Business district recommendation method, device and equipment based on spatio-temporal trajectory characteristics, medium and product

Abstract

The invention discloses a business district recommendation method and device based on space-time trajectory characteristics, equipment, a medium and a product, and relates to the technical field of business district recommendation, and the method comprises the steps: generating a user dynamic trajectory representation and an area dynamic trajectory representation according to mobile phone signaling data based on a space-time large model; constructing user multi-source features according to the user dynamic trajectory characterization, and constructing regional multi-source features according to the regional dynamic trajectory characterization; inputting the user multi-source features into a user tower of a double-tower recommendation model to obtain a user representation vector, and inputting the regional multi-source features into a regional tower of the double-tower recommendation model to obtain a regional representation vector; and determining a business district recommendation result according to the user representation vector and the region representation vector. The technical problem that a recommendation result of a traditional business district recommendation scheme is not accurate enough can be solved, so that an appropriate business district is recommended to the user.

Description

Technical Field

[0001] The present application relates to the technical field of business district recommendation, and in particular to a business district recommendation method, device, equipment, medium and product based on spatiotemporal trajectory features. Background Art

[0002] Traditional business district recommendation solutions integrate heterogeneous multi-source data of cities to obtain the urban spatiotemporal knowledge graph model, thereby constructing an urban knowledge graph to learn the dynamic characteristics of the six domains of interest points, categories, brands, business districts, regions, and users over time, as well as the dynamic characteristics of the relationships over time, and thus obtain the long-term and short-term preferences of users for interest points. For the dynamic characteristics of the learned knowledge graph entities and relationships, combined with the long-term and short-term preferences of users, recommendation results are generated. However, this business district recommendation solution only learns the dynamic characteristics of the six domains of interest points, categories, brands, business districts, regions, and users over time, as well as the dynamic characteristics of their relationships over time. It lacks the linkage and integration of online and offline information between users and regions, and does not adequately characterize the characteristics of users and regions, which will lead to inaccurate final business district recommendation results. Summary of the invention

[0003] The main purpose of this application is to provide a method, device, equipment, medium and product for recommending a business district based on spatiotemporal trajectory characteristics, aiming to solve the technical problem that the traditional business district recommendation scheme lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation results being not accurate enough.

[0004] To achieve the above purpose, the present application proposes a method for recommending a business district based on spatiotemporal trajectory features, and the method for recommending a business district based on spatiotemporal trajectory features includes:

[0005] Based on the spatiotemporal big model, the user dynamic trajectory representation and regional dynamic trajectory representation are generated according to the mobile phone signaling data;

[0006] constructing a user multi-source feature according to the user dynamic trajectory representation, and constructing a regional multi-source feature according to the regional dynamic trajectory representation;

[0007] Inputting the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and inputting the region multi-source features into the region tower of the dual-tower recommendation model to obtain a region representation vector;

[0008] A business district recommendation result is determined according to the user characterization vector and the region characterization vector.

[0009] In one embodiment, the step of generating a user dynamic trajectory representation and a regional dynamic trajectory representation based on the spatiotemporal large model and the mobile phone signaling data comprises:

[0010] Get mobile phone signaling data;

[0011] Eliminate abnormal data in the mobile phone signaling data to obtain target mobile phone signaling data;

[0012] Based on the grid division result of the spatial region, generating a user spatiotemporal triple sequence according to the target mobile phone signaling data;

[0013] The user spatiotemporal triple sequence is input into a spatiotemporal large model to obtain a user dynamic trajectory representation and a region dynamic trajectory representation, wherein the spatiotemporal large model is constructed using a generative pre-trained transformation model.

