Information prediction method and device, electronic equipment and storage medium

Abstract

This application relates to the field of machine learning technology, and more particularly to an information prediction method, apparatus, electronic device, and storage medium, specifically applicable to map scenarios or travel scenarios. The method includes: generating a target relationship graph based on the interaction relationship features between the object to be predicted and the points of interest to be predicted, and the object relationship features between the objects to be predicted; inputting the historical interaction features, the access features, and the object relationship features into a target feature fusion network, aggregating the historical interaction features and the access features of the neighboring nodes of each node through the target feature fusion network to obtain access prediction information of the object to be predicted for the target points of interest; and determining the target access object corresponding to the target points of interest from the objects to be predicted based on the access prediction information. This application can improve the accuracy of access object prediction.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to an information prediction method, apparatus, electronic device, and storage medium. Background Technology

[0002] In map and travel scenarios, with the increasing popularity of location-aware social media applications, researchers have conducted extensive research on point-of-interest (POI) prediction. Most existing PPI research explores this from the user's perspective, helping users discover new points of interest to enrich their experience; however, this doesn't meet the needs of real-world data mining. For example, given a point of interest, service providers at that point want to predict which users will visit it next, as knowing in advance allows for better personalized service. However, existing user-POI-based prediction methods suffer from inaccurate predictions, leading to inaccurate predictions of users who will visit a given point of interest, thus hindering the provision of personalized services. Summary of the Invention

[0003] The technical problem to be solved by this application is to provide an information prediction method, apparatus, electronic device and storage medium that can improve the accuracy of the fusion features of each object to be predicted, and then, given a point of interest, predict the object that will visit the point of interest based on the accurate fusion features, thereby improving the accuracy of the prediction of the visiting object and facilitating the provision of personalized services to the object to be visited.

[0004] To address the aforementioned technical problems, this application provides an information prediction method, which may include:

[0005] Based on the interaction relationship features between the object to be predicted and the points of interest to be predicted, as well as the object relationship features between the objects to be predicted, a target relationship graph is generated; each node in the target relationship graph corresponds to an object to be predicted; the interaction relationship features of the objects to be predicted include the historical interaction features between the objects to be predicted and the points of interest to be predicted, and the access features of the objects to be predicted to the points of interest to be predicted, wherein the access features characterize the ease with which the objects to be predicted access the points of interest to be predicted;

[0006] The historical interaction features, the access features, and the object relationship features are input into the target feature fusion network. The target feature fusion network aggregates the historical interaction features and access features of the neighboring nodes of each node to update the historical interaction features and access features of each node, thereby obtaining the fusion features of the object to be predicted corresponding to each node.

[0007] The fusion features of the object to be predicted and the interest point features of the target interest point are input into the target prediction network to predict access information, thereby obtaining the access prediction information of the object to be predicted to the target interest point; the target interest point is at least one of the interest points to be predicted.

[0008] Based on the access prediction information, the target access object corresponding to the target point of interest is determined from the objects to be predicted.

[0009] On the other hand, this application provides an information prediction device, comprising:

[0010] The relationship graph generation module is used to generate a target relationship graph based on the interaction relationship features between the object to be predicted and the points of interest to be predicted, as well as the object relationship features between the objects to be predicted. Each node in the target relationship graph corresponds to an object to be predicted. The interaction relationship features of the object to be predicted include the historical interaction features between the object to be predicted and the points of interest to be predicted, as well as the access features of the object to be predicted to the points of interest to be predicted. The access features characterize the ease with which the object to be predicted can access the points of interest to be predicted.

[0011] The feature fusion module is used to input the historical interaction features, the access features, and the object relationship features into the target feature fusion network, and to aggregate the historical interaction features and access features of the neighboring nodes of each node through the target feature fusion network, so as to update the historical interaction features and access features of each node, and obtain the fusion features of the object to be predicted corresponding to each node.

[0012] The information prediction module is used to input the fusion features of the object to be predicted and the interest point features of the target interest point into the target prediction network to predict access information, thereby obtaining the access prediction information of the object to be predicted to the target interest point; the target interest point is at least one of the interest points to be predicted.

[0013] The object determination module is used to determine the target access object corresponding to the target point of interest from the objects to be predicted based on the access prediction information.

[0014] On the other hand, this application provides an electronic device including a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the information prediction method as described above.

[0015] On the other hand, this application provides a computer storage medium storing at least one instruction or at least one program, wherein the at least one instruction or at least one program is loaded by a processor and executed as described above in the information prediction method.

[0016] On the other hand, this application provides a computer program product including computer instructions stored in a computer-readable storage medium; a processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions to cause the electronic device to perform the above-described information prediction method.

[0017] Implementing the embodiments of this application has the following beneficial effects:

[0018] In this embodiment, a target relationship graph is generated based on the interaction relationship features between the object to be predicted and the points of interest to be predicted, as well as the object relationship features between the objects to be predicted, to facilitate the fusion of interaction relationship features and object relationship features. The interaction relationship features include historical interaction features and access features that characterize the ease with which the object to be predicted accesses the points of interest to be predicted. This allows the interaction relationship features of the object to be predicted to better reflect the association information between the object to be predicted and the points of interest to be predicted from multiple dimensions, improving the accuracy of the fused features of the object to be predicted. Furthermore, for each node in the target relationship graph, the target feature fusion network can aggregate the historical interaction features and access features of each node's neighboring nodes, and update the historical interaction features and access features of each node based on the aggregation results, obtaining the fused features of the object to be predicted for each node. This ensures that each object to be predicted can update its own interaction relationship features based on the interaction relationship features of its friend objects, further improving the accuracy of the fused features of the object to be predicted. The accurate fused features of the object to be predicted can deeply express the hidden association information between the object to be predicted and each point of interest to be predicted. Therefore, based on the given target interest point and the accurate fusion features of the object to be predicted, the prediction of the object that is about to visit the target interest point can be improved, which makes it easier to provide personalized services for the object that is about to visit. Attached Figure Description

