A community search method, system, program, and storage medium with embedded preference enhancement

Abstract

This invention provides a method, system, program, and storage medium for community search with enhanced embedding preferences. The system includes a social network node embedding device, a community search device, a random walk sequence acquisition device, and a system control device. The method dynamically combines embedding learning and community search, utilizing a Check2Vec model to learn node embeddings, improving the modeling ability of check-in information. In community search, it comprehensively considers node attributes and check-in information to obtain more accurate user-location binary communities. The advantages of this invention include multimodal information integration, applicability to multiple data domains, and flexible adjustability, providing a powerful means to improve the effectiveness and personalization of community search.

Description

Technical Field

[0001] This invention belongs to the field of social network technology, specifically relating to a community search method, system, program, and storage medium with embedded preference enhancement. Background Technology

[0002] Generally speaking, community search (CS) problems in large-scale graphs can be divided into community search based on graph structure and community search based on a combination of graph attributes and graph structure [Fang Y, Huang X, Qin L, et al.A survey of community search overbig graphs[J].VLDB J.,2020,29(1):353-392.].

[0003] Typically, the goal of Computer Science (CS) is to search for high-quality communities online based on query requests. Specifically, given an isomorphic graph G and a query node Q, the objective is to find a community, or dense subgraph, containing Q that satisfies the following properties:

[0004] Connectivity, meaning that the nodes in a community are connected;

[0005] Cohesion (internal cohesion) refers to the close connection between nodes in a community, which is a special measure of excellence.

[0006] In graph-based community search, the main focus is on the cohesion of the community structure. The cohesion of the community structure is typically measured using classic subgraph cohesion metrics, primarily k-core, k-truss, and k-clique metrics [Zhang C, Zhang Y, Zhang W, et al. Efficient Maximal Spatial CliqueEnumeration[C] / / 35th IEEE International Conference on Data Engineering,ICDE2019,Macao,China,April 8-11,2019.IEEE,2019:878-889.], as shown below:

[0007] A k-core is a maximal subgraph of G, where each node in the subgraph has a degree of at least k.

[0008] k-truss[Akbas E,Zhao P. Truss-based Community Search: a Truss-equivalence Based Indexing Approach[J].Proc.VLDB Endow.,2017,10(11):1298-1309.] is the largest subgraph of G, where each edge is contained in at least (k-2) triangles.

[0009] k-clique[Lin YK,Lin WC,Wang C CK-Closest Points and Maximum CliquePruning for Efficient and Effective 3-D Laser Scan Matching[J].IEEE RoboticsAutom.Lett.,2022,7(2):1471-1477.] is a set of k nodes of G, and each pair of nodes has an edge.

[0010] Each of the three structures has its advantages and disadvantages. Currently, the most researched community search method is based on k-core. Compared to the other two, the k-core structure consumes the least computational resources because it uses node degree as the subgraph density control condition. k-clique uses complete subgraphs as the density control condition and is often used to identify overlapping communities. The k-truss-based community search method is often used for problems with high requirements for structural cohesion. This algorithm uses the number of triangles containing an edge as the subgraph density control condition. The k-truss structure usually has higher structural cohesion than other methods, but its computational resource consumption is also relatively large.

[0011] Classical community search algorithms are mainly based on isomorphic graphs, such as the user social networks of Facebook, Twitter, and Amazon. Some studies have also extracted the user social network of LBSN for community search. Sozio et al. [Sozio M, Gionis A. The community-search problem and how to plan a successful cocktailparty[C] / / RAO B, KRISHNAPURAM B, TOMKINS A, et al. Proceedings of the 16th ACMSIGKDD International Conference on Knowledge Discovery and Data Mining, Washington, DC, USA, July 25-28, 2010. ACM, 2010: 939-948.] proposed the first global algorithm to find (kd)-cores containing q. Zhu et al. [Zhu Junchao, Wang Chaokun. Community search method under complex conditions[J]. Journal of Software, 2019(03vo 30): 552-572.] proposed community search under complex conditions and conducted experiments on the LBSN dataset. However, these works assume non-spatial graphs and ignore the location information of nodes.

