Method, device and equipment for recommending points of interest, storage medium and program product

Abstract

The invention provides an interest point recommendation method and device, electronic equipment, a computer readable storage medium and a computer program product, and relates to the technical field of artificial intelligence such as data processing, interest point analysis and intelligent recommendation. A specific embodiment of the method comprises: generating an associated predicted value based on characteristic characterization and prompt characterization of a point of interest and characteristic characterization and prompt characterization of a candidate point of interest, the prompt characterization being generated based on the characteristic characterization; and determining the candidate interest point as a recommended interest point of the interest point in response to the condition that the associated prediction value is greater than or equal to a prediction value threshold.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, specifically to the field of artificial intelligence technology such as data processing, point of interest analysis, and intelligent recommendation, and especially to methods, devices, electronic devices, computer-readable storage media, and computer program products for recommending points of interest. Background Art

[0002] Points of Interest (POI) refer to meaningful places marked on a map or in an app. For example, these places can be tourist attractions, restaurants, hotels, shops, gas stations, etc. POIs can provide users with useful information to help them find places or services of interest in a specific area.

[0003] In this context, the POIs previously used and searched can be referred to, and based on the associations between POIs, related and recommended POIs can be provided to users more proactively, in the hope of facilitating users and improving user experience. Alternatively, based on the associations between the reference POI and other POIs provided by the user for reference, other POIs that are associated with the reference POI can be recommended to the user. Therefore, how to improve the prediction accuracy and quality of the associations between POIs is worthy of attention and an urgent need. Summary of the invention

[0004] The embodiments of the present disclosure provide a method, an apparatus, an electronic device, a computer-readable storage medium, and a computer program product for recommending points of interest.

[0005] In a first aspect, an embodiment of the present disclosure proposes a method for recommending points of interest, including: generating an associated prediction value based on a characteristic representation and a prompt representation of the point of interest, and a characteristic representation and a prompt representation of a candidate point of interest, wherein the prompt representation is generated based on the characteristic representation; in response to the associated prediction value being greater than or equal to a prediction value threshold, determining the candidate point of interest as a recommended point of interest.

[0006] In a second aspect, an embodiment of the present disclosure proposes a device for recommending points of interest, including: an associated prediction value generating unit, configured to generate an associated prediction value based on a characteristic representation and a prompt representation of the point of interest, a characteristic representation and a prompt representation of a candidate point of interest, wherein the prompt representation is generated based on the characteristic representation; and a recommended point of interest determining unit, configured to determine a candidate point of interest as a recommended point of interest in response to an associated prediction value being greater than or equal to a prediction value threshold.

[0007] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that when the at least one processor executes, it is possible to implement the method for recommending points of interest as described in any implementation method in the first aspect.

[0008] In a fourth aspect, an embodiment of the present disclosure provides a non-transitory computer-readable storage medium storing computer instructions, which are used to enable a computer to implement the method for recommending points of interest as described in any implementation manner in the first aspect when executed.

[0009] In a fifth aspect, an embodiment of the present disclosure provides a computer program product including a computer program, which, when executed by a processor, can implement the method for recommending points of interest as described in any implementation manner in the first aspect.

[0010] The method, apparatus, electronic device, computer-readable storage medium, and computer program product for recommending points of interest provided by the embodiments of the present disclosure generate an associated prediction value based on the characteristic representation and prompt representation of the point of interest, and the characteristic representation and prompt representation of the candidate point of interest, wherein the prompt representation is generated based on the characteristic representation; in response to the associated prediction value being greater than or equal to a prediction value threshold, the candidate point of interest is determined as a recommended point of interest.

[0011] The present disclosure provides more information about points of interest by combining the characteristic representation of fixed information that is not easy to change and the prompt representation of dynamic information. This can improve the matching and recommendation quality of points of interest in the recommendation scheme for recommending related and other points of interest based on points of interest.

[0012] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Other features, objects and advantages of the present disclosure will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0014] Figure 1 is an exemplary system architecture in which the present disclosure may be applied;

[0015] Figure 2 A flowchart of a method for recommending points of interest provided by an embodiment of the present disclosure;

[0016] Figure 3A flowchart of another method for recommending points of interest provided by an embodiment of the present disclosure;

[0017] Figure 4 A flowchart of a method for recommending points of interest in an application scenario provided by an embodiment of the present disclosure;

[0018] Figure 5 A structural block diagram of a device for recommending points of interest provided in an embodiment of the present disclosure;

[0019] Figure 6 A schematic diagram of the structure of an electronic device suitable for executing a method for recommending points of interest provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0020] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description. It should be noted that the embodiments in the present disclosure and the features in the embodiments may be combined with each other without conflict.

[0021] In addition, in the technical solutions involved in the present disclosure, the acquisition, storage, use, processing, transportation, provision and disclosure of user personal information involved (for example, the points of interest used as reference for finding recommended interests in the subsequent disclosure, which may be provided by the user or determined based on the analysis results of the user's access records) shall comply with the relevant laws and regulations and shall not violate public order and good morals.

[0022] Figure 1 An exemplary system architecture 100 is shown to which embodiments of the method, apparatus, electronic device, and computer-readable storage medium for recommending points of interest of the present disclosure can be applied.

[0023] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0024] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 and server 105 can be installed with various applications for realizing information communication between the two, such as location recommendation applications, service location query applications, instant messaging applications, etc.

[0025] Terminal devices 101, 102, 103 and server 105 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, etc.; when terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above, which can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here. When server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server; when the server is software, it can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here.

