A personalized point of interest recommendation method and system based on heterogeneous graph

Abstract

The present invention relates to a personalized point of interest recommendation method and system based on heterogeneous graphs. The method comprises the following steps: S1 obtains user-point of interest check-in records; S2 constructs user-point of interest-time heterogeneous subgraphs and user-category heterogeneous subgraphs based on the check-in records; S3 calculates the user's point of interest preference probability and category preference probability for the point of interest based on the heterogeneous subgraphs; S4 calculates the user's general check-in probability for the point of interest based on the check-in records, and calculates the popularity of the point of interest; S5 calculates the user's geographical location preference check-in probability for the point of interest under the influence of geographical location preference based on the latitude and longitude of the point of interest and the check-in records; S6 predicts the user's overall check-in probability for the point of interest based on the point of interest preference probability, category preference probability, point of interest popularity, and geographical location preference check-in probability; S7 recommends points of interest to the user based on the overall check-in probability. A variety of important factors are comprehensively considered when recommending points of interest, thereby improving the accuracy of the recommendation of points of interest.

Description

Technical Field

[0001] The present invention belongs to the technical field of point of interest recommendation, and in particular relates to a method and system for personalized point of interest recommendation based on heterogeneous graphs. Background Art

[0002] With the rapid development of mobile devices and wireless networks, various location-based social networks (LBSNs), such as Weibo, Foursquare, and Yelp, have emerged. Point of Interest (POI) recommendation, as one of the most popular applications of LBSN services, provides users with suggestions on places to visit. Different from traditional e-commerce recommendations, POI recommendations have their own characteristics. First, the target of POI recommendation is a real geographical location, such as a restaurant or a cinema, connecting the user's online network to the real world. Second, traditional e-commerce recommendations usually utilize users' purchase history, while POI recommendations make full use of users' historical check-in behavior.

[0003] Although the current research on POI recommendation has achieved great results, user-personalized POI recommendation still faces challenges: how to use limited user information and location information to extract more objective features to alleviate the data sparsity in POI recommendation; how to integrate the relevant content features of POI, select appropriate learning models, and extract key factors that can distinguish user preferences. Previous studies rarely consider the various preferences of users in the areas where the points of interest are located. In addition, most studies based on geographical influence do not consider the influence of category and popularity, which actually affect check-in behavior. Effectively integrating these main aspects remains challenging, but will help improve the POI recommendation system. Summary of the invention

[0004] In view of the above problems existing in the prior art, the present invention provides a personalized POI recommendation method and system based on heterogeneous graphs, which makes full use of user check-in sequence, POI category, POI popularity and other contents to make personalized POI recommendations to users.

[0005] The present invention adopts the following technical solutions:

[0006] A personalized point of interest recommendation method based on heterogeneous graphs, comprising the steps of:

[0007] S1. Obtain user-point of interest check-in records;

[0008] S2. Based on the user-POI check-in records, construct the user-POI-time heterogeneous subgraph and the user-category heterogeneous subgraph;

[0009] S3, calculating the user's interest point preference probability for the interest point based on the user-interest point-time heterogeneous subgraph, and calculating the user's category preference probability for the category based on the user-category heterogeneous subgraph;

[0010] S4, calculating the user's general check-in probability for the interest point based on the user-interest point check-in record, and calculating the interest point popularity based on the general check-in probability;

[0011] S5. Calculate the probability of the user's geographic location preference check-in for the point of interest based on the latitude and longitude of the point of interest and the check-in record under the influence of the geographic location preference;

[0012] S6. Predicting the user's overall check-in probability for a point of interest based on the point of interest preference probability, category preference probability, point of interest popularity, and location preference check-in probability;

[0013] S7. Recommend points of interest to the user based on the overall check-in probability.