[0012] Community search on LBSNs is currently a hot research topic. For example, in the SAC algorithm [Fang Y, Wang Z, Cheng R, et al. On Spatial-Aware Community Search[J].IEEE Trans.Knowl.DataEng.,2019,31(4):783-798.], it is assumed that each user is associated with a location and a spatially-aware community (SAC) is identified. This paper studies the problem of online community search on spatial graphs, and defines communities that satisfy connectivity, structural cohesion, and spatial cohesion as spatially-aware communities. However, this paper associates each user with a location, ignoring the embedding information of LBSNs and the attribute information of nodes. Guo et al. [Guo F, Yuan Y, Wang G, et al. Multi-attributed Community Search in Road-social Networks[C] / / 37th IEEE International Conference on Data Engineering,ICDE 2021, Chania, Greece, April 19-22, 2021.IEEE, 2021:109-120.] proposed a multi-attributed community search (MAC) method based on road networks to address the problem of inaccurate results caused by imprecise user input preference vectors. However, this paper only considers user attribute information and ignores location attribute information and embedded heterogeneous information.Junghoon Kim et al. [Kim J, Guo T, Feng K, et al. Densely Connected User Community and Location Cluster Search in Location-Based Social Networks[C] / / MAIER D, POTTINGER R, DOAN A, et al. Proceedings of the 2020 International Conference on Management of Data, SIGMOD Conference 2020, [Portland, OR, USA], June 14-19, 2020. ACM, 2020: 2199-2209.] proposed the GeoSocial Community Search (GCS) problem, which integrates location information given a location-based social network and a set of query nodes. This paper studies a community search algorithm with the goal of maximizing embedding density. By traversing the k-core structures containing all query nodes in the user network graph and the location subgraph, the algorithm selects a pair of k-core structures that maximizes the embedding density as the result set. Summary of the Invention

[0013] The purpose of this invention is to provide a community search method, system, program, and storage medium with embedded preference enhancement.

[0014] The objective of this invention is achieved through the following technical solution:

[0015] A community search method with embedded preference enhancement, the specific steps of which are as follows:

[0016] Step 1: Construct a model Check2Vec containing LBSN user node and location node embeddings. Ensure node proximity by maximizing the cosine similarity between the four types of nodes in the hypergraph (location, time, semantics, and user domain) and their best-fit line.

[0017] Step 2: In the improved community search algorithm AGSCS+, the check-in vector representations of users and location nodes learned by Check2Vec, as well as the corresponding best-fit lines, are used to perform community search;

[0018] Step 3: The AGSCS+ algorithm combines node attribute similarity and check-in similarity when calculating the score, and adjusts the impact of attribute information and check-in information on the search results community by inputting parameter p;

[0019] Step 4: Perform jump walk sequence sampling in the user-location bipartite graph using a random walk strategy to obtain sequences of user and location nodes for learning check-in hyperedge information;

[0020] Step 5: The Check2Vec algorithm learns node embeddings and uses the idea of ​​the best-fit line in linear regression to maximize the cosine similarity between the node embeddings and the best-fit line.

[0021] Step 6: Use negative sampling techniques to learn vector representations of users and location nodes to ensure better capture of the relationships between nodes;

[0022] Step 7: In the community search phase, the optimal user-location binary community is obtained by comprehensively considering the attribute similarity and check-in similarity of users and location nodes through the comprehensive scoring calculation formula in the AGSCS+ algorithm.

[0023] A community search system with enhanced embedding preferences includes a social network node embedding device, a community search device, a random walk sequence acquisition device, and a system control device.

[0024] The social network node embedding device learns the vector representations of users and location nodes, optimizes cosine similarity through a node embedding learning algorithm, and combines a random walk strategy to sample node sequences and embedding hyperedge sequences to provide informative node embedding representations for community search.