[0019] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1This is a schematic diagram of the implementation environment provided in the embodiments of this application;

[0021] Figure 2 This is a flowchart of an information prediction method provided in an embodiment of this application;

[0022] Figure 3 This is a flowchart of the distance feature determination method provided in the embodiments of this application;

[0023] Figure 4 This is a flowchart of a historical interaction feature generation method provided in an embodiment of this application;

[0024] Figure 5 This is a flowchart of a predictive network update method provided in an embodiment of this application;

[0025] Figure 6 This is a schematic diagram of a network based on point-of-interest interaction provided in an embodiment of this application;

[0026] Figure 7 This is a schematic diagram of the information prediction process provided in the embodiments of this application;

[0027] Figure 8 This is a schematic diagram of an information prediction device provided in an embodiment of this application;

[0028] Figure 9 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0031] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0032] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0033] Please see Figure 1 The illustration shows an implementation environment provided in the embodiments of this application. The implementation environment may include at least one target terminal 110 and a prediction server 120, which can communicate with each other via a network.

[0034] Specifically, the target terminal 110 can send an information prediction request to the prediction server 120, which may include a target point of interest (POI). Upon receiving the information prediction request, the prediction server 120 can acquire the POI features of the target POI and input the fusion features of the POI and the target object to be predicted into the target prediction network for access information prediction, thereby obtaining the target access object that will visit the target POI among the objects to be predicted. The prediction server 120 can be a server corresponding to a map platform, which can predict the target access object that will visit the target POI based on the information prediction request sent by the target terminal 110 based on the map platform. Alternatively, the prediction server 120 can be a server corresponding to a travel platform, which can predict the target access object that will visit the target POI based on the information prediction request sent by the target terminal 110 based on the travel platform. The map platform can be a platform that provides map services, such as trip navigation and route planning; the travel platform can be a platform that provides travel services, such as ride-hailing and car rental services.

[0035] The target terminal 110 can communicate with the prediction server 120 based on a browser / server (B / S) or client / server (C / S) model. The target terminal 110 may include physical devices such as smartphones, tablets, laptops, digital assistants, smart wearable devices, in-vehicle terminals, and servers, and may also include software running on the physical device, such as applications. The operating system running on the target terminal 110 in this embodiment may include, but is not limited to, Android, iOS, Linux, and Windows.

[0036] The prediction server 120 and the target terminal 110 can establish a communication connection via wired or wireless means. The prediction server 120 may include a stand-alone server, a distributed server, or a server cluster consisting of multiple servers, wherein the server may be a cloud server.

[0037] To address the technical problem in existing technologies where the predicted users who will visit a given point of interest are inaccurate, thus hindering the provision of personalized services, this application provides an information prediction method. The method can be executed by the aforementioned prediction server. (See also...) Figure 2 The method may include:

[0038] S210. Based on the interaction relationship features between the object to be predicted and the point of interest to be predicted, and the object relationship features between the objects to be predicted, a target relationship graph is generated; each node in the target relationship graph corresponds to an object to be predicted; the interaction relationship features of the object to be predicted include the historical interaction features between the object to be predicted and the point of interest to be predicted, and the access features of the object to be predicted to the point of interest to be predicted, wherein the access features characterize the ease with which the object to be predicted accesses the point of interest to be predicted.

[0039] In this embodiment, the interaction relationship features can characterize the interaction information between the object to be predicted and the points of interest to be predicted. These features can include the interaction relationship features between any object to be predicted and multiple points of interest to be predicted. Object relationship features can characterize the social relationship information between the objects to be predicted, such as the intimacy or relevance between them. These features can include the object relationship features between any object to be predicted and multiple objects to be predicted, such as the relevance between them. Relevance can characterize the degree of intimacy or similarity between the objects to be predicted. In this embodiment, the map platform or travel platform can display multiple points of interest to be predicted. The object to be predicted can interact with these points of interest based on the map platform or travel platform, thereby obtaining the interaction relationship features between the object to be predicted and the points of interest to be predicted.

[0040] Based on the interaction characteristics between the objects to be predicted and the points of interest to be predicted, as well as the object relationship characteristics between the objects to be predicted, a target relationship graph can be generated. The target relationship graph includes nodes corresponding to each object to be predicted, and edges are formed between the nodes. For any two nodes, if there is a connecting edge between the two nodes, it means that there is a social relationship between the objects to be predicted corresponding to the two nodes, and the degree of association between the two objects to be predicted can be determined based on the object relationship characteristics. If there is no connecting edge between the two nodes, it means that there is no social relationship between the objects to be predicted corresponding to the two nodes, and the degree of association between the two objects to be predicted is 0.

[0041] Furthermore, the interaction relationship features of the object to be predicted include historical interaction features and access features. Historical interaction features can include the interaction features of any object to be predicted to multiple points of interest to be predicted within a historical time period, and access features can include the access features of any object to be predicted to multiple points of interest to be predicted. The ease with which the object to be predicted can access the points of interest to be predicted can be determined based on at least one of the following information: distance, traffic conditions, etc.

[0042] S220. Input the historical interaction features, the access features, and the object relationship features into the target feature fusion network. Aggregate the historical interaction features and access features of the neighboring nodes of each node through the target feature fusion network to update the historical interaction features and access features of each node, thereby obtaining the fusion features of the object to be predicted corresponding to each node.