[0014] As a preferred solution, in step S1, user-point of interest check-in record information F = {u i ,l j ,t s ,c p}; User set U = {u 1 ,u 2 ,…u n}; Interest point set L = {l 1 ,l 2 ,…l m}; Check-in time set T = {t 1 ,t 2 ,…t z}; Category set C = {c 1 ,c 2 ,…c w}; i=1,2,3,...,n; j=1,2,3,...,m; s=1,2,3,...,z; p=1,2,3,...,w.

[0015] As a preferred solution, in step S2, the nodes in the user-POI-time heterogeneous subgraph are user, POI, and time, and the nodes in the user-category heterogeneous subgraph are user and category.

[0016] As a preferred solution, step S3 includes the following steps:

[0017] S3.1. For each node in the user-interest-time heterogeneous subgraph and the user-category heterogeneous subgraph, obtain the neighboring node set of the node through sampling strategy;

[0018] S3.2, mapping function that maps nodes to embedding vectors;

[0019] S3.3, based on the neighbor node set and mapping function, solve the probability formula of the neighbor node appearance of the node;

[0020] S3.4, calculate the objective function based on the probability formula of the neighboring nodes;

[0021] S3.5. Based on the objective function, we obtain the vector representation of each node, including the user vector Point of interest vector Category vector

[0022] S3.6. Based on user vector Point of interest vector Calculate the probability of interest point preference based on the user vector Category vector Compute class preference probabilities.

[0023] As a preferred solution, the calculation formula for the interest point preference probability is:

[0024]

[0025] score(u i ,l) represents user u i For the interest point preference probability of interest point l, T represents the transposition;

[0026] The formula for calculating category preference probability is:

[0027]

[0028] score(u i ,c) represents user u i The class preference probability for class c.

[0029] As a preferred solution, in step S4, user u i The general check-in probability calculation formula at point of interest l is:

[0030]

[0031] in, Represents user u i Check-in frequency at point of interest l, Represents the check-in frequency of all users in the user set at the point of interest l.

[0032] As a preferred solution, in step S4, the popularity of the point of interest l The calculation formula is:

[0033]

[0034] As a preferred solution, step S5 includes the following steps:

[0035] S5.1. Calculate the total number of check-ins of the user;

[0036] S5.2, initialize the cluster radius ∈ and the number of interest point clusters;

[0037] S5.3. Input the user's check-in record, cluster radius, and the minimum number of points of interest in a cluster, and calculate the number of clusters of points of interest and the clusters where the points of interest are located;

[0038] S5.4, when the number of clusters of interest points divided by the total number of check-ins is less than the preset threshold, add ∈+0.1 and repeat step S5.3, otherwise go to step S5.5;

[0039] S5.5, obtaining the clusters where all the points of interest are located, and calculating the user's visit center for each cluster based on the longitude and latitude of the points of interest;

[0040] S5.6. Based on the calculated visit center, calculate the user's geographic location preference check-in probability for the point of interest under the influence of the geographic location preference.

[0041] As a preferred solution, in step S6: a linear framework is used to combine the influence of the interest point preference probability, the category preference probability, the interest point popularity, and the geographical location preference check-in probability to predict the user's overall check-in probability for the interest point;

[0042] In step S7, specifically, the overall check-in probabilities predicted by the user for all points of interest are sorted, and the point of interest with the highest overall check-in probability is selected and recommended to the user.

[0043] A personalized point of interest recommendation system based on heterogeneous graphs is also provided, comprising a record acquisition module, a subgraph construction module, a calculation module, and a recommendation module; the calculation module comprises a first calculation unit, a second calculation unit, a third calculation unit, a fourth calculation unit, and a fifth calculation unit;

[0044] The subgraph construction module is connected to the record acquisition module, the first calculation unit, and the second calculation unit respectively. The record acquisition module is also connected to the third calculation unit and the fourth calculation unit respectively. The first calculation unit, the second calculation unit, the third calculation unit, and the fourth calculation unit are respectively connected to the fifth calculation unit. The fifth calculation unit is also connected to the recommendation module.