[0025] The community search device uses the learned node vectors to construct a user-location bipartite graph, calculates the comprehensive score of the nodes, and obtains the optimal community through the community search algorithm.

[0026] The random walk sequence acquisition device is a sequence acquisition engine that generates random walk sequences for users and location nodes and samples random walk sequences embedded with hyperedges.

[0027] The system control device is a central control unit that coordinates the workflow of each device and manages data transmission and information exchange.

[0028] Furthermore, the main task of the social network node embedding device is to learn the vector representations of users and location nodes. Through a node embedding learning algorithm, it optimizes the cosine similarity in the manner of best fitting the line, thereby capturing the correlation between nodes. Through a random walk strategy, it generates sequences of users and location nodes, as well as sequences of sampled embedding hyperedges, providing rich and meaningful node embedding representations for community search.

[0029] Furthermore, the community search device utilizes the node vectors learned by Check2Vec to construct a user-location bipartite graph, realizing the core algorithm of community search. By calculating the comprehensive score of nodes and comprehensively considering the weights of attribute information and embedding information, the system can more accurately evaluate the similarity between nodes. Finally, the optimal community is found through the community search algorithm, realizing community search based on embedding information.

[0030] Furthermore, the random walk sequence acquisition device is responsible for generating random walk sequences for users and location nodes; by using a jump-walk strategy, it solves the problem of no relationship between structural relationships and location nodes, ensuring that the generated sequences are more diverse.

[0031] Furthermore, the system control device acts as the central control unit of the entire system, responsible for coordinating the workflow between the Check2Vec model learning device, the community search algorithm device, and the random walk sequence acquisition device; through effective data transmission and information exchange, it ensures that each device operates in an orderly manner and works collaboratively.

[0032] A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that: when the computer program / instructions are executed by a processor, they implement the steps of an embedded preference-enhanced community search method.

[0033] A computer program product, comprising a computer program / instructions, characterized in that: when the computer program / instructions are executed by a processor, they implement the steps of an embedded preference-enhanced community search method.

[0034] The beneficial effects of this invention are as follows:

[0035] This invention combines embedding learning and community search, utilizing the Check2Vec model to learn node embeddings, thereby improving the modeling ability of check-in information. Furthermore, it comprehensively considers node attributes and check-in information in community search to obtain more accurate user-location binary communities. This invention offers advantages such as multimodal information integration, applicability to multiple data domains, and flexible adjustability, providing a powerful means to improve the effectiveness and personalization of community search. Attached Figure Description

[0036] Figure 1 Schematic diagram of the LBSN described in this invention;

[0037] Figure 2 A structural block diagram of the system described in this invention;

[0038] Figure 3 This represents a schematic diagram of learning embedded hyperedges;

[0039] Figure 4 A diagram illustrating a random walk strategy;

[0040] Figure 5 Comparison chart of runtime when parameter p is changed;

[0041] Figure 6 A graph showing attribute similarity and check-in density on Yelp;

[0042] Figure 7 A graph showing attribute similarity and check-in density on Gowalla;

[0043] Figure 8 A graph showing attribute similarity and check-in density on Foursquare;

[0044] Figure 9 A comparison chart of running times on Yelp;

[0045] Figure 10 A comparison chart of attribute scores;

[0046] Figure 11 This is a comparison chart of structural density;

[0047] Figure 12 This is a chart comparing check-in density. Detailed Implementation

[0048] The present invention will now be further described with reference to the accompanying drawings.

[0049] Figure 1 A schematic diagram of the LBSN described in this invention, for example:

[0050] User U goes to breakfast shop (p1) in the morning, starts working at the company (p2) in the morning, has lunch at the restaurant (p3) at noon, goes to the supermarket (p4) after lunch, then returns to the company (p2) to continue working, goes to the movie theater (p5) after get off work, and returns to the apartment (p6) in the evening. Therefore, User U checks in at multiple locations throughout the day, and User U may have rated these locations on social media platforms. Meanwhile, these locations are constantly checking in by many other people, such as colleagues and clients at the company, and employees and customers at the shopping mall.