[0043] The target feature fusion network has the function of aggregating the features of the neighboring nodes of each node in the target relationship graph to update the features of each node itself. Specifically, when historical interaction features, access features, and object relationship features are input into the target feature fusion network, the network can aggregate the historical interaction features of the neighboring nodes of each node and the access features of the neighboring nodes of each node. Based on the aggregation results, the historical interaction features and access features of each node are updated to obtain the fusion features of the object to be predicted for each node.

[0044] Further, in this embodiment, the input of the historical interaction features, the access features, and the object relationship features into the target feature fusion network, and the aggregation of the historical interaction features and access features of the neighboring nodes of each node through the target feature fusion network, to update the historical interaction features and access features of each node, thereby obtaining the fusion features of the object to be predicted corresponding to each node, including:

[0045] The degree of association between each node and its neighboring nodes is determined based on the object relationship characteristics.

[0046] The historical interaction features, access features, and object relationship features are input into the target feature fusion network, so that the target feature fusion network determines the aggregation weight of each neighbor node based on the correlation between each node and its neighbor nodes; performs feature aggregation based on the aggregation weight of each neighbor node and the historical interaction features of each neighbor node, and performs feature aggregation based on the aggregation weight of each neighbor node and the access features of each neighbor node to obtain the aggregation features of each node; and performs feature fusion based on the aggregation features of each node, the historical interaction features of each node, and the access features of each node to obtain the fused features of each node.

[0047] As described above, the target relationship graph includes nodes corresponding to each object to be predicted, with edges forming between nodes. For any two nodes, if there is a connecting edge between them, it indicates that there is a social relationship between the objects to be predicted corresponding to these two nodes. Furthermore, the degree of association between these two objects to be predicted can be determined based on the object relationship features. Accordingly, in the target relationship graph, the weight of the connecting edge between any two nodes is the degree of association between the objects to be predicted corresponding to these two nodes. Similarly, in the feature aggregation process of each node's neighboring nodes, the aggregation weight of each neighboring node's features in the feature aggregation process can be determined based on the weight of the connecting edge between each node and its neighboring nodes. That is, the larger the aggregation weight, the greater the weight of the corresponding neighboring node's features in the feature aggregation process, and vice versa.

[0048] Specifically, in the feature aggregation process, the historical interaction features and access features of neighboring nodes can be aggregated separately to obtain aggregated interaction features and aggregated access features. Then, based on the aggregated interaction features and aggregated access features, the aggregated features of each node are obtained. The aggregated interaction features in the aggregated features of each node are then fused with the historical interaction features of each node to obtain the first fused feature. The aggregated access features in the aggregated features of each node are then fused with the access features of each node to obtain the second fused feature. Based on the first fused feature and the second fused feature, the fused feature of each node is obtained. Here, feature fusion can specifically be feature concatenation or feature weighting, etc.

[0049] In one example, the objects to be predicted, a, b, and c, correspond to nodes a, b, and c in the target relationship graph, respectively. The node feature of node a is t1, the node feature of node b is t2, and the node feature of node c is t3. Node a has two neighboring nodes, b and c. The weight of the edge connecting node b and node a is m, and the weight of the edge connecting node c and node a is n. The aggregate weight of node b is m / (m+n), the aggregate weight of node c is n / (m+n), and the aggregate feature of node a is t2*m / (m+n)+t3*n / (m+n). Furthermore, the fusion feature of node a can be obtained as t2*m / (m+n)+t3*n / (m+n)+t1.

[0050] In this embodiment, when it is determined that each node contains multiple neighboring nodes, the aggregation weight of each neighboring node in the feature aggregation process can be determined based on the correlation between each node and its neighboring nodes. This results in the features of neighboring nodes with higher correlation having higher weights and the features of neighboring nodes with lower correlation having lower weights. This allows the features of each node to be influenced by the features of its neighboring nodes and to update its own features according to their correlation, thereby improving the accuracy of the fused features of the predicted object corresponding to each node.

[0051] S230. Input the fusion features of the object to be predicted and the interest point features of the target interest point into the target prediction network to predict access information, and obtain the access prediction information of the object to be predicted to the target interest point; the target interest point is at least one of the interest points to be predicted.

[0052] The features of a target point of interest can be determined based on its preset location, such as its latitude and longitude coordinates; or they can be determined based on its type, theme, or other characteristics. The target point of interest can be any given point of interest among those to be predicted.

[0053] By inputting the fusion features of the object to be predicted and the interest features of the target interest point into the target prediction network for access information prediction, the target prediction network can match the interest features of the target interest point with the fusion features of the object to be predicted to determine the probability that each object to be predicted will access the target interest point. In this embodiment, the access prediction information of the object to be predicted to the target interest point can include the predicted probability that each object to be predicted will access the target interest point.

[0054] S240. Based on the access prediction information, determine the target access object corresponding to the target point of interest from the objects to be predicted.

[0055] When the access prediction information includes the prediction probability, the target access object that will be accessed from each object to be predicted can be determined from each object to be predicted; for example, the object to be predicted with a prediction probability greater than a preset probability can be determined as the target access object, or a preset number of objects with a relatively large prediction probability can be determined as the target access object.

[0056] Given the real-time distance between the target access object and the target point of interest, the predicted access time of the target access object to the target point of interest can be further predicted, so that corresponding personalized services can be prepared for the target access object in advance of the predicted access time.