[0026] The server 105 can provide various services through various built-in applications. Taking the location recommendation application that can provide location recommendation services as an example, the server 105 can achieve the following effects when running the location recommendation application: first, the points of interest are obtained from the terminal devices 101, 102, and 103 through the network 104; then, the server 105 generates an associated prediction value based on the characteristic representation and prompt representation of the points of interest, and the characteristic representation and prompt representation of the candidate points of interest, wherein the prompt representation is generated based on the characteristic representation; then, in response to the associated prediction value being greater than or equal to the prediction value threshold, the server 105 determines the candidate point of interest as a recommended point of interest.

[0027] It should be noted that, in addition to being obtained from the terminal devices 101, 102, 103 through the network 104, the points of interest can also be pre-stored locally in the server 105 in various ways. Therefore, when the server 105 detects that these data have been stored locally (for example, starting to process the previously reserved pending points of interest matching and recommendation relationship determination tasks), it can choose to directly obtain these data locally. In this case, the exemplary system architecture 100 may also not include the terminal devices 101, 102, 103 and the network 104.

[0028] Since storing points of interest, generating characteristic representations and prompt representations of points of interest, etc. require more computing resources and stronger computing power, the methods for recommending points of interest provided in the subsequent embodiments of the present disclosure are generally performed by the server 105 with stronger computing power and more computing resources, and accordingly, the device for recommending points of interest is generally also set in the server 105. However, it should also be pointed out that when the terminal devices 101, 102, and 103 also have computing power and computing resources that meet the requirements, the terminal devices 101, 102, and 103 can also complete the above-mentioned various operations that are handed over to the server 105 through the location recommendation application installed thereon, and then output the same result as the server 105. Especially in the case where there are multiple terminal devices with different computing powers at the same time, but when the location recommendation application determines that the terminal device where it is located has stronger computing power and more remaining computing resources, it can let the terminal device perform the above-mentioned operations, thereby appropriately reducing the computing pressure of the server 105, and accordingly, the device for recommending points of interest can also be set in the terminal devices 101, 102, and 103. In this case, the exemplary system architecture 100 may also not include the server 105 and the network 104 .

[0029] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0030] Please refer to Figure 2 , Figure 2 A flowchart of a method for recommending points of interest provided by an embodiment of the present disclosure, wherein process 200 includes the following steps:

[0031] Step 201, generating a correlation prediction value based on the characteristic representation and prompt representation of the interest point, and the characteristic representation and prompt representation of the candidate interest point;

[0032] This step is intended to be performed by the execution subject of the method for recommending points of interest (e.g. Figure 1 The server 105 shown generates an associated prediction value based on the characteristic representation and prompt representation of the interest point, and the characteristic representation and prompt representation of the candidate interest point.

[0033] The characteristic representation of a point of interest usually corresponds to the "fixed information that is not easily changed" of the point of interest. For example, the characteristic representation can be determined based on at least one of the following information of the point of interest (or POI): the geographical location of the point of interest, for example, the longitude and latitude coordinates of the geographical location corresponding to the point of interest, the specific street or city where the point of interest is located; the specific type of the point of interest, such as the restaurant, hotel, medical institution, educational institution to which the point of interest specifically points, or a more specific type, for example, whether the restaurant provides Chinese food, Western food, fast food, etc.; the specific name of the point of interest; the associated pictures of the point of interest, for example, the appearance and internal environment of the place corresponding to the point of interest; user comments of visiting users associated with the point of interest; the contact information of the place corresponding to the point of interest; and the business hours corresponding to the point of interest, etc.

[0034] Accordingly, the characteristic representation of the interest point can at least be based on the interest point dynamic relationship diagram under multiple time periods. arrive learning and processing to obtain the characteristic representation of interest points.

[0035] In the following, for the sake of ease of understanding, we will directly use “points of interest” as the input for learning and processing.

[0036] Furthermore, the prompt representation of the point of interest is often related to the change of time and is determined based on the dynamic change information of the point of interest. For example, the prompt representation can be used to correspondingly represent and reflect the traffic status of the point of interest at different time points (for example, the corresponding access traffic of the point of interest at different times, for example, when the point of interest is a restaurant, the prompt representation can be determined based on the user traffic information of the restaurant at different time points), for example, the association characteristics of the point of interest with other points of interest at a specific time point. For example, at "lunch time", for the points of interest representing shopping malls and the points of interest representing restaurants, for the points of interest representing shopping malls, the points of interest representing restaurants may be based on the time point of "lunch time", so that the points of interest representing restaurants have an additional degree of association compared to the points of interest representing cinemas and medical institutions. Or, the points of interest representing restaurants may be based on the time point of "lunch time", so that they have a higher degree of association than the correlation between the points of interest in shopping malls and the points of interest in restaurants corresponding to the midnight time period.

[0037] Therefore, by using the hint representation, we can "adjust" and "correct" the correlation between different points of interest in a more fine-grained manner and in combination with dynamic information, so as to more fine-grainedly and accurately mine those points of interest that have a closer correlation at the corresponding time point. For example, by using the hint representation, we can enhance or weaken the correlation determined by the characteristic representation based on different time points to mine recommended points of interest that are more likely to be selected by users.

[0038] Hint representations can usually be extracted and determined based on feature representations. For example, feature representations of points of interest can be processed through self-supervised learning tasks to determine the dynamic changes of the representations of points of interest and to determine hint representations of points of interest. For example, learning the changes of the characteristic of the flow of people at points of interest over time.

[0039] It should be understood that the characteristic representation may also be generated by other devices other than the execution entity (eg, terminal devices 101, 102, 103) and provided to the execution entity (eg, server 105), and the present disclosure is not intended to limit this.

[0040] Therefore, by jointly utilizing the characteristic representation of the "inherent information" indicating the point of interest and the prompt representation of the "dynamic information" indicating the point of interest, it is possible to analyze the correlation between the inherent information based on the characteristic representation, and based on the prompt representation, provide a reference and adjustment basis for the correlation analysis based on the "inherent information", so that in the process of matching and recommending points of interest, the "information" of the above two aspects can be combined to perform more fine-grained matching, so as to determine whether there is a satisfactory correlation between the points of interest and determine the recommended points of interest.