[0051] Figure 2 The diagram shown is a structural block diagram of the system described in this invention, comprising:

[0052] Social Network Node Embedding Device: This device learns vector representations of user and location nodes. It optimizes cosine similarity through a node embedding learning algorithm and combines a random walk strategy to sample node sequences and embedding hyperedge sequences, providing informative node embedding representations for community search. This vector representation plays a crucial role in subsequent community search, helping the system better understand the relationships between users, locations, and embedding information.

[0053] The community search device utilizes node vectors learned from Check2Vec to construct a user-location bipartite graph, implementing the core algorithm for community search. By calculating a comprehensive score for each node and considering the weights of attribute and embedding information, the system can more accurately assess the similarity between nodes. Ultimately, the optimal community is found through the community search algorithm, achieving community search based on embedding information. The device aims to provide more intelligent and personalized community search results, better meeting user needs.

[0054] Random Walk Sequence Acquisition Device: This device acts as a sequence acquisition engine, generating random walk sequences for users and location nodes, and sampling random walk sequences with embedded hyperedges. By employing a skip-walk strategy, it addresses the issue of non-existent relationships between structural relationships and location nodes, ensuring greater diversity in the generated sequences. These sequences are not only used for node embedding learning but also provide rich sequence information for community search, helping the system to more comprehensively understand the relationships between users and location nodes, thus improving the model's robustness and generalization ability.

[0055] System Control Unit: This unit acts as the central control unit, responsible for coordinating the workflows between the Check2Vec model learning unit, the community search algorithm unit, and the random walk sequence acquisition unit. Through effective data transmission and information exchange, it ensures the orderly operation and collaborative work of each unit. The overall system control unit's role is to improve the overall efficiency of the system, ensuring that all components can cooperate organically, thereby enabling the smooth operation of the embedded preference-enhanced community search system.

[0056] Figure 3 To represent the learning embedding hyperedge diagram, specifically:

[0057] For user nodes in a user-location bipartite graph, the random walk scheme of this invention performs a classic random walk algorithm on user nodes and location nodes based on a path guided by check-in edges. For example, for each encountered user node, a set of check-in hyperedges is sampled from the corresponding user at the user node. During iteration at each user node in the random walk sequence, learning is done not only from heterogeneous edges such as {user-location-user…}, but also from the check-in hyperedges {location, time, semantics, user} at the corresponding user node. In other words, the node embedding learning process alternates between the user-location heterogeneous structure and the check-in hyperedges. To perform a classic random walk on a heterogeneous network, this invention uses the same strategy as existing work [13,23], generating a sequence of length at each user node, and determining the walk order based on the relationship between the user and the location.

[0058] Furthermore, to balance the impact of attributes and sign-in hyperedges on the learned node embeddings, Check2Vec introduces an adjustable parameter p to control the emphasis on both. Specifically, this invention uses parameter p to determine the influence of attributes and sign-in on node attribute scores. The following sections detail two types of node embedding learning processes that preserve the n-directed node proximity of the hyperedges. This invention first introduces the AGSCS community search algorithm combined with representation learning, then describes the random walk sequence acquisition algorithm and the method for learning sign-in hyperedges.

[0059] Figure 4 This is a diagram illustrating a random walk strategy, specifically:

[0060] To better learn the information of the check-in hyperedges, a jump-walk sequence was chosen as the random walk sequence for learning the check-in hyperedges. When learning the check-in vector representation of a user node, for a given context window of length k, k contexts are iterated for each user node u, resulting in k pairs of user nodes. Specifically, a window of length k is generated to the left of user node u. Through a random position node connected to u, another user node connected to that position node is found. This process is repeated twice to complete the window sequence sampling for u, resulting in k pairs of user nodes for learning. Simultaneously, for user node u, k pairs of check-in hyperedges also need to be learned, for a total of 2k edges.