[0057] In this embodiment, a target relationship graph is generated based on the interaction relationship features between the object to be predicted and the points of interest to be predicted, as well as the object relationship features between the objects to be predicted, to facilitate the fusion of interaction relationship features and object relationship features. The interaction relationship features include historical interaction features and access features that characterize the ease with which the object to be predicted accesses the points of interest to be predicted. This allows the interaction relationship features of the object to be predicted to better reflect the association information between the object to be predicted and the points of interest to be predicted from multiple dimensions, improving the accuracy of the fused features of the object to be predicted. Furthermore, for each node in the target relationship graph, the target feature fusion network can aggregate the historical interaction features and access features of each node's neighboring nodes, and update the historical interaction features and access features of each node based on the aggregation results, obtaining the fused features of the object to be predicted for each node. This ensures that each object to be predicted can update its own interaction relationship features based on the interaction relationship features of its friend objects, further improving the accuracy of the fused features of the object to be predicted. The accurate fused features of the object to be predicted can deeply express the hidden association information between the object to be predicted and each point of interest to be predicted. Therefore, based on the given target interest point and the accurate fusion features of the object to be predicted, the prediction of the object that is about to visit the target interest point can be improved, which makes it easier to provide personalized services for the object that is about to visit.

[0058] Specifically, the access characteristics of the object to be predicted include the distance characteristics between the object to be predicted and the point of interest to be predicted; please refer to the relevant section. Figure 3 It illustrates a distance feature determination method, which may include:

[0059] S310. For any object to be predicted, obtain the real-time location of the object to be predicted.

[0060] The real-time location of any object to be predicted can be the real-time latitude and longitude coordinates of any object to be predicted, or the real-time location of any object to be predicted, etc. This embodiment does not make specific limitations.

[0061] S320. Based on the real-time location of any object to be predicted and the preset location of each point of interest to be predicted, determine the real-time distance between any object to be predicted and any point of interest to be predicted.

[0062] The preset location of each point of interest to be predicted can be its latitude and longitude coordinates. When the real-time location of any object to be predicted is its real-time latitude and longitude coordinates, the real-time distance between any object to be predicted and any point of interest to be predicted can be determined based on the latitude and longitude coordinates of the point of interest to be predicted and the real-time latitude and longitude coordinates of any object to be predicted. When the real-time location of any object to be predicted is its real-time region, the real-time distance can be determined based on the latitude and longitude coordinates of the point of interest to be predicted and the latitude and longitude coordinates of a target point in the real-time region. The target point can be the regional center point of the real-time region or any point in the real-time region.

[0063] S330. Normalize the real-time distance between any object to be predicted and any point of interest to be predicted to obtain the distance features between any object to be predicted and any point of interest to be predicted.

[0064] Specifically, the distance feature can be obtained using the following formula:

[0065]

[0066] Where lc represents the real-time position of object u, l represents any point of interest to be predicted, and D(lc, l) represents the distance between the real-time position of object u and the preset position of any point of interest to be predicted. max This represents the farthest distance between object u and each point of interest.

[0067] In this embodiment, the distance feature between any object to be predicted and any point of interest to be predicted can be determined based on the real-time distance between the object to be predicted and the point of interest to be predicted. The distance feature characterizes the ease of access; for example, a shorter distance indicates greater ease of access, while a larger distance indicates lower ease. By applying the distance feature to subsequent object prediction, the probability of an object accessing a point of interest can be predicted, making the object features used in the prediction process more diverse and the prediction results more accurate. Furthermore, normalizing the real-time distance between any object to be predicted and any point of interest to be predicted makes the data more concentrated, preventing excessively large values ​​from affecting model predictions and avoiding overflow or underflow, thus improving the stability of the model's data processing.

[0068] Furthermore, in this embodiment, the access characteristics of the object to be predicted also include traffic characteristics between the real-time location of the object to be predicted and the preset location of the point of interest to be predicted;

[0069] The method further includes:

[0070] Obtain at least one candidate route from the real-time location of any object to be predicted to a preset location of any point of interest to be predicted, as well as the real-time traffic conditions of each candidate route;

[0071] Based on the at least one candidate route and the real-time traffic conditions of each candidate route, the traffic characteristics of any object to be predicted and any point of interest to be predicted are determined.

[0072] The traffic features in this embodiment may include real-time traffic conditions determined based on each candidate route. Real-time traffic conditions may include the congestion status of the roads corresponding to the candidate routes and the travel modes corresponding to the candidate routes. The travel modes supported by a candidate route are those supported by that route; for example, some candidate routes only support walking, cycling, and electric vehicles, while others support walking, cycling, electric vehicles, private cars, buses, subways, etc. Correspondingly, the lower the congestion level of a candidate route and the more travel modes it supports, the better the traffic features of that candidate route characterize the ease with which the target object can access the predicted point of interest. Conversely, the higher the congestion level of a candidate route and the fewer travel modes it supports, the worse the traffic features of that candidate route characterize the ease with which the target object can access the predicted point of interest.

[0073] In this embodiment, the corresponding traffic features can be determined by combining the candidate routes between the object to be predicted and the point of interest to be predicted, as well as the real-time traffic information of the candidate routes. These traffic features can characterize the ease with which the object to be predicted can access the point of interest to be predicted. By applying the traffic features to the subsequent object prediction process, the probability of the object to be predicted accessing the point of interest to be predicted can be predicted by combining the traffic features between the object to be predicted and the point of interest to be predicted. This makes the object features used in the prediction process more diversified, and thus makes the prediction results more accurate.

[0074] In this embodiment, the historical interaction features of the object to be predicted can be generated based on the historical interaction events between the object to be predicted and the point of interest to be predicted; please refer to [link / reference] for details. Figure 4 It illustrates a method for generating historical interaction features, which may include:

[0075] S410. Obtain historical interaction events between the object to be predicted and the point of interest to be predicted; the historical interaction events include interactive operation events of the object to be predicted on the target point of interest platform, historical access events of the object to be predicted to the point of interest, and regional entry events of the object to be predicted entering the associated area of ​​the preset location of the point of interest.