[0041] Similarly, for the candidate interest points, the corresponding characteristic representation and prompt representation can also be obtained in the same manner as described above, and will not be repeated here.

[0042] It should be noted that the characteristic representations of the points of interest, the candidate points of interest, and the prompt representations can be directly obtained by the above-mentioned execution subject from a local storage device, or from a non-local storage device (such as Figure 1 The local storage device may be a data storage module provided in the above-mentioned execution subject, such as a server hard disk, in which case the characteristic representation and prompt representation of the point of interest may be quickly read locally; the non-local storage device may also be any other electronic device provided for storing data, such as some user terminals, in which case the above-mentioned execution subject may obtain the required characteristic representation and prompt representation of the point of interest by sending an acquisition command to the electronic device.

[0043] The execution entity can generate an associated prediction value based on the characteristic representation and prompt representation of the point of interest and the characteristic representation and prompt representation of other candidate points of interest (candidate POIs), so as to facilitate the subsequent search for other points of interest matching the point of interest (i.e., points of interest that can be recommended) based on the associated prediction value.

[0044] For example, the execution subject may generate the similarity between the feature representations and the similarity between the prompt representations respectively, add the two similarities (or weighted addition), and use the sum of the similarities as the association prediction value.

[0045] In some embodiments, the execution subject may generate characteristic representations and prompt representations of other candidate points of interest in a similar manner before matching the points of interest. Thus, in the process of matching the points of interest, the execution subject may directly use the characteristic representations and prompt representations of these candidate points of interest to improve the matching efficiency. For example, in a scenario where points of interest are recommended to users in real time based on the points of interest provided by the user, the execution subject may use the characteristic representations and prompt representations of these generated candidate points of interest for matching and recommendation.

[0046] In some embodiments, the execution subject chooses to generate characteristic representations and prompt representations corresponding to all "points of interest" for each determined point of interest. Then, based on these characteristic representations and prompt representations, corresponding association prediction values ​​are generated for each pair of points of interest to analyze and mine the "association between points of interest." In this way, the association, matching, and recommendation relationship between points of interest can be determined through the analysis of the points of interest by the execution subject. Therefore, in the subsequent case of query demand (for example, when the user expects to query or be recommended points of interest), the execution subject can directly provide recommendation and matching results according to the points of interest selected by the user based on the determined association, matching, and recommendation relationship.

[0047] For the purpose of easy understanding, the following will only provide examples and explanations based on the scenario in which the execution subject preliminarily matches and mines the association relationships of the identified and determined points of interest.

[0048] Accordingly, if there are two or more candidate points of interest, the executing entity can choose to generate association prediction values ​​between the point of interest and each candidate point of interest respectively, so as to analyze and find the points of interest that have a matching or associated relationship with the point of interest among multiple or more points of interest, which can be used as recommended points of interest recommended as associated points of interest for the point of interest.

[0049] Step 202: In response to the associated prediction value being greater than or equal to the prediction value threshold, determining the candidate point of interest as a recommended point of interest.

[0050] This step is based on the above step 201. If the associated prediction value generated in step 201 is greater than or equal to the prediction value threshold (that is, a predetermined prediction value that can be reliably considered to be associated between two points of interest), the execution entity can respond to this to determine the candidate point of interest as a recommended point of interest.

[0051] The method for recommending points of interest provided by the embodiment of the present disclosure provides more information about points of interest by jointly providing characteristic representations of fixed information that is not easily changed and prompt representations of dynamic information. Thus, the matching and recommendation quality of points of interest in the recommendation scheme for recommending related and other points of interest based on points of interest can be improved.

[0052] In some embodiments, in order to improve the processing, generation efficiency and quality of feature representation, you can choose to use the interest point representation generation model, or more specifically, use the graph neural network (GNN) layer of the interest point representation generation model to process interest points to generate feature representations of interest points.

[0053] For example, sample interest points can be used as input in advance, and the sample characteristic representation of the sample interest points can be used as output. The GNN layer in the interest point representation generation model can be trained with a preset first loss function as supervision, so that the GNN layer has the ability to process interest points and generate corresponding sample characteristic representations.

[0054] Accordingly, when the value corresponding to the preset first loss function meets the corresponding training requirements (for example, it has been minimized, converged to the target value, etc.), the execution subject responds, completes the training of the GNN layer, and determines that the GNN layer has the ability to process interest points and generate corresponding sample feature representations.

[0055] Subsequently, the execution subject can use the interest point representation generation model including the trained GNN layer to process the interest points using the GNN layer to generate characteristic representations of the interest points, thereby processing the interest points more efficiently and with the same quality to generate characteristic representations of the interest points.

[0056] In some optional implementations of this embodiment, the interest point representation generation model may also be configured with a prompt representation generation layer for processing the characteristic representation of the interest point to generate a prompt representation. Thus, the interest point representation generation model may be used to process the interest point in a coherent and holistic manner to generate corresponding characteristic representations and prompt representations.

[0057] Accordingly, in such a case, the execution subject can similarly use the prompt representation generation layer of the interest point representation generation model to process the feature representation and generate the prompt representation of the interest point. For example, the prompt representation layer can be located after the GNN layer to process the feature representation output by the GNN layer to obtain the corresponding prompt representation.

[0058] As discussed above, dynamic information may also be multi-dimensional. Therefore, the prompt representation may actually be a combination of prompt representations in at least two different dimensions, so that the prompt representation can provide more reference information to achieve more accurate and high-quality interest point matching and recommended actions.

[0059] Specifically, the cue representation can actually include information in both the temporal dimension and the spatial dimension.

[0060] In this case, in some embodiments, the prompt representation generation layer can actually be composed of a first prompt representation generation layer and a second prompt representation generation layer. The first prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in the time dimension. The second prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in the spatial dimension.