[0045] Record acquisition module, used to obtain user-point of interest check-in records;

[0046] The subgraph construction module is used to construct user-POI-time heterogeneous subgraphs and user-category heterogeneous subgraphs based on user-POI check-in records;

[0047] A first calculation unit, configured to calculate the user's interest point preference probability for the interest point according to the user-interest point-time heterogeneous subgraph;

[0048] A second calculation unit, used for calculating the category preference probability of the user for the category according to the user-category heterogeneous subgraph;

[0049] A third calculation unit, configured to calculate a general check-in probability of a user for a point of interest according to the user-point of interest check-in record, and calculate the popularity of the point of interest based on the general check-in probability;

[0050] A fourth calculation unit, configured to calculate a user's geographic location preference check-in probability for the point of interest under the influence of the geographic location preference according to the latitude and longitude of the point of interest and the check-in record;

[0051] a fifth calculation unit, configured to predict the user's overall check-in probability for a point of interest based on the point of interest preference probability, the category preference probability, the point of interest popularity, and the geographical location preference check-in probability;

[0052] The recommendation module is used to recommend points of interest to users based on the overall check-in probability.

[0053] The beneficial effects of the present invention are:

[0054] When making recommendations, we comprehensively consider important factors such as user preference, user category preference, popularity of POIs, and distance between POIs. Finally, we design a linear fusion framework to linearly fuse the above contents, predict the overall check-in probability of users to POIs, and recommend POIs to users based on the overall check-in probability, thus improving the accuracy of POI recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention 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 of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 It is a flow chart of a personalized point of interest recommendation method based on heterogeneous graphs according to the present invention;

[0057] Figure 2 It is a framework diagram of a personalized point of interest recommendation method based on heterogeneous graphs described in the present invention;

[0058] Figure 3 It is the user-interest-point-time heterogeneous subgraph and the user-category heterogeneous subgraph;

[0059] Figure 4 It is a schematic diagram of user, interest point and category vector generation;

[0060] Figure 5 It is a structural schematic diagram of a personalized point of interest recommendation system based on heterogeneous graphs described in the present invention. DETAILED DESCRIPTION

[0061] The following describes the embodiments of the present invention through specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0062] Embodiment 1:

[0063] Reference Figure 1 , 2 This embodiment provides a personalized POI recommendation method based on heterogeneous graphs, comprising the steps of:

[0064] S1. Obtain user-point of interest check-in records;

[0065] S2. Based on the user-POI check-in records, construct the user-POI-time heterogeneous subgraph and the user-category heterogeneous subgraph;

[0066] S3, calculating the user's interest point preference probability for the interest point based on the user-interest point-time heterogeneous subgraph, and calculating the user's category preference probability for the category based on the user-category heterogeneous subgraph;

[0067] S4, calculating the user's general check-in probability for the interest point based on the user-interest point check-in record, and calculating the interest point popularity based on the general check-in probability;

[0068] S5. Calculate the probability of the user's geographic location preference check-in for the point of interest based on the latitude and longitude of the point of interest and the check-in record under the influence of the geographic location preference;

[0069] S6. Predicting the user's overall check-in probability for a point of interest based on the point of interest preference probability, category preference probability, point of interest popularity, and location preference check-in probability;

[0070] S7. Recommend points of interest to the user based on the overall check-in probability.

[0071] It can be seen that the present invention comprehensively considers important factors such as user preference, user category preference, popularity of points of interest, distance between points of interest, etc. when making recommendations. Finally, the above contents are linearly fused to predict the overall check-in probability of users to points of interest, and based on the overall check-in probability, points of interest are recommended to users, thereby improving the accuracy of point of interest recommendation.