[0076] A target interest point platform can be a platform provided to the target object, enabling interaction with various target interest points. This platform can be an application or a webpage. Interaction events on the target interest point platform by the target object can include check-in events, comment events, like events, favorite events, share events, and view events. For check-in events, information such as the check-in object's identifier, check-in time, checked-in interest point, and check-in location can be included. Comment events refer to the target object's comments on various interest points. Furthermore, the sentiment tendency of the comments can be determined based on their content. Sentiment tendency can be positive, negative, or neutral. Positive sentiment indicates a recommendation for the interest point and a high probability of revisiting it; negative sentiment indicates a disrespect for the interest point and a low probability of revisiting it; neutral sentiment indicates a possible revisit or non-revisit. A sharing event can be an event where a first predictor shares a point of interest with a second predictor. The first predictor can be the sharing object among multiple predictors, and the second predictor can be the object being shared with among multiple predictors. Accordingly, it can be determined that the first and second predictors have a relatively high probability of visiting the shared point of interest. That is, a sharing event can not only represent a high degree of interest in the shared point of interest by both the sharing object and the object being shared with, but also a high degree of social connection between the sharing object and the object being shared with. A viewing event can be an action by a predictor to view a point of interest, indicating that the predictor is interested in that point of interest and is likely to visit it.

[0077] Historical access events can be events where the object to be predicted visits a point of interest within a historical event period. Specifically, these events may include the object's identifier, access time, access location, and the point of interest's identifier. Region entry events can be events where the object to be predicted enters a region associated with a point of interest. The associated region can be a region adjacent to the point of interest or a region within a preset area centered on the point of interest. Region entry events may specifically include the object's identifier, entry time, the region's identifier, and the entry location.

[0078] S420. Based on the interactive operation event, the historical access event, and the area entry event, feature extraction processing is performed respectively to obtain interactive operation features, first interaction features, and second interaction features.

[0079] Feature extraction based on interactive operation events yields interactive operation features; feature extraction based on historical access events yields first interactive features; and feature extraction based on region entry events yields second interactive features.

[0080] For example, for a sign-in event, the formula for calculating sign-in characteristics is as follows:

[0081]

[0082] Where A(u) represents the number of times object u checks in at point of interest l, and A represents the total number of times all objects check in at point of interest l.

[0083] Correspondingly, for events such as commenting, liking, collecting, sharing, and viewing, corresponding comment features, like features, collect features, share features, and view features can be generated. Furthermore, interactive operation features can be generated based on check-in features, comment features, like features, collect features, share features, and view features.

[0084] For historical access events, the formula for calculating the first interaction feature is as follows:

[0085]

[0086] Where A(u,l) represents the set visited by object u at point of interest l, and A(u) represents the number of times object u checks in at point of interest l.

[0087] For region entry events, the formula for calculating the second interaction feature is as follows:

[0088]

[0089] Where P(l)) represents the associated region of point of interest l, A(u,P(l)) represents the set of visits of object u in the associated region, and A(u) represents the number of times object u checks in at point of interest l.

[0090] S430. Based on the interactive operation features, the first interactive features, and the second interactive features, the historical interactive features of the object to be predicted are obtained.

[0091] Based on the above feature extraction method, interactive operation features, first interaction features, and second interaction features can be obtained. By performing feature weighting or feature concatenation on the interactive operation features, first interaction features, and second interaction features, the historical interaction features of the object to be predicted can be obtained.

[0092] Specifically, in this embodiment, an object feature generator can also be used to obtain the historical interaction features of each object to be predicted. Specifically, by inputting the historical interaction event dataset into the object feature generator, the corresponding interactive operation features, the first interaction feature, and the second interaction feature can be generated.

[0093] In this embodiment, by acquiring multiple dimensions of information such as interactive operation events, historical access events, and area entry events of the object to be predicted, the corresponding interactive operation features, first interaction features, and second interaction features are determined respectively. Then, the historical interaction features of the object to be predicted are generated based on multiple dimensions of information such as interactive operation, historical access, and associated area entry, thereby improving the accuracy and comprehensiveness of the historical interaction feature generation.

[0094] Furthermore, the information prediction method in this embodiment may also include:

[0095] If the preset object relationship features do not include the social relationship information of the objects to be predicted on the target point of interest platform, the preset object relationship features are updated based on the historical interaction operations of each object to be predicted on the target point of interest platform to obtain the object relationship features between the objects to be predicted.

[0096] The target point of interest platform can be a map platform or a travel platform, etc. The preset relationship features may be social relationship information between the objects to be predicted obtained from other social platforms or networks, which may not include the social relationships on the target point of interest platform. As can be seen from the above, the objects to be predicted can socialize based on the target point of interest platform. For example, a sharing event can be an event in which the first object to be predicted shares an interest point with the second object to be predicted based on the target point of interest platform. The first object to be predicted can be the sharing object among multiple objects to be predicted, and the second object to be predicted can be the object being shared with among multiple objects to be predicted. Accordingly, it can be determined that the first object to be predicted and the second object to be predicted have a high probability of visiting the shared interest point. That is, the sharing event can not only represent that the sharing object and the object being shared with have a high degree of interest in the shared interest point, but also represent that the sharing object and the object being shared with have a high degree of social correlation.