[0061] As a result, the dynamic information of interest points in the time dimension and the space dimension can be aggregated in the "prompt representation".

[0062] In some optional implementations of this embodiment, if different dimensions exist, a prompt representation combination layer may be correspondingly configured in the interest point representation generation model to combine information of the interest point in different dimensions.

[0063] For example, in the case of information including two dimensions, time dimension and space dimension, the prompt representation combination layer in the interest point representation generation model can be used to combine the prompt representation of the interest point in the time dimension generated by the first prompt representation generation layer and the prompt representation of the interest point in the spatial dimension generated by the second prompt representation, and finally obtain the (complete) prompt representation corresponding to the interest point.

[0064] For example, the prompt representation combination layer can be connected to the first prompt representation generation layer and the second prompt representation generation layer respectively. For example, the prompt representation combination layer can provide two different interfaces or channels to simultaneously and separately connect to the first prompt representation generation layer and the second prompt representation generation layer.

[0065] Subsequently, the prompt representation combination layer may receive the prompt representations generated by the first prompt representation generation layer and the second prompt representation generation layer in corresponding dimensions respectively, and then combine the two to obtain a final prompt representation.

[0066] Therefore, by utilizing the hint representation combination layer, hint representations on different dimensions can be directly “combined and aggregated” in the interest point representation generation model, so as to utilize a separate “hint representation” to provide more information and improve the quality of the hint representation.

[0067] In some optional implementations of the present embodiment, for an interest point representation generation model that includes at least a graph neural network layer and a prompt representation generation layer, in order to improve the training and update efficiency in such an interest point representation generation model, it is also possible to configure the graph neural network layer and the prompt representation generation layer to be updated separately.

[0068] Specifically, if the execution entity receives a sample characteristic representation for the prompt representation layer and a sample prompt representation corresponding to the sample characteristic representation, the execution entity may select parameters of the fixed graph neural network layer in response thereto.

[0069] Then, the sample characteristic representation is used as the input of the prompt representation layer, and the sample prompt representation corresponding to the sample characteristic representation is used as the output. Based on the preset second loss function as supervision, the prompt representation layer is updated and trained.

[0070] Therefore, for example, when the prompt representation generation performance changes due to the adaptive adjustment of the "focus point" of dynamic information caused by different scenes and angles, and the interest point representation generation model needs to be adaptively adjusted and updated for training, the adaptation adjustment of the interest point representation generation model can be completed more easily, lightly and efficiently (for example, such an update method does not require training of the graph neural network layer).

[0071] In some embodiments, a prompt representation of the interest point in the time dimension can be generated based at least on the average sum of the characteristic representations of the interest point at each time point in a group of time nodes (for example, such a method can be implemented by the execution entity, or by the execution entity using the first prompt representation layer).

[0072] Specifically, for the interest point v i , the execution subject can choose to obtain the interest point at a set of time nodes {t1, t2...t n Characterization of the characteristics on} (where n is a positive integer)

[0073] Then, the execution subject may choose to generate the prompt representation of the interest point in the time dimension based on at least the average sum of the characteristic representations of the interest point at each time point in a set of time nodes based on the following formula (1): i :

[0074]

[0075] Where T is a set of time nodes {t1, t2...t n}.

[0076] Therefore, through the prompt representation in the time dimension, the independent representation of each point of interest across time can be retained, and the dynamic information in the time dimension can be introduced as a reference. In this way, context information from a time evolution perspective can be provided for subsequent processing actions to improve the matching and recommendation quality of points of interest.

[0077] In some optional implementations of this embodiment, when generating a prompt representation of the interest point in the time dimension, i In the process, random noise can also be added to enhance the robustness and expressiveness of the interest point representation generation model through random noise, so that the interest point representation generation model can explore more diverse (feature, vector) spaces.

[0078] Specifically, the hint representation of the interest point in the time dimension is i It can be further expressed as the following formula (2):

[0079]

[0080] Where T is a set of time nodes {t1, t2...t n},w i is random noise, and the vector of random noise satisfies the uniform distribution from -σ to σ, that is, w i ~u(-σ,σ).

[0081] In some optional implementations of this embodiment, σ can be based on the total number of interest points, candidate interest points, and prompt representation e i The actual representation dimension (e.g., vector dimension) is determined.

[0082] For example, σ can be determined based on the following formula (3):

[0083]

[0084] Where |V| represents the total number of the above-mentioned points of interest and candidate points of interest, and d is the prompt representation e i The actual expression dimension of , and a is a constant determined based on the actual scenario and prior knowledge.

[0085] In some embodiments, the hint representation of the interest point in the spatial dimension is generated based on the following method (for example, such a method can be implemented by the execution subject, or by the execution subject using the second hint representation layer):

[0086] First, a set of reference interest points is determined based on the maximum influence distance of the interest points.

[0087] Specifically, the execution subject may choose to process the interest point map including each interest point, and determine the farthest distance that each interest point can affect.

[0088] For example, for the interest point v i , the maximum impact distance ρ i , can be determined based on the following formula (4):

[0089] ρ i =max{d i ,j|((v i , v j )∈ε i} (4)

[0090] Among them, ε i Indicates the point of interest v i The set of all interest points with adjacent relationships, interest point v j is i , distance from the point of interest v i The farthest, affected point of interest.

[0091] Then, the execution subject determines the corresponding “maximum influence distance” for each interest point and candidate interest point in a similar way, and obtains the “maximum influence distance” set {ρ1, .., ρ |V|}.

[0092] Then execute the main body, based on the interest point v i A set of reference interest points is determined based on the corresponding maximum influence distance. The difference between the maximum influence distance of the reference interest point in the set of reference interest points and the maximum influence distance of the interest point is less than or equal to a difference threshold (the difference threshold is usually set based on the "difference" allowed in practice).