[0061] Because the community search algorithm AGSCS of this invention needs to consider both user and location node types simultaneously, it also needs to learn the reverse sign-in representation of location nodes. The learning of location nodes follows a similar process, except that the roles of user and location are reversed, resulting in 2k edges for learning. Similar to this learning process on the user graph, the learning of location nodes is also performed in the same way, such as... Figure 4 It demonstrates a two-way movement method from user to location and from location to user.

[0062] To verify the effectiveness of this invention, experiments were conducted on three real datasets: Yelp, Foursquare, and Gowalla. The dataset attributes are shown in Table 1, and the dataset size and index construction time are shown in Table 2.

[0063] Table 1

[0064]

[0065] Table 2

[0066]

[0067] The evaluation metrics include runtime, attribute similarity, structure density, and embedding density. Specifically, the total number of sampled hyperedges is first determined by combining the number of check-in data points and the PageRank value of the reference node. That is, for a user u and location v, if the number of check-in data points is n and the PageRank value is pr, then the number of sampled hyperedges for this user is... (Round down). Secondly, to sample more recent embedded hyperedges, a sampling ratio is set. First, the user-to-location check-in data is sorted chronologically, with more recent dates having a higher sampling probability. For example, between user u and location v, the number of sampled hyperedges is... Divide the n embedded data into 10 time periods t according to time. i For each time period i ∈ [1, 10], a larger proportion of samples are taken for more recent times (smaller i). If there are not enough samples in a given time period, samples are taken sequentially from the next time period. This method allows us to sample effective check-in embedding information between user u and location v, thus obtaining a dataset for representation learning.

[0068] For the Check2Vec method of this invention, the same random walk parameters (l=80, r=10, k=10) are used as those for DeepWalk and Node2vec. If not specified, the node embedding dimension d of all methods in all experiments is uniformly set to 128 in tensor form, and the number of negative samples γ is set to 10.

[0069] First, experiments were conducted by changing the value of parameter p, mainly to observe the changes in attribute similarity and embedding density after changing parameter p. The experimental results regarding runtime are as follows: Figure 5 As shown, the overall trend fluctuates within a certain range, indicating that changing the parameter p does not affect the experiment time. This is because the main time consumption of the algorithm in this invention comes from the construction of the location graph and the search of the k-TTC infrastructure. Whether or not the parameter p is introduced only changes the number of nodes within this k-TTC range, but the total time remains at a similar level.

[0070] Figure 6These figures show how attribute similarity and check-in density of communities on Yelp change with parameter p. Overall, attribute similarity decreases as parameter p increases, while check-in density increases with p, but the rate of increase slows down after p = 0.4. Combining these two figures, it can be seen that, as shown in Definition 5, parameter p causes attribute similarity and check-in density to exhibit an inverse relationship. It can be observed that at p = 0.45, attribute similarity has a relatively high value, while check-in density has entered a plateau phase with slow increases. Therefore, setting the parameter to p = 0.45 can achieve higher attribute similarity and check-in density, allowing the searched communities to simultaneously satisfy attribute similarity and check-in preference.

[0071] akin, Figure 7 The experiment was conducted on the Gowalla dataset. It was found that the attribute similarity on the Gowalla dataset generally decreases as the parameter p increases. Unlike Yelp, the range of variation in Gowalla is relatively large. This may be because Gowalla's attributes are self-generated, resulting in greater randomness and a larger range of variation. The check-in density, however, is similar to Yelp, showing an initial increase followed by a plateau with increasing parameter p, exhibiting a stable growth after p = 0.35. Figure 8 The experimental results on the Foursquare dataset show that the trends in attribute similarity and check-in density are the same as those of Yelp and Gowalla. However, unlike Yelp and Gowalla, Foursquare shows the least variation in attribute similarity and check-in density. This is because Foursquare has the smallest proportion of check-in data, and as a sparse graph, check-in has a relatively small impact on the results.