[0097] In addition to sharing between objects, social relationships on target interest platforms can also include interactive operations such as viewing object profiles and following objects. Accordingly, the preset object relationship features can be improved based on the social relationships between objects to be predicted on the target interest platform. On the one hand, this can improve the comprehensiveness and accuracy of the expression of social relationships between objects. On the other hand, social relationships on the interest platform can better reflect the interaction between objects based on interest points, thus making the prediction and recommendation more in line with interest point scenarios.

[0098] In object-interest prediction scenarios, the interaction information of objects with interest points may change, the social relationships between objects may also change, and new objects may be added to the scenario. Therefore, it is necessary to dynamically adjust the target feature fusion network and the target prediction network to adapt to the dynamically changing scenario; please refer to [link / reference] for details. Figure 5 It illustrates a method for predicting network updates, which may include:

[0099] S510. Obtain the incremental data corresponding to the current time window; the incremental data includes incremental interaction relationship features, incremental object relationship features, and incremental access tags.

[0100] Incremental interaction relationship features can include new interaction relationship features between existing objects and points of interest, and / or interaction relationship features between newly added objects and points of interest. Incremental object relationship features can include an increase or decrease in social relationships between any two existing objects; or an increase in social relationships between existing and newly added objects, or an increase in social relationships between newly added objects, etc. Incremental access tags can be updated tags of existing objects to points of interest, or access tags of newly added objects to points of interest.

[0101] S520. Based on the incremental interaction relationship features and the incremental object relationship features, determine the new node in the target relationship graph, the interaction relationship features of the new node, and the neighbor nodes of the new node.

[0102] When the incremental data includes incremental interaction relationship features, incremental object relationship features, and incremental access tags of the new object, a new node corresponding to the new object can be generated and added to the target relationship graph, while the neighbor nodes corresponding to the new node are determined.

[0103] S530. Based on the newly added node, the interaction relationship features of the newly added node, the neighboring nodes of the newly added node, and the incremental access label, the target feature fusion network and the target prediction network are updated to obtain the updated target feature fusion network and the updated target prediction network.

[0104] Both the target feature fusion network and the target prediction network are updated based on the newly added nodes, the interaction features of the newly added nodes, the neighboring nodes of the newly added nodes, and the frontal access labels. That is, only the network structure parameters need to be updated based on incremental data, without retraining the entire network, thereby improving the network training efficiency and keeping the network in a dynamic update state, which is more in line with real-time data. In turn, it can improve the efficiency and accuracy of information prediction based on the updated target prediction network and the updated target prediction network.

[0105] The following is a specific example illustrating the implementation process of the information prediction method in this application. In this embodiment, the object to be predicted can specifically be a user interacting with a point of interest. Please refer to [link / reference]. Figure 6 The diagram illustrates a network based on points of interest (POIs) interaction. This network includes two different types of entities: users and POIs. Users include u1, u2, u3, and u4, while POIs include l1, l2, l3, l4, and l5. Each user can check in at multiple POIs within a given time period, and each POI can be checked in by multiple users. For example, user u1 checked in at POIs l1 and l2, user u2 checked in at POIs l1, l2, and l4, user u3 checked in at POIs l4, l3, and l5, and user u4 checked in at POIs l3 and l5.

[0106] Please see Figure 7 It illustrates an information prediction process, specifically acquiring user interaction relationship features and object relationship features. The interaction relationship features can be generated based on an object feature generator. These features may include interaction operation features determined by interactive operation events, first interaction features determined by historical access events, second interaction features determined by area entry events, and access features representing access convenience. The interaction relationship features may include interaction relationship features between each user and multiple points of interest. The interaction relationship features corresponding to users 1 to n are U1, U2, U3…U… n The object relationship features can specifically be a user social relationship matrix A representing the social relationships between users 1 to n, thereby representing the interaction relationship features U1, U2, U3...U n The user social relationship matrix A is input into the target feature fusion network to obtain the fusion features corresponding to users 1 to n. Then, the interest point information Pi of the target interest points and the fusion features corresponding to users 1 to n are input into the target prediction network to obtain the predicted probabilities O1, O2, O3...O of users 1 to n visiting the target interest points. nThen, users whose predicted probability is greater than the preset probability are identified as users who are about to visit the target point of interest, or users with a relatively high predicted probability are identified as users who are about to visit the target point of interest.

[0107] In this embodiment, during the specific network model training process, the parameters of the object feature generator, the target feature fusion network, and the target prediction network can be adjusted based on the objective function. Specifically, for the target feature fusion network, parameters such as the weights and biases of the convolutional layers can be modified; for the target prediction network, parameters such as the weights and biases of the fully connected layers can be modified, thereby obtaining the object feature generator, the target feature fusion network, and the target prediction network. The objective function can specifically adopt the following cross-entropy loss function:

[0108]

[0109] Where x represents a sample, y represents the actual label, a represents the prediction result, and n represents the total number of samples.

[0110] It should be noted that any of the methods described above in this embodiment can be combined based on the actual implementation situation and have corresponding beneficial effects, which will not be elaborated here.

[0111] Please see Figure 8 This embodiment also provides an information prediction device, which may include:

[0112] The relationship graph generation module 810 is used to generate a target relationship graph based on the interaction relationship features between the object to be predicted and the point of interest to be predicted, as well as the object relationship features between the objects to be predicted; each node in the target relationship graph corresponds to an object to be predicted; the interaction relationship features of the object to be predicted include the historical interaction features between the object to be predicted and the point of interest to be predicted, as well as the access features of the object to be predicted to the point of interest to be predicted, wherein the access features characterize the ease with which the object to be predicted accesses the point of interest to be predicted;

[0113] The feature fusion module 820 is used to input the historical interaction features, the access features, and the object relationship features into the target feature fusion network, and to aggregate the historical interaction features and access features of the neighboring nodes of each node through the target feature fusion network, so as to update the historical interaction features and access features of each node, and obtain the fusion features of the object to be predicted corresponding to each node.