[0093] In some embodiments, the execution entity may also select and maintain a set of interest points based on the threshold standard, so as to extract a set of reference interest points corresponding to the interest point using the set of interest points.

[0094] Specifically, the execution subject can determine the “maximum influence distance” set {ρ1, .., ρ |V|}, and then discretize them, and store the "maximum influence distances" that can be considered to belong to each other and can be referenced into the same "interest point set". That is, in each interest point set, for any interest point, other interest points can be used as reference interest points that meet the requirements.

[0095] Next, the execution subject can generate a hint representation of the interest point in the spatial dimension based on at least the average pooling result of the characteristic representation of the reference interest point at each time point in a set of time nodes. For example, the execution subject can generate a hint representation S(ρ) of the interest point in the spatial dimension based on the following formula (5): i ):

[0096]

[0097] Where l is the total number of interest points and a set of reference interest points, bin(ρ i ) is the point of interest v i A corresponding set of reference points of interest, is a reference interest point v in a set of reference interest points k Characterization at time node t.

[0098] Therefore, the "interest point influence" of the interest point in the spatial dimension can be determined based on the reference interest point with the maximum influence distance similarity after discretization, so as to introduce dynamic information in the spatial dimension as a reference to improve the matching and recommendation quality of the interest point.

[0099] In some optional implementations of this embodiment, when generating a prompt representation of an interest point in the spatial dimension, random noise may be similarly introduced to enhance the learning and processing capabilities in the spatial dimension.

[0100] That is, in the process of generating a prompt representation of the interest point in the spatial dimension by averaging the pooling results of the feature representation of the reference interest point at each time point in a set of time nodes, the execution entity can choose to combine the average pooling results of the feature representation of the reference interest point at each time point in a set of time nodes and random noise to generate a prompt representation of the interest point in the spatial dimension.

[0101] Specifically, the hint representation S(ρ) of the interest point in the spatial dimension i ) can be further expressed as the following formula (6):

[0102]

[0103] in, is random noise, and the vector of random noise satisfies the uniform distribution from -σ to σ, that is, σ can be similarly determined based on the method discussed above, and will not be repeated here.

[0104] It should be understood that, for random noise, if there is a prompt representation that actually includes both time and space dimensions, the execution entity may also choose to add the random noise only to the result of combining the two.

[0105] In some embodiments, in the process of generating an associated prediction value based on the characteristic representation and prompt representation of the point of interest, and the characteristic representation and prompt representation of the candidate point of interest, the execution entity may actually choose to use the prompt representation of the point of interest in the spatial dimension to drive the time-aware bilinear function to generate an associated prediction value based on the characteristic representation and prompt representation of the point of interest, and the characteristic representation and prompt representation of the candidate point of interest.

[0106] Temporal Bilinear Function is a mathematical tool for modeling systems with time dependence, which can be used to process time series data to capture complex patterns of temporal changes in the data.

[0107] Therefore, using the perceptual bilinear function, it is possible to more accurately determine and capture dynamic information based on time changes. Take the prompt representation that provides information on both the time dimension and the space dimension as an example. Later, for the case where only one dimension is considered, it can be similarly achieved by "deleting" the corresponding part.

[0108] To this end, we can first determine the interest point v based on feature representation and prompt representation. i Actual prompt representation P i (e.g., a combination of spatial and temporal dimensions, for example, the process may be implemented by a hint representation combination layer), P i It can be expressed as the following formula (7):

[0109] P i =e i +S(ρ i ) (7)

[0110] Next, the execution subject can be based on the point of interest v i Characterization at time node t and prompt representation P i , the execution subject can choose to use the hint representation of the interest point in the spatial dimension to drive the time-aware bilinear function based on the following formula (8) to generate the characteristic representation and hint representation of the interest point, and the correlation prediction value between the characteristic representation and hint representation of the candidate interest point

[0111]

[0112] in, is the activation function (e.g., based on the interest point v i a temporal perception bilinear function driven by cue representation in the spatial dimension), is the point of interest v iThe characteristic representation at time node t, for each time t the diagonal matrix represents the learnable weight matrix for relation r, is the point of interest v j The characteristic representation at time node t indicates the representation P j is the point of interest v j Characterization of the characteristics, interest point v j It can be used to determine whether the query is for the point of interest v i Matched candidate interest points.

[0113] In some embodiments, for the associated prediction value, the execution entity may also select a multi-time task based on a cross entropy loss function to update it to further improve its quality.

[0114] Please refer to Figure 3 , Figure 3 A flowchart of another process of recommending points of interest provided by an embodiment of the present disclosure, wherein process 300 includes the following steps:

[0115] Step 301, generating a correlation prediction value based on the characteristic representation and prompt representation of the interest point, and the characteristic representation and prompt representation of the candidate interest point;

[0116] The above step 301 and the above step 302 Figure 2 In the process 200 shown, step 301 is equivalent and can be implemented similarly with reference to the above description, and will not be repeated here.

[0117] Step 302, using a multi-time task based on a cross entropy loss function, updating the associated prediction value to obtain an updated associated prediction value;

[0118] The cross entropy loss function can be used to evaluate and optimize the performance of time series prediction in multi-time tasks. The cross entropy loss function can be configured based on the understanding of Shannon entropy and KL divergence. Shannon entropy measures the uncertainty of information, while KL divergence is used to measure the difference between two probability distributions. Accordingly, by minimizing the cross entropy loss, its predicted distribution can be gradually adjusted, thereby improving the accuracy of the processing results.

[0119] Therefore, the multi-time task based on the cross entropy loss function can make the associated prediction value measure the gap between the prediction result and the actual label in terms of dynamic information associated with time, so as to improve the prediction quality of the associated prediction value.