[0014] In one embodiment, the steps of constructing a user multi-source feature according to the user dynamic trajectory representation and constructing a region multi-source feature according to the region dynamic trajectory representation include:

[0015] Using the user dynamic trajectory representation as user trajectory information, and determining the pedestrian flow characteristics of the business district according to the regional dynamic trajectory representation;

[0016] The user multi-source features are constructed based on the user trajectory information, user basic attributes, user call status, user Internet information, user travel information and the user's historical visited business district vector list, and the regional multi-source features are constructed based on the business district crowd flow characteristics, business district basic attributes, business district energy level characteristics and business district business format information.

[0017] In one embodiment, the steps of constructing user multi-source features according to the user trajectory information, user basic attributes, user call status, user Internet access information, user travel information, and a user's historically visited business district vector list, and constructing regional multi-source features according to the business district crowd flow characteristics, business district basic attributes, business district energy level characteristics, and business district business format information include:

[0018] Classifying the user multi-source feature and the region multi-source feature respectively to obtain a first discrete feature, a first continuous feature, a first vector feature, a second discrete feature, a second continuous feature, and a second vector feature;

[0019] Performing one-hot encoding on the first discrete feature and the second discrete feature respectively to obtain a first user dense vector and a first region dense vector;

[0020] Binning the first continuous feature and the second continuous feature respectively to obtain a second user dense vector and a second region dense vector;

[0021] Generate a new user multi-source feature based on the first user dense vector, the second user dense vector, the user's historical visited business district vector list, and the first vector feature;

[0022] A new regional multi-source feature is generated based on the first regional dense vector, the second regional dense vector, and the second vector feature.

[0023] In one embodiment, the user tower includes a user feature embedding layer, a multi-head self-attention mechanism layer, a user feature cross layer, a user deep neural network layer, and a user normalization layer; wherein the step of inputting the user multi-source features into the user tower of the dual-tower model to obtain the user representation vector includes:

[0024] Input the multi-source features of the user into the user tower, and process the user's historical visited business district vector list through the multi-head self-attention mechanism layer to obtain a user history vector;

[0025] Processing the user multi-source features and the user history vector through the user feature embedding layer to obtain a user splicing vector feature;

[0026] Processing the user concatenated vector features through the user feature cross layer to obtain a high-order representation of the user after crossover of different features;

[0027] Processing the user concatenated vector features and the user high-order representation through the user deep neural network layer to obtain an initial user vector;

[0028] The initial user vector is processed by the user normalization layer to obtain a user representation vector.

[0029] In one embodiment, the regional tower includes a regional feature embedding layer, a regional feature cross layer, a regional deep neural network layer, and a regional normalization layer; the step of inputting the regional multi-source features into the regional tower of the dual-tower model to obtain the user representation vector includes:

[0030] Inputting the regional multi-source features into the regional tower, processing the regional multi-source features through the regional feature embedding layer to obtain regional splicing vector features;

[0031] Processing the regional splicing vector features through the regional feature cross layer to obtain a high-order representation of the region after crossover of different features;

[0032] Processing the regional splicing vector features and the regional high-order representation through the regional deep neural network layer to obtain an initial regional vector;

[0033] The initial region vector is processed by the region normalization layer to obtain a region representation vector.

[0034] In one embodiment, before the step of inputting the user multi-source features into the user tower of the dual-tower recommendation model to obtain the user representation vector, and the step of inputting the region multi-source features into the region tower of the dual-tower recommendation model to obtain the region representation vector, the step further includes:

[0035] Input the training samples into the initial dual-tower recommendation model to obtain the predicted probability;

[0036] Based on the true value label of the training sample and the predicted probability, determining a cross entropy loss value according to a cross entropy loss function;

[0037] The initial dual-tower recommendation model is updated according to the cross entropy loss value to obtain a dual-tower recommendation model.

[0038] In one embodiment, the step of determining a business district recommendation result according to the user representation vector and the region representation vector includes:

[0039] Determine a dot product calculation result of the user characterization vector and the region characterization vector;

[0040] The region representation vector with the maximum dot product calculation result is used as the target region representation vector;

[0041] A target business district corresponding to the target area representation vector is determined, and the target business district is used as a business district recommendation result.