[0114] The information prediction module 830 is used to input the fusion features of the object to be predicted and the interest point features of the target interest point into the target prediction network to predict access information, thereby obtaining the access prediction information of the object to be predicted to the target interest point; the target interest point is at least one of the interest points to be predicted.

[0115] The object determination module 840 is used to determine the target access object corresponding to the target point of interest from the objects to be predicted based on the access prediction information.

[0116] Furthermore, the access characteristics of the object to be predicted include the distance characteristics between the object to be predicted and the point of interest to be predicted;

[0117] The device further includes a distance feature determination module, used for:

[0118] For any object to be predicted, obtain the real-time location of the object to be predicted.

[0119] Based on the real-time location of any object to be predicted and the preset location of each point of interest to be predicted, the real-time distance between any object to be predicted and any point of interest to be predicted is determined.

[0120] The real-time distance between any object to be predicted and any point of interest to be predicted is normalized to obtain the distance features between any object to be predicted and any point of interest to be predicted.

[0121] Furthermore, the access characteristics of the object to be predicted also include traffic characteristics between the real-time location of the object to be predicted and the preset location of the point of interest to be predicted;

[0122] The device further includes a traffic feature determination module, used for:

[0123] Obtain at least one candidate route from the real-time location of any object to be predicted to a preset location of any point of interest to be predicted, as well as the real-time traffic conditions of each candidate route;

[0124] Based on the at least one candidate route and the real-time traffic conditions of each candidate route, the traffic characteristics of any object to be predicted and any point of interest to be predicted are determined.

[0125] Furthermore, the device also includes a historical interaction feature determination module, used for:

[0126] The historical interaction events between the object to be predicted and the point of interest to be predicted are obtained. The historical interaction events include the interaction operation events of the object to be predicted on the target point of interest platform, the historical access events of the object to be predicted to the point of interest, and the regional entry events of the object to be predicted when it enters the associated area of ​​the preset location of the point of interest.

[0127] Based on the interactive operation event, the historical access event, and the area entry event, feature extraction processing is performed to obtain interactive operation features, first interaction features, and second interaction features.

[0128] The historical interaction features of the object to be predicted are obtained based on the interactive operation features, the first interaction features, and the second interaction features.

[0129] Furthermore, the device also includes an object relationship feature update module, used for:

[0130] If the preset object relationship features do not include the social relationship information of the objects to be predicted on the target point of interest platform, the preset object relationship features are updated based on the historical interaction operations of each object to be predicted on the target point of interest platform to obtain the object relationship features between the objects to be predicted.

[0131] Furthermore, the feature fusion module is used for:

[0132] The degree of association between each node and its neighboring nodes is determined based on the object relationship characteristics.

[0133] The historical interaction features, access features, and object relationship features are input into the target feature fusion network, so that the target feature fusion network determines the aggregation weight of each neighbor node based on the correlation between each node and its neighbor nodes; performs feature aggregation based on the aggregation weight of each neighbor node and the historical interaction features of each neighbor node, and performs feature aggregation based on the aggregation weight of each neighbor node and the access features of each neighbor node to obtain the aggregation features of each node; and performs feature fusion based on the aggregation features of each node, the historical interaction features of each node, and the access features of each node to obtain the fused features of each node.

[0134] Furthermore, the device also includes a network update module for:

[0135] Obtain incremental data corresponding to the current time window; the incremental data includes incremental interaction relationship features, incremental object relationship features, and incremental access tags.

[0136] Based on the incremental interaction relationship features and the incremental object relationship features, the newly added nodes, the interaction relationship features of the newly added nodes, and the neighbor nodes of the newly added nodes in the target relationship graph are determined.

[0137] Based on the newly added node, the interaction relationship features of the newly added node, the neighboring nodes of the newly added node, and the incremental access label, the target feature fusion network and the target prediction network are updated to obtain the updated target feature fusion network and the updated target prediction network.

[0138] The apparatus provided in the above embodiments can execute the methods provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in the above embodiments can be found in the methods provided in any embodiment of this application.

[0139] This embodiment also provides a computer-readable storage medium storing at least one instruction or at least one program, which is loaded by a processor and executed as any of the methods described above in this embodiment.

[0140] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform any of the methods described above.

[0141] Figure 9 This is a block diagram illustrating an electronic device for information prediction according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an information prediction method.

[0142] Those skilled in the art will understand that Figure 9The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0143] This specification provides the operational steps of the methods described in the embodiments or flowcharts, but more or fewer operational steps may be included based on conventional or non-inventive labor. The steps and order listed in the embodiments are merely one possible execution order among many steps and do not represent the only execution order. In actual system or interrupt product execution, the methods shown in the embodiments or drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0144] The structure shown in this embodiment is only a partial structure related to the solution of this application and does not constitute a limitation on the device to which the solution of this application is applied. Specific devices may include more or fewer components than shown, or combinations of certain components, or arrangements of different components. It should be understood that the methods, apparatuses, etc., disclosed in this embodiment can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or unit modules through some interfaces.