[0120] Specifically, the execution subject can use the following formula (9) to calculate the associated prediction value: To update:

[0121]

[0122] Where T is a set of time nodes {t1, t2...t n}, Represents a set of interest points used to learn and train the association relationship between interest points. represents the truth value of the relation at time t.

[0123] Step 303: in response to the updated associated prediction value being greater than or equal to the prediction value threshold, determining the candidate point of interest as a recommended point of interest.

[0124] In this step, the executing entity can further use the "updated associated prediction value" as the detection object in a manner similar to that discussed above for step 203, and when the updated associated prediction value is greater than or equal to the prediction value threshold (for example, the prediction value threshold involved in the above step 203), select a response to determine the candidate point of interest as a recommended point of interest.

[0125] To deepen understanding, the present disclosure also provides a specific implementation solution in combination with a specific application scenario, see Figure 4 The process 400 shown in FIG. 4 is shown in FIG. 400. For ease of understanding, reference may be made to Figure 1 An exemplary system architecture 100 is shown for illustration.

[0126] For example, the server 105 may be the execution subject of the process 400 .

[0127] In process 400 , the point of interest 410 may be input into the point of interest representation generation model 420 by the server 105 , so that the server 105 may process the point of interest 410 using the graph neural network 421 of the point of interest representation generation model 420 to obtain a characteristic representation 413 of the point of interest 410 .

[0128] Then, the server 105 may continue to utilize the prompt representation generation layer 422 of the interest point representation generation model 420 to process the characteristic representation 413 to generate a prompt representation 432 in the spatial dimension of the interest point 410; and utilize the prompt representation generation layer 423 to process the characteristic representation 413 to generate a prompt representation 431 in the time dimension of the interest point 410.

[0129] Next, the server 105 uses the prompt representation combination layer 424 of the interest point representation generation model 420 to combine the prompt representation 431 in the time dimension and the prompt representation 432 in the space dimension to obtain the prompt representation 435 .

[0130] Accordingly, after obtaining the prompt representation 435 , the server 105 may choose to add random noise to the prompt representation 435 to obtain the prompt representation 435 ′.

[0131] It should be understood that the action of adding random noise may also be configured to be implemented by the interest point representation generation model 420 , and the present disclosure is not intended to be limited in this regard.

[0132] After obtaining the characteristic representation 413 and the prompt representation 435', the server 105 can determine the interest point 410 and the candidate interest point 441, determine the interest point 410 and the candidate interest point 442... and determine the corresponding association prediction value of the interest point 410 and the candidate interest point 44N (N is a positive integer) based on the characteristic representation 413 and the prompt representation 435' respectively.

[0133] For example, for candidate interest point 441 , server 105 may determine association prediction value 471 between interest point 410 and candidate interest point 441 based on characteristic representation 413 and prompt representation 435 ′, characteristic representation 451 corresponding to candidate interest point 441 , and prompt representation 461 .

[0134] Similarly, server 105 may determine an association prediction value 472 between point of interest 410 and candidate point of interest 442 based on characteristic representation 452 and prompt representation 462 corresponding to candidate point of interest 442... and determine an association prediction value 47N between point of interest 410 and candidate point of interest 44N based on characteristic representation 45N and prompt representation 46N corresponding to candidate point of interest 44N.

[0135] Exemplarily, if the associated prediction values ​​471 and 472 are greater than or equal to the prediction value threshold, the server 105 may accordingly determine that the candidate points of interest 441 and 442 are recommended points of interest for the point of interest 410 .

[0136] Further references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a device for recommending points of interest. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0137] like Figure 5 As shown, the device 500 for recommending interest points of this embodiment may include: an associated prediction value generating unit 501 and a recommended interest point determining unit 502. The associated prediction value generating unit 501 is configured to generate an associated prediction value based on the characteristic representation and prompt representation of the interest point, the characteristic representation and prompt representation of the candidate interest point, wherein the prompt representation is generated based on the characteristic representation; the recommended interest point determining unit 502 is configured to determine the candidate interest point as a recommended interest point in response to the associated prediction value being greater than or equal to the prediction value threshold.

[0138] In the device for recommending points of interest 500, the specific processing of the associated prediction value generating unit 501 and the recommended points of interest determining unit 502 and the technical effects thereof can be referred to in the respective Figure 2 The relevant descriptions of steps 201 - 202 in the corresponding embodiment are not repeated here.

[0139] In some optional implementations of this embodiment, the device 500 also includes: a feature representation generation unit, which is configured to use the graph neural network layer of the interest point representation generation model to process the interest point and generate a feature representation of the interest point.

[0140] In some optional implementations of this embodiment, the apparatus 500 further includes: a prompt representation generation unit configured to utilize the prompt representation generation layer of the interest point representation generation model to process the characteristic representation and generate the prompt representation of the interest point.

[0141] In some optional implementations of this embodiment, the prompt representation generation layer includes a first prompt representation generation layer and a second prompt representation generation layer, the first prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in the time dimension, and the second prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in the spatial dimension.

[0142] In some optional implementations of this embodiment, the device 500 also includes: a prompt representation combination unit, which is configured to use the prompt representation combination layer of the interest point representation generation model to combine the prompt representation of the interest point in the time dimension and the prompt representation of the interest point in the spatial dimension to obtain the prompt representation of the interest point.

[0143] In some optional implementations of this embodiment, the prompt representation includes a prompt representation of the interest point in the time dimension, and the prompt representation of the interest point in the time dimension is generated based on the following method: at least based on the average sum of the characteristic representations of the interest point at each time point in a group of time nodes, the prompt representation of the interest point in the time dimension is generated.

[0144] In some optional implementations of the present embodiment, a prompt representation of the input interest point in the time dimension is generated based at least on the average sum of the characteristic representations of the input interest point at each time point in a group of time nodes, including: combining the average sum of the characteristic representations of the input interest point at each time point in a group of time nodes with random noise to generate a prompt representation of the input interest point in the time dimension.