[0072] Specifically:

[0073] For the convenience of description, the relevant symbols are defined as follows:

[0074] F={u i ,l j ,t s ,c p}: User-point of interest check-in record information;

[0075] U={u 1 ,u 2 ,…u n}: user collection;

[0076] L = {l 1 ,l 2 ,…l m}: interest point set;

[0077] T={t 1 ,t 2 ,…t z}: Check-in time collection;

[0078] C={c 1 ,c 2 ,…c w}: category collection;

[0079] User vector;

[0080] Interest point vector;

[0081] Category vector;

[0082] score(u i ,l): represents user u i The probability of interest point preference for interest point l;

[0083] score(u i ,c): represents user u i The category preference probability for category c;

[0084] User i The general check-in probability at point of interest l;

[0085] The popularity of point of interest l;

[0086] User i The set of interest point clusters;

[0087] User i The points of interest clustering cluster visit center set;

[0088] User i In cluster a i Check-in frequency;

[0089] Point of interest l and cluster a to be checked in i Distance to the access center;

[0090] User i The probability of checking in based on the geographic location preference of point of interest l;

[0091] User i The overall check-in probability for point of interest l.

[0092] In step S1, user-point of interest check-in record information F = {u i ,l j ,t s ,c p}; User set U = {u 1 ,u 2 ,…u n}; Interest point set L = {l 1 ,l 2 ,…l m}; Check-in time set T = {t 1 ,t 2 ,…t z}; Category set C = {c 1 ,c 2 ,…c w}; i=1,2,3,...,n; j=1,2,3,...,m; s=1,2,3,...,z; p=1,2,3,...,w.

[0093] Reference Figure 3In step S2, a user-POI-time heterogeneous subgraph is constructed, in which the nodes are users, POIs and time. If a user visits a POI at a certain time point, then a user-POI edge and a POI-time edge are generated between the user and the POI. A user-category heterogeneous subgraph is constructed, in which the nodes are users and categories. If a user visits a POI of a certain category, then a user-category edge is generated between the user and the category.

[0094] In step S3, the user's probability of preference for the point of interest and the user's probability of preference for the category are calculated, and the specific process is as follows:

[0095] For each node v in the user-interest-time heterogeneous subgraph and the user-category heterogeneous subgraph i , N s (v i ) is the node v sampled by the sampling strategy i The sampling strategy is still the same as the Deepwalk algorithm, which uses a random walk to obtain the node's neighbor sequence. The difference from the random walk is that a biased random walk is used here. Hyperparameters p and q are introduced to control the random walk strategy. Assume that the random walk node v j The passing edge (v j ,v i ) Arrives at node v i , node v i The probability of the next visited vertex x is calculated according to the following formula:

[0096]

[0097] Among them, is the next visited node x and the current vertex v i The previous vertex v j If q is large, then vertex v i The next step of walking tends to visit node v i The neighboring nodes of the vertex in the previous step constitute a width walking strategy. If q is small, then vertex v i The next step of walking tends to visit node v i Nodes with far vertices in the previous step constitute a deep walk strategy;

[0098] f(v i ) is to put the node v i Mapping function for embedding vector, Pro(N s (v i )|f(v i )) is the i In the case of neighbor node set N s (vi ) appears, V is the set of nodes in the graph; the user vector is calculated using the Node2Vec algorithm and the interest point vector The goal of Node2Vec optimization is to maximize the probability of neighboring nodes appearing under each node condition. The formula is as follows:

[0099]

[0100] Assume that a source node v is given i Under this condition, the probability of its neighbor nodes appearing is independent of the other nodes in the neighbor set. By introducing the naive Bayes hypothesis, Pro(N s (v i )|f(v i )) can be expressed as the following formula:

[0101]

[0102] Assuming that a node shares the same set of embedding vectors when it is a source node and a neighbor node, then for a given node v i In the case of node v j The probability of occurrence is as follows:

[0103]

[0104] Based on the above two assumptions, the final objective function is expressed as follows:

[0105]

[0106] in is the normalization factor, Through the objective function, we can get the vector representation of each node, and the user vector is represented as The interest point vector is represented as The category vector is represented as

[0107] The obtained user vector Point of interest vector and the category vector Use the cosine similarity function to calculate user u i For the preference of point of interest l and user u i For the preference of category c, the calculation formula is as follows:

[0108]

[0109]

[0110] score(ui ,l) represents user u i For the interest point preference probability of interest point l, T represents the transposition; score(u i ,c) represents user u i For the category preference probability of category c, the above process can refer to Figure 4 shown.