[0072] In summary, these experimental results demonstrate that the parameter p has a relatively small impact on runtime. As the value of parameter p increases, the attribute similarity of the community decreases, while the check-in density increases. This is consistent with the scoring calculation formula. By changing the value of parameter p, the influence ratio of attributes and check-ins can be altered. On different datasets, by selecting an appropriate p value, the community can simultaneously satisfy both high check-in density and attribute similarity. For example, on Yelp, Gowalla, and Foursquare, the parameter can be selected as 0.45, 0.35, and 0.5, respectively. Therefore, the experiments prove that the method of this invention can enable the community results to meet the attribute similarity and check-in preference requirements of the AGSCS problem.

[0073] Next, this invention continues to experiment on the effectiveness of representation learning methods based on three algorithms: GCS, AGSCS, and AGSCS+. First, a comparison is made in terms of runtime, such as... Figure 9As shown, similar to previous results, GCS consumes significantly more time than AGSCS and AGSCS+ on query nodes of different sizes, while the time consumption of AGSCS and AGSCS+ is not significantly different.

[0074] Below is a comparison of the attribute scoring performance of the three algorithms: GCS, AGSCS, and AGSCS+. Figure 10 The results are from the Yelp, Gowalla, and Foursquare datasets. It can be seen that attribute scores exhibit the same trend across all datasets. GCS attributes score significantly lower than AGSCS and AGSCS+ across all three datasets, indicating that considering node attributes has a significant impact on the community. Without considering node attributes, the searched community may contain a large number of low-quality nodes irrelevant to the queried node. AGSCS+ attribute scores are slightly lower than AGSCS, but the difference is not large on the Yelp and Gowalla datasets. This suggests that the attribute scores of the communities obtained by all algorithms proposed in this invention are related to the attribute characteristics of the nodes themselves in the dataset.

[0075] Similarly, such as Figure 11 The diagram shows a comparison of the structural similarity of the three algorithms, GCS, AGSCS, and AGSCS+, under the same settings. It can be seen that AGSCS and AGSCS+ have greater structural similarity than GCS, indicating that the k-TTC structure used as a structural constraint in this invention can effectively improve the cohesion of the community structure. The structural similarity of AGSCS and AGSCS+ is basically the same. This is because these two algorithms ensure that they are connected truss structures with maximum support values ​​when searching the preceding UL-Community. Furthermore, regardless of attribute optimization or the introduction of sign-in hyperedges, the algorithms always optimize the community under high structural constraints, thus maintaining their structural similarity at the same level.

[0076] Figure 12 The results shown are a comparison of sign-in density. First, we comprehensively observe the sign-in density of AGSCS and AGSCS+. Figure 10 The attribute scores show that although AGSCS+ has a slightly lower attribute score than AGSCS, its sign-in density is much higher. This is because AGSCS+ incorporates sign-in information into its score calculation before attribute optimization. Therefore, during Pareto optimization, it may remove some nodes with high attribute scores but poor sign-in performance, while retaining nodes with low attribute scores that increase community sign-in density. The final optimized solution is a comprehensive optimal solution that considers both attribute and sign-in hyperedge information. Although this may slightly decrease the attribute score, it significantly improves the sign-in density.

[0077] Then, observing the comparison results of check-in density between AGSCS+ and GCS, it can be seen that the community check-in density obtained by AGSCS+ is generally slightly lower than that of GCS on different datasets. This is because the principle of the GCS method is to take check-in density as the sole optimization objective, which does not impose strong constraints on the community structure and does not require the similarity of attributes between the community and the query node. Although the community obtained maximizes the check-in density, it contains many redundant nodes that are unrelated to the query node, which cannot meet the user's personalized query expectations. In summary, AGSCS+ is a community search method that can simultaneously satisfy the community's structural cohesion, attribute similarity, and check-in preference. Experiments have proven the effectiveness of the method of this invention.