[0145] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0146] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this specification can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0147] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An information prediction method, characterized in that, include: Based on the interaction relationship features between the objects to be predicted and the points of interest to be predicted, as well as the object relationship features between the objects to be predicted, a target relationship graph is generated. Each node in the target relationship graph corresponds to a target object to be predicted; the interaction relationship features of the target object to be predicted include the historical interaction features between the target object to be predicted and the target interest point to be predicted, as well as the access features of the target object to the target interest point to be predicted, wherein the access features characterize the ease with which the target object to access the target interest point to be predicted. The historical interaction features, the access features, and the object relationship features are input into the target feature fusion network. The target feature fusion network aggregates the historical interaction features and access features of the neighboring nodes of each node to update the historical interaction features and access features of each node, thereby obtaining the fusion features of the object to be predicted corresponding to each node. The fusion features of the object to be predicted and the interest point features of the target interest point are input into the target prediction network to predict access information, thereby obtaining the access prediction information of the object to be predicted to the target interest point; the target interest point is at least one of the interest points to be predicted. Based on the access prediction information, the target access object corresponding to the target point of interest is determined from the objects to be predicted.

2. The method according to claim 1, characterized in that, The access characteristics of the object to be predicted include the distance characteristics between the object to be predicted and the point of interest to be predicted; The method further includes: For any object to be predicted, obtain the real-time location of the object to be predicted. Based on the real-time location of any object to be predicted and the preset location of each point of interest to be predicted, the real-time distance between any object to be predicted and any point of interest to be predicted is determined. The real-time distance between any object to be predicted and any point of interest to be predicted is normalized to obtain the distance features between any object to be predicted and any point of interest to be predicted.

3. The method according to claim 2, characterized in that, The access characteristics of the object to be predicted also include the traffic characteristics between the real-time location of the object to be predicted and the preset location of the point of interest to be predicted; The method further includes: Obtain at least one candidate route from the real-time location of any object to be predicted to a preset location of any point of interest to be predicted, as well as the real-time traffic conditions of each candidate route; Based on the at least one candidate route and the real-time traffic conditions of each candidate route, the traffic characteristics of any object to be predicted and any point of interest to be predicted are determined.

4. The method according to claim 1, characterized in that, Before generating the target relationship graph based on the interaction relationship features between the object to be predicted and the interest point to be predicted, and the object relationship features between the objects to be predicted, the method further includes: The historical interaction events between the object to be predicted and the point of interest to be predicted are obtained. The historical interaction events include the interaction operation events of the object to be predicted on the target point of interest platform, the historical access events of the object to be predicted to the point of interest, and the regional entry events of the object to be predicted when it enters the associated area of ​​the preset location of the point of interest. Based on the interactive operation event, the historical access event, and the area entry event, feature extraction processing is performed to obtain interactive operation features, first interaction features, and second interaction features. The historical interaction features of the object to be predicted are obtained based on the interactive operation features, the first interaction features, and the second interaction features.

5. The method according to claim 4, characterized in that, The method further includes: If the preset object relationship features do not include the social relationship information of the objects to be predicted on the target point of interest platform, the preset object relationship features are updated based on the historical interaction operations of each object to be predicted on the target point of interest platform to obtain the object relationship features between the objects to be predicted.

6. The method according to claim 1, characterized in that, The step involves inputting the historical interaction features, access features, and object relationship features into a target feature fusion network. This network then aggregates the historical interaction features and access features of each node's neighboring nodes to update the historical interaction features and access features of each node, resulting in the fused features of the object to be predicted for each node. This process includes: The degree of association between each node and its neighboring nodes is determined based on the object relationship characteristics. The historical interaction features, access features, and object relationship features are input into the target feature fusion network, so that the target feature fusion network determines the aggregation weight of each neighbor node based on the correlation between each node and its neighbor nodes; performs feature aggregation based on the aggregation weight of each neighbor node and the historical interaction features of each neighbor node, and performs feature aggregation based on the aggregation weight of each neighbor node and the access features of each neighbor node to obtain the aggregation features of each node; and performs feature fusion based on the aggregation features of each node, the historical interaction features of each node, and the access features of each node to obtain the fused features of each node.

7. The method according to claim 1, characterized in that, The method further includes: Obtain incremental data corresponding to the current time window; the incremental data includes incremental interaction relationship features, incremental object relationship features, and incremental access tags. Based on the incremental interaction relationship features and the incremental object relationship features, the newly added nodes, the interaction relationship features of the newly added nodes, and the neighbor nodes of the newly added nodes in the target relationship graph are determined. Based on the newly added node, the interaction relationship features of the newly added node, the neighboring nodes of the newly added node, and the incremental access label, the target feature fusion network and the target prediction network are updated to obtain the updated target feature fusion network and the updated target prediction network.

8. An information prediction device, characterized in that, include: The relationship graph generation module is used to generate a target relationship graph based on the interaction relationship features between the object to be predicted and the interest point to be predicted, as well as the object relationship features between the objects to be predicted. Each node in the target relationship graph corresponds to a target object to be predicted; the interaction relationship features of the target object to be predicted include the historical interaction features between the target object to be predicted and the target interest point to be predicted, as well as the access features of the target object to the target interest point to be predicted, wherein the access features characterize the ease with which the target object to access the target interest point to be predicted. The feature fusion module is used to input the historical interaction features, the access features, and the object relationship features into the target feature fusion network, and to aggregate the historical interaction features and access features of the neighboring nodes of each node through the target feature fusion network, so as to update the historical interaction features and access features of each node, and obtain the fusion features of the object to be predicted corresponding to each node. The information prediction module is used to input the fusion features of the object to be predicted and the interest point features of the target interest point into the target prediction network to predict access information, thereby obtaining the access prediction information of the object to be predicted to the target interest point; the target interest point is at least one of the interest points to be predicted. The object determination module is used to determine the target access object corresponding to the target point of interest from the objects to be predicted based on the access prediction information.

9. An electronic device, characterized in that, The device includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the information prediction method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The storage medium stores at least one instruction or at least one program, which is loaded and executed by a processor according to any one of claims 1 to 7.

Images