[0145] In some optional implementations of the present embodiment, the prompt representation includes a prompt representation of the interest point in the spatial dimension, and the prompt representation of the interest point in the spatial dimension is generated based on the following method: based on the maximum influence distance of the interest point, a group of reference interest points is determined, wherein the difference between the maximum influence distance of the reference interest point in a group of reference interest points and the maximum influence distance of the interest point is less than or equal to a difference threshold; at least based on the average pooling result of the characteristic representation of the reference interest point at each time point in a group of time nodes, the prompt representation of the interest point in the spatial dimension is generated.

[0146] In some optional implementations of the present embodiment, a prompt representation of the interest point in the spatial dimension is generated based at least on the average pooling result of the characteristic representation of the reference interest point at each time point in a set of time nodes, including: combining the average pooling result of the characteristic representation of the reference interest point at each time point in a set of time nodes and random noise to generate a prompt representation of the interest point in the spatial dimension.

[0147] In some optional implementations of the present embodiment, the device 500 also includes: a graph neural network layer fixing unit, configured to fix the parameters of the graph neural network layer in response to receiving a sample characteristic representation for the prompt representation layer and a sample prompt representation corresponding to the sample characteristic representation; a prompt representation layer updating unit, configured to use the sample characteristic representation as an input to the prompt representation layer and the sample prompt representation corresponding to the sample characteristic representation as an output, and to update and train the prompt representation layer based on a preset loss function as supervision.

[0148] In some optional implementations of the present embodiment, the associated prediction value generating unit 501 is further configured to utilize the prompt representation of the interest point in the spatial dimension to drive the time-aware bilinear function, and to generate an associated prediction value based on the characteristic representation and prompt representation of the interest point, and the characteristic representation and prompt representation of the candidate interest point.

[0149] In some optional implementations of this embodiment, the device 500 also includes: an associated prediction value updating unit, configured to update the associated prediction value using a multi-time task based on a cross entropy loss function to obtain an updated associated prediction value; and a recommended interest point determination unit 502, further configured to, in response to the updated associated prediction value being greater than or equal to a prediction value threshold, determine the candidate interest point as a recommended interest point.

[0150] This embodiment exists as an apparatus embodiment corresponding to the above method embodiment. The apparatus for recommending points of interest provided by this embodiment provides more information about points of interest by jointly providing characteristic representations of fixed information that is not easily changed and prompt representations of dynamic information. As a result, the matching and recommendation quality of points of interest in the recommendation scheme for recommending related and other points of interest based on points of interest can be improved.

[0151] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0152] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0153] like Figure 6 As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0154] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0155] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 601 performs the various methods and processes described above, such as a method for recommending points of interest. For example, in some embodiments, the method for recommending points of interest may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the method for recommending points of interest described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform the method for recommending points of interest in any other appropriate manner (e.g., by means of firmware).

[0156] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0157] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0158] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0159] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0160] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0161] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and virtual private servers (VPS) services. The server can also be divided into a server of a distributed system, or a server combined with a blockchain.

[0162] According to the technical solution of the embodiment of the present disclosure, more information about points of interest is provided by jointly providing characteristic representations of fixed information that is not easily changed and prompt representations of dynamic information. This can improve the matching and recommendation quality of points of interest in the recommendation scheme for recommending related and other points of interest based on points of interest.

[0163] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions provided by this disclosure can be achieved, and this document does not limit this.

[0164] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for recommending points of interest, comprising: generating an associated prediction value based on the characteristic representation and the prompt representation of the interest point, the characteristic representation and the prompt representation of the candidate interest point, wherein the prompt representation is generated based on the characteristic representation; In response to the associated prediction value being greater than or equal to a prediction value threshold, determining the candidate point of interest as a recommended point of interest.

2. The method according to claim 1, further comprising: The interest points are processed using a graph neural network layer of an interest point representation generation model to generate feature representations of the interest points.

3. The method according to claim 2, further comprising: The feature representation is processed by using the prompt representation generation layer of the interest point representation generation model to generate a prompt representation of the interest point.

4. The method according to claim 3, wherein: The prompt representation generation layer includes a first prompt representation generation layer and a second prompt representation generation layer. The first prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in a time dimension. The second prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in a spatial dimension.

5. The method according to claim 4, further comprising: The prompt representation combination layer of the interest point representation generation model is used to combine the prompt representation of the interest point in the time dimension and the prompt representation of the interest point in the space dimension to obtain the prompt representation of the interest point.

6. The method according to claim 1, wherein: The prompt representation includes a prompt representation of the interest point in a time dimension, and the prompt representation of the interest point in a time dimension is generated based on the following method: A prompt representation of the interest point in the time dimension is generated based at least on an average summation result of the characteristic representations of the interest point at each time point in a group of time nodes.

7. The method according to claim 6, wherein: The step of generating a prompt representation of the input interest point in a time dimension based at least on an average sum of characteristic representations of the input interest point at each time point in a group of time nodes comprises: The average summation result of the characteristic representation of the input interest point at each time point in a group of time nodes is combined with random noise to generate a prompt representation of the input interest point in the time dimension.

8. The method according to claim 1, wherein: The prompt representation includes a prompt representation of the interest point in a spatial dimension, and the prompt representation of the interest point in a spatial dimension is generated based on the following method: Determining a group of reference points of interest based on the maximum influence distance of the point of interest, wherein the difference between the maximum influence distance of the reference points of interest in the group of reference points of interest and the maximum influence distance of the point of interest is less than or equal to a difference threshold; A prompt representation of the interest point in the spatial dimension is generated based at least on an average pooling result of the characteristic representation of the reference interest point at each time point in a group of time nodes.