[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A community search method with embedded preference enhancement, characterized in that: The specific steps are as follows: Step 1: Construct a model Check2Vec containing LBSN user node and location node embeddings. Ensure node proximity by maximizing the cosine similarity between the four types of nodes in the hypergraph (location, time, semantics, and user domain) and their best-fit line. Step 2: In the improved community search algorithm AGSCS+, the embedded vector representations of users and location nodes learned by Check2Vec, as well as the corresponding best-fit lines, are used to perform community search; Step 3: The AGSCS+ algorithm combines node attribute similarity and embedding similarity when calculating scores, and adjusts the impact of attribute information and embedding information on the search results community by inputting parameter p; Step 4: Perform jump walk sequence sampling in the user-location bipartite graph using a random walk strategy to obtain sequences of user and location nodes for learning embedded hyperedge information; Step 5: The Check2Vec algorithm learns node embeddings and uses the idea of ​​the best-fit line in linear regression to maximize the cosine similarity between the node embeddings and the best-fit line. Step 6: Use negative sampling techniques to learn vector representations of users and location nodes to ensure better capture of the relationships between nodes; Step 7: In the community search phase, the optimal user-location binary community is obtained by comprehensively considering the attribute similarity and embedding similarity of users and location nodes through the comprehensive scoring calculation formula in the AGSCS+ algorithm.

2. A community search system with embedded preference enhancement using the method of claim 1, characterized in that: This includes a social network node embedding device, a community search device, a random walk sequence acquisition device, and a system control device; The social network node embedding device learns the vector representations of users and location nodes, optimizes cosine similarity through a node embedding learning algorithm, and combines a random walk strategy to sample node sequences and embedding hyperedge sequences to provide informative node embedding representations for community search. The community search device uses the learned node vectors to construct a user-location bipartite graph, calculates the comprehensive score of the nodes, and obtains the optimal community through the community search algorithm. The random walk sequence acquisition device is a sequence acquisition engine that generates random walk sequences for users and location nodes and samples random walk sequences embedded with hyperedges. The system control device is a central control unit that coordinates the workflow of each device and manages data transmission and information exchange.

3. The community search system with embedded preference enhancement according to claim 2, characterized in that: The task of the social network node embedding device is to learn the vector representations of users and location nodes, optimize the cosine similarity in a best-fit line manner through a node embedding learning algorithm, thereby capturing the correlation between nodes; and generate sequences of users and location nodes, as well as sequences of sampled embedding hyperedges through a random walk strategy, providing rich and meaningful node embedding representations for community search.

4. The community search system with embedded preference enhancement according to claim 2, characterized in that: The community search device uses node vectors learned by Check2Vec to construct a user-location bipartite graph, realizing the core algorithm of community search. By calculating the comprehensive score of nodes and taking into account the weights of attribute information and embedding information, the system can more accurately evaluate the similarity between nodes. Finally, the optimal community is found through the community search algorithm, realizing community search based on embedding information.

5. The community search system with embedded preference enhancement according to claim 2, characterized in that: The random walk sequence acquisition device is responsible for generating random walk sequences for users and location nodes; by using a jump-walk strategy, it solves the problem of no relationship between structural relationships and location nodes, ensuring that the generated sequences are more diverse.

6. The community search system with embedded preference enhancement according to claim 2, characterized in that: The system control device acts as the central control unit of the entire system, responsible for coordinating the workflow between the Check2Vec model learning device, the community search algorithm device, and the random walk sequence acquisition device; through effective data transmission and information exchange, it ensures that each device operates in an orderly manner and works collaboratively.

7. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that: When the computer program / instructions are executed by the processor, they implement the steps of the method described in claim 1.

8. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by the processor, they implement the steps of the method described in claim 1.

Images