[0111] In step S4, the user u is calculated based on the user-point of interest check-in record information. i For a general check-in probability of a certain point of interest l, the calculation formula is as follows:

[0112]

[0113] in Represents user u i Check-in frequency at point of interest l, Represents the check-in frequency of all users in the user set at the point of interest l; the popularity of the point of interest l is further calculated by the general check-in probability of the user for a certain point of interest The calculation formula is as follows:

[0114]

[0115] In step S5, based on the latitude and longitude of the point of interest and the sign-in record, the probability of the user's geographical location preference sign-in for the point of interest under the influence of the geographical location preference is calculated. The specific process is as follows:

[0116] S5.1. For each user u in the user set i , calculate user u i The total number of check-ins is recorded as:

[0117] S5.2, initialize the cluster radius ∈ = 0.1, initialize the number of clusters of interest points clusternum = 0;

[0118] S5.3. Input user u i The check-in records, cluster radius ∈, and the minimum number of points of interest in a cluster MinPts are used. All points of interest in the given point of interest set are marked as "unvisited" using the DBSCAN algorithm. DBSCAN randomly selects an unvisited point of interest p, marks p as "visited", and checks whether p's ∈-neighborhood contains at least MinPts objects. If not, p is marked as a noise point. Otherwise, a new cluster a is created for p i , and put all objects in p's ∈-neighborhood into the candidate set N. DBSCAN iteratively adds objects in N that do not belong to other clusters to a iIn this process, for the object P′ marked as “unvisited” in N, DBSCAN marks it as “visited” and checks its ∈-neighborhood. If P′’s ∈-neighborhood contains at least MinPts objects, all objects in P′’s ∈-domain are added to N. DBSCAN continues to add objects to a i , until a i It cannot be expanded, that is, until N is empty. At this time, cluster a i After the generation is completed, the number of clusters of interest points and the clusters where the interest points are located are finally calculated;

[0119] S5.4, when the number of clusters of interest points, clusternum, is divided by When it is less than a certain threshold δ, set ∈ + 0.1 and repeat step S5.3, otherwise go to step S5.5;

[0120] S5.5. Get the final user u i All clusters of Then get user u i For each cluster, the visit center in The calculation formula is as follows:

[0121]

[0122] in Represents user u i Check-in frequency at point of interest l, Represents user u i In cluster a i Check-in frequency, lat l Indicates the latitude of the point of interest l, lon l represents the longitude of the point of interest l;

[0123] S5.6. Calculate the user u under the influence of geographical location i The probability of geographic location preference check-in for the point of interest l is as follows:

[0124]

[0125] in, Represents user u i In cluster a i Check-in frequency, Represents user u i The check-in frequency in each cluster, Represents the interest point l and cluster a to be checked in i The distance to the access center is calculated as follows:

[0126]

[0127] Among them, cluster a i The longitude of the visit center point is expressed as Latitude is expressed as

[0128] In step S6: a linear framework is used to combine the influence of interest point preference probability, category preference probability, interest point popularity, and location preference check-in probability to predict user u i The overall check-in probability of point of interest l is calculated as follows:

[0129]

[0130] Among them, α represents the relative importance of category influence, β represents the relative importance of POI popularity, and γ represents the relative importance of geographical influence;

[0131] In step S7, specifically, the overall check-in probabilities predicted by the user for all points of interest are sorted, and the point of interest with the highest overall check-in probability is selected and recommended to the user.

