distribution is determined by combining the Gaussian mean vector of the interest point, the variance vector of the interest point, and the fixed point vector.

7. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 6, characterized in that: Compute the internal aggregate representation of the user's trajectory, including: ,in, is the parameter matrix of the self-attention layer, is the embedding trajectory matrix, is the trajectory spatiotemporal embedding matrix, is the mask matrix.

8. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 1, characterized in that: According to the global trajectory graph, a basic transfer matrix is ​​generated in combination with the self-attention mechanism, and the objective probability distribution is aggregated and learned with the two-hop transfer matrix, including: Use interest point feature embedding to obtain interest point node embedding; Use self-attention mechanism to capture the dynamic transfer distribution between interest points; Calculating a normalized matrix according to the global trajectory graph; Determine a first-hop transfer matrix according to the dynamic transfer distribution and the normalized matrix, and then determine a second-hop transfer matrix according to the first-hop transfer matrix; The first-hop transfer matrix and the second-hop transfer matrix are aggregated, and the weight of each interest point in the second-hop transfer matrix is ​​adjusted using a set parameter to obtain an objective probability distribution.

9. The method for recommending points of interest based on the fusion of subjective and objective factors as claimed in claim 1, characterized in that: The subjective preference probability distribution is merged with the objective probability distribution to obtain a final probability distribution, including: The subjective preference probability distribution and the objective probability distribution are weighted and summed to obtain a final probability distribution.

10. A point of interest recommendation system based on the fusion of subjective and objective factors, characterized in that: The process includes: A data acquisition unit is configured to: acquire the user's historical check-in data and location information of points of interest; A data processing unit is configured to: construct a corresponding trajectory space relationship matrix, a trajectory time relationship matrix and a global trajectory graph according to the historical check-in data and the location information; The multimodal spatiotemporal embedding unit is configured to: embed the multimodal trajectory information of the user according to the trajectory spatial relationship matrix and the trajectory temporal relationship matrix to obtain a trajectory spatiotemporal embedding matrix; The subjective preference probability calculation unit is configured to: obtain the sequential embedding points of the user's historical trajectory according to the user's embedding trajectory matrix and the trajectory spatiotemporal embedding matrix, using the self-attention layer and the recurrent neural network, and use the Mahalanobis distance as a measure of the embedding of the sequential embedding points and the points of interest to obtain the subjective preference probability distribution of the next point of interest; The objective probability generation unit is configured to: generate a basic transfer matrix according to the global trajectory graph in combination with a self-attention mechanism, and aggregate and learn the objective probability distribution with the two-hop transfer matrix; The optimal interest point recommendation unit is configured to: merge the subjective preference probability distribution with the objective probability distribution to obtain a final probability distribution, determine an interest ranking list according to the final probability distribution, and determine the optimal interest point according to the interest ranking list.

Images