[0042] In addition, to achieve the above purpose, the present application also proposes a device for recommending a business district based on spatiotemporal trajectory features, and the device for recommending a business district based on spatiotemporal trajectory features includes:

[0043] A generation module, used to generate user dynamic trajectory representation and regional dynamic trajectory representation based on the spatiotemporal large model and mobile phone signaling data;

[0044] A construction module, configured to construct a user multi-source feature according to the user dynamic trajectory representation, and to construct a region multi-source feature according to the region dynamic trajectory representation;

[0045] An input module, configured to input the user multi-source features into a user tower of a dual-tower recommendation model to obtain a user representation vector, and to input the region multi-source features into a region tower of a dual-tower recommendation model to obtain a region representation vector;

[0046] The determination module is used to determine the business district recommendation result according to the user characterization vector and the area characterization vector.

[0047] In addition, to achieve the above-mentioned purpose, the present application also proposes a business district recommendation device based on spatiotemporal trajectory features, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the business district recommendation method based on spatiotemporal trajectory features as described above.

[0048] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the business district recommendation method based on spatiotemporal trajectory features as described above are implemented.

[0049] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the business district recommendation method based on spatiotemporal trajectory features as described above are implemented.

[0050] One or more technical solutions proposed in this application have at least the following technical effects:

[0051] The method, device, equipment, medium and product for recommending a business district based on spatiotemporal trajectory features proposed in the present application generate user dynamic trajectory representation and regional dynamic trajectory representation based on mobile phone signaling data based on a spatiotemporal large model; construct user multi-source features based on the user dynamic trajectory representation, and construct regional multi-source features based on the regional dynamic trajectory representation; input the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and input the regional multi-source features into the regional tower of the dual-tower recommendation model to obtain a regional representation vector; determine the business district recommendation result based on the user representation vector and the regional representation vector, and solve the technical problem that the traditional business district recommendation scheme lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation result being not accurate enough. Compared with the prior art, the present application adopts a spatiotemporal large model to extract hidden information between users and regions, and learns the user's regular daily behavior and preference behavior from massive signaling trajectory data to form user dynamic trajectory representation and regional dynamic trajectory representation, and then adopts a separable dual-tower deep model to learn the correlation information between users and business districts, thereby providing users with more accurate business district recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0054] Figure 1 A flowchart diagram of the first embodiment of the method for recommending a business district based on spatiotemporal trajectory features of the present application;

[0055] Figure 2 A schematic diagram of the structure of the dual-tower recommendation model provided in Example 1 of the method for recommending a business district based on spatiotemporal trajectory features of the present application;

[0056] Figure 3 A schematic diagram of the overall process framework provided for the first embodiment of the method for recommending a business district based on spatiotemporal trajectory features of the present application;

[0057] Figure 4 A flowchart diagram of Embodiment 2 of the method for recommending a business district based on spatiotemporal trajectory features of the present application;

[0058] Figure 5 This is a schematic diagram of the module structure of a device for recommending a business district based on spatiotemporal trajectory features according to an embodiment of the present application;

[0059] Figure 6 This is a schematic diagram of the device structure of the hardware operating environment involved in the business district recommendation method based on spatiotemporal trajectory features in an embodiment of the present application.

[0060] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0061] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0062] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0063] The main solution of the embodiment of the present application is: based on the spatiotemporal big model, generate user dynamic trajectory representation and regional dynamic trajectory representation according to mobile phone signaling data; construct user multi-source features according to the user dynamic trajectory representation, and construct regional multi-source features according to the regional dynamic trajectory representation; input the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and input the regional multi-source features into the regional tower of the dual-tower recommendation model to obtain a regional representation vector; determine the business district recommendation result according to the user representation vector and the regional representation vector.