[0132] Embodiment 2:

[0133] Reference Figure 5 , This embodiment provides a personalized POI recommendation system based on heterogeneous graphs, based on the recommendation method described in Embodiment 1, including a record acquisition module, a subgraph construction module, a calculation module, and a recommendation module; the calculation module includes a first calculation unit, a second calculation unit, a third calculation unit, a fourth calculation unit, and a fifth calculation unit;

[0134] The subgraph construction module is connected to the record acquisition module, the first calculation unit, and the second calculation unit respectively. The record acquisition module is also connected to the third calculation unit and the fourth calculation unit respectively. The first calculation unit, the second calculation unit, the third calculation unit, and the fourth calculation unit are respectively connected to the fifth calculation unit. The fifth calculation unit is also connected to the recommendation module.

[0135] Record acquisition module, used to obtain user-point of interest check-in records;

[0136] The subgraph construction module is used to construct user-POI-time heterogeneous subgraphs and user-category heterogeneous subgraphs based on user-POI check-in records;

[0137] A first calculation unit, configured to calculate the user's interest point preference probability for the interest point according to the user-interest point-time heterogeneous subgraph;

[0138] A second calculation unit, used for calculating the category preference probability of the user for the category according to the user-category heterogeneous subgraph;

[0139] A third calculation unit, configured to calculate a general check-in probability of a user for a point of interest according to the user-point of interest check-in record, and calculate the popularity of the point of interest based on the general check-in probability;

[0140] A fourth calculation unit, configured to calculate a user's geographic location preference check-in probability for the point of interest under the influence of the geographic location preference according to the latitude and longitude of the point of interest and the check-in record;

[0141] a fifth calculation unit, configured to predict the user's overall check-in probability for a point of interest based on the point of interest preference probability, the category preference probability, the point of interest popularity, and the geographical location preference check-in probability;

[0142] The recommendation module is used to recommend points of interest to users based on the overall check-in probability.

[0143] It should be noted that the personalized POI recommendation system based on heterogeneous graphs provided in this embodiment is similar to that in the first embodiment and will not be described in detail here.

[0144] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope of the present invention.

Claims

1. A personalized point of interest recommendation method based on heterogeneous graph, characterized in that: Includes steps: S1. Obtain user-point of interest check-in records; S2. Based on the user-POI check-in records, construct the user-POI-time heterogeneous subgraph and the user-category heterogeneous subgraph; S3, calculating the user's interest point preference probability for the interest point based on the user-interest point-time heterogeneous subgraph, and calculating the user's category preference probability for the category based on the user-category heterogeneous subgraph; S4, calculating the user's general check-in probability for the interest point based on the user-interest point check-in record, and calculating the interest point popularity based on the general check-in probability; S5. Calculate the probability of the user's geographic location preference check-in for the point of interest based on the latitude and longitude of the point of interest and the check-in record under the influence of the geographic location preference; S6. Predicting the user's overall check-in probability for a point of interest based on the point of interest preference probability, category preference probability, point of interest popularity, and location preference check-in probability; S7, recommending points of interest to users based on the overall check-in probability; In step S1, user-point of interest check-in record information F = {u i ,l j ,t s ,c p }; User set U = {u1,u2,…u n }; Interest point set L = {l1,l2,…l m }; Check-in time set T = {t1, t2, ... t z }; Category set C = {c1, c2, ... c w }; i=1,2,3,...,n; j=1,2,3,...,m; s=1,2,3,...,z; p=1,2,3,...,w; In step S2, the nodes in the user-POI-time heterogeneous subgraph are user, POI, and time, and the nodes in the user-category heterogeneous subgraph are user and category; Step S3 includes the following steps: S3.

1. For each node in the user-interest-time heterogeneous subgraph and the user-category heterogeneous subgraph, obtain the neighboring node set of the node through sampling strategy; S3.2, mapping function that maps nodes to embedding vectors; S3.3, based on the neighbor node set and mapping function, solve the probability formula of the neighbor node appearance of the node; S3.4, calculate the objective function based on the probability formula of the neighboring nodes; S3.