9. The method according to claim 8, wherein: The generating a prompt representation of the interest point in a spatial dimension based at least on an average pooling result of the characteristic representation of the reference interest point at each time point in a group of time nodes comprises: An average pooling result based on the characteristic representation of the reference interest point at each time point in a group of time nodes and random noise are combined to generate a prompt representation of the interest point in the spatial dimension.

10. The method according to claim 3, further comprising: In response to receiving a sample characteristic representation for the prompt representation layer and a sample prompt representation corresponding to the sample characteristic representation, fixing parameters of the graph neural network layer; The sample characteristic representation is used as the input of the prompt representation layer, the sample prompt representation corresponding to the sample characteristic representation is used as the output, and the prompt representation layer is updated and trained based on a preset loss function as supervision.

11. The method according to claim 6, wherein: The generating of the associated prediction value based on the characteristic representation and the prompt representation of the interest point, the characteristic representation and the prompt representation of the candidate interest point, comprises: The hint representation of the interest point in the spatial dimension is used to drive the time-aware bilinear function, and the associated prediction value is generated based on the characteristic representation and hint representation of the interest point, and the characteristic representation and hint representation of the candidate interest point.

12. The method according to claim 11, further comprising: Using a multi-time task based on a cross entropy loss function, updating the associated prediction value to obtain an updated associated prediction value; as well as In response to the associated prediction value being greater than or equal to a prediction value threshold, determining the candidate point of interest as the recommended point of interest includes: In response to the updated associated prediction value being greater than or equal to a prediction value threshold, the candidate point of interest is determined as a recommended point of interest.

13. A device for recommending points of interest, comprising: an association prediction value generating unit configured to generate an association prediction value based on the characteristic representation and the prompt representation of the interest point, the characteristic representation and the prompt representation of the candidate interest point, wherein the prompt representation is generated based on the characteristic representation; The recommended interest point determination unit is configured to determine the candidate interest point as a recommended interest point for the interest point in response to the associated prediction value being greater than or equal to a prediction value threshold.

14. The apparatus according to claim 13, further comprising: The feature representation generation unit is configured to process the interest point using the graph neural network layer of the interest point representation generation model to generate the feature representation of the interest point.

15. The apparatus according to claim 14, further comprising: The prompt representation generating unit is configured to process the feature representation by using the prompt representation generating layer of the interest point representation generating model to generate the prompt representation of the interest point.

16. The device according to claim 15, wherein: The prompt representation generation layer includes a first prompt representation generation layer and a second prompt representation generation layer. The first prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in a time dimension. The second prompt representation layer processes the characteristic representation to generate a prompt representation of the interest point in a spatial dimension.

17. The apparatus according to claim 16, further comprising: The prompt representation combining unit is configured to use the prompt representation combining layer of the interest point representation generation model to combine the prompt representation of the interest point in the time dimension and the prompt representation of the interest point in the space dimension to obtain the prompt representation of the interest point.

18. The device according to claim 13, wherein: The prompt representation includes a prompt representation of the interest point in a time dimension, and the prompt representation of the interest point in a time dimension is generated based on the following method: A prompt representation of the interest point in the time dimension is generated based at least on an average summation result of the characteristic representations of the interest point at each time point in a group of time nodes.

19. The device according to claim 18, wherein: The step of generating a prompt representation of the input interest point in a time dimension based at least on an average sum of characteristic representations of the input interest point at each time point in a group of time nodes comprises: The average summation result of the characteristic representation of the input interest point at each time point in a group of time nodes is combined with random noise to generate a prompt representation of the input interest point in the time dimension.

20. The device according to claim 13, wherein: The prompt representation includes a prompt representation of the interest point in a spatial dimension, and the prompt representation of the interest point in a spatial dimension is generated based on the following method: Determining a group of reference points of interest based on the maximum influence distance of the point of interest, wherein the difference between the maximum influence distance of the reference points of interest in the group of reference points of interest and the maximum influence distance of the point of interest is less than or equal to a difference threshold; A prompt representation of the interest point in the spatial dimension is generated based at least on an average pooling result of the characteristic representation of the reference interest point at each time point in a group of time nodes.

21. The device according to claim 20, wherein: The generating a prompt representation of the interest point in a spatial dimension based at least on an average pooling result of the characteristic representation of the reference interest point at each time point in a group of time nodes comprises: An average pooling result based on the characteristic representation of the reference interest point at each time point in a group of time nodes and random noise are combined to generate a prompt representation of the interest point in the spatial dimension.

22. The apparatus according to claim 15, further comprising: A graph neural network layer fixing unit, configured to fix parameters of the graph neural network layer in response to receiving a sample characteristic representation for the prompt representation layer and a sample prompt representation corresponding to the sample characteristic representation; The prompt representation layer updating unit is configured to use the sample characteristic representation as the input of the prompt representation layer, and the sample prompt representation corresponding to the sample characteristic representation as the output, and update the prompt representation layer based on a preset loss function as supervision.

23. The device according to claim 18, wherein The associated prediction value generating unit is further configured to utilize the prompt representation of the interest point in the spatial dimension to drive the time-aware bilinear function, and to generate an associated prediction value based on the characteristic representation and prompt representation of the interest point, and the characteristic representation and prompt representation of the candidate interest point.

24. The apparatus according to claim 23, further comprising: an associated prediction value updating unit, configured to update the associated prediction value by using a multi-time task based on a cross entropy loss function to obtain an updated associated prediction value; as well as The recommended interest point determination unit is further configured to, in response to the updated associated prediction value being greater than or equal to a prediction value threshold, determine the candidate interest point as the recommended interest point of the interest point.

25. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for recommending points of interest according to any one of claims 1 to 12.

26. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method for recommending points of interest according to any one of claims 1 to 12.

27. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method for recommending points of interest according to any one of claims 1 to 12.

Images