5. Based on the objective function, we obtain the vector representation of each node, including the user vector Point of interest vector Category vector S3.

6. Based on user vector Point of interest vector Calculate the probability of interest point preference based on the user vector Category vector Calculate category preference probabilities; The calculation formula of interest point preference probability is: score(u i ,l) represents user u i For the interest point preference probability of interest point l, T represents the transposition; The formula for calculating category preference probability is: score(u i ,c) represents user u i The category preference probability for category c; Step S5 includes the steps of: S5.

1. Calculate the total number of check-ins of the user; S5.2, initialize the cluster radius ∈ and the number of interest point clusters; S5.

3. Input the user's check-in record, cluster radius, and the minimum number of points of interest in a cluster, and calculate the number of clusters of points of interest and the clusters where the points of interest are located; S5.4, when the number of clusters of interest points divided by the total number of check-ins is less than the preset threshold, add ∈+0.1 and repeat step S5.3, otherwise go to step S5.5; S5.5, obtaining the clusters where all the points of interest are located, and calculating the user's visit center for each cluster based on the longitude and latitude of the points of interest; S5.6, based on the calculated access center, calculating the user's geographic location preference check-in probability for the point of interest under the influence of the geographic location preference; In step S6: using a linear framework to combine the influence of the POI preference probability, the category preference probability, the POI popularity, and the geographical location preference check-in probability, to predict the user's overall check-in probability for the POI; In step S7, specifically, the overall check-in probabilities predicted by the user for all points of interest are sorted, and the point of interest with the highest overall check-in probability is selected and recommended to the user.

2. The personalized POI recommendation method based on heterogeneous graph according to claim 1, characterized in that: In step S4, user u i The general check-in probability calculation formula at point of interest l is: in, Represents user u i Check-in frequency at point of interest l, Represents the check-in frequency of all users in the user set at the point of interest l.

3. The personalized POI recommendation method based on heterogeneous graph according to claim 2, characterized in that: In step S4, the popularity of point of interest l The calculation formula is:

4. A personalized point of interest recommendation system based on heterogeneous graph, based on a personalized point of interest recommendation method based on heterogeneous graph according to any one of claims 1 to 3, characterized in that: It includes a record acquisition module, a subgraph construction module, a calculation module, and a recommendation module; the calculation module includes a first calculation unit, a second calculation unit, a third calculation unit, a fourth calculation unit, and a fifth calculation unit; The subgraph construction module is connected to the record acquisition module, the first calculation unit, and the second calculation unit respectively. The record acquisition module is also connected to the third calculation unit and the fourth calculation unit respectively. The first calculation unit, the second calculation unit, the third calculation unit, and the fourth calculation unit are respectively connected to the fifth calculation unit. The fifth calculation unit is also connected to the recommendation module. Record acquisition module, used to obtain user-point of interest check-in records; The subgraph construction module is used to construct user-POI-time heterogeneous subgraphs and user-category heterogeneous subgraphs based on user-POI check-in records; A first calculation unit, configured to calculate the user's interest point preference probability for the interest point according to the user-interest point-time heterogeneous subgraph; A second calculation unit, used for calculating the category preference probability of the user for the category according to the user-category heterogeneous subgraph; A third calculation unit, configured to calculate a general check-in probability of a user for a point of interest according to the user-point of interest check-in record, and calculate the popularity of the point of interest based on the general check-in probability; A fourth calculation unit, configured to calculate a user's geographic location preference check-in probability for the point of interest under the influence of the geographic location preference according to the latitude and longitude of the point of interest and the check-in record; a fifth calculation unit, configured to predict the user's overall check-in probability for a point of interest based on the point of interest preference probability, the category preference probability, the point of interest popularity, and the geographical location preference check-in probability; The recommendation module is used to recommend points of interest to users based on the overall check-in probability.

Images