[0064] It can be seen from the above embodiments that the present application generates user dynamic trajectory representation and regional dynamic trajectory representation based on mobile phone signaling data based on a large spatiotemporal model; constructs user multi-source features based on the user dynamic trajectory representation, and constructs regional multi-source features based on the regional dynamic trajectory representation; inputs the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and inputs the regional multi-source features into the regional tower of the dual-tower recommendation model to obtain a regional representation vector; determines the business district recommendation result based on the user representation vector and the regional representation vector, thereby solving the technical problem that the traditional business district recommendation scheme lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation result being not accurate enough. Compared with the prior art, the present application uses a large spatiotemporal model to extract hidden information between users and regions, and learns users' regular daily behaviors and preference behaviors from massive signaling trajectory data to form user dynamic trajectory representation and regional dynamic trajectory representation, and then uses a separable dual-tower deep model to learn the correlation information between users and business districts, thereby providing users with more accurate business district recommendations.

[0065] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a business district recommendation device based on spatiotemporal trajectory features, etc. The following takes the business district recommendation based on spatiotemporal trajectory features as an example to illustrate this embodiment and the following embodiments.

[0066] Based on this, the present application embodiment provides a method for recommending a business district based on spatiotemporal trajectory features. Figure 1 , Figure 1 This is a flowchart of the first embodiment of the business district recommendation method based on spatiotemporal trajectory features of the present application.

[0067] In this embodiment, the business district recommendation method based on spatiotemporal trajectory features includes steps S10 to S40:

[0068] Step S10, based on the spatiotemporal large model, generating a user dynamic trajectory representation and a regional dynamic trajectory representation according to the mobile phone signaling data;

[0069] It should be noted that the spatiotemporal model is based on the group spatiotemporal location data and uses the original mobile phone signaling data to generate individual-level, full-time domain high-precision feature vectors to achieve spatiotemporal perception of users and urban spatial areas. The spatiotemporal model extracts spatiotemporal features from massive data and provides a high-dimensional general vector for user / region learning tasks based on user trajectory sequence learning methods.

[0070] In a feasible implementation manner, the step of generating a user dynamic trajectory representation and a regional dynamic trajectory representation based on the mobile phone signaling data based on the spatiotemporal large model includes: acquiring mobile phone signaling data; eliminating abnormal data in the mobile phone signaling data to obtain target mobile phone signaling data; generating a user spatiotemporal triple sequence based on the target mobile phone signaling data based on a grid division result of a spatial area; inputting the user spatiotemporal triple sequence into the spatiotemporal large model to obtain a user dynamic trajectory representation and a regional dynamic trajectory representation, wherein the spatiotemporal large model is constructed using a generative pre-trained transformation model.

[0071] It should be noted that the spatiotemporal large model is constructed using the Generative Pre-trained Transformer 3 (GPT3 model), which is composed of a deep learning structure (Transformer structure) based on multiple stacked self-attention mechanisms. The model uses the self-attention mechanism to capture long-distance dependencies in sequence data. The model predicts where the user may go next and is trained in a self-supervisory manner, so that the spatiotemporal pre-trained large model can realize spatiotemporal position data encoding technology based on semantic understanding of user trajectories.

[0072] It should be noted that before inputting the mobile phone signaling data into the spatiotemporal model, it is necessary to remove abnormal data such as ping-pong, repetition, drift, and redundancy in the mobile phone signaling data to obtain the target mobile phone signaling data. For the spatial area, 200m*200m grid division and encoding (i.e. the grid division result of the spatial area) are performed, and then the user's spatiotemporal triple sequence is generated according to the target mobile phone signaling data:

[0073] [(u i ,a1,t1),(u i ,a2,t2),...,(u i ,a n ,t n )] is used to represent the spatiotemporal sequence of users arranged in chronological order, where u i Represents a user, a i represents the area, t i Indicates time.

[0074] It should be noted that the spatiotemporal model maintains the shared weights of the regional matrix during training. After the model is trained, it infers a high-dimensional general dynamic representation vector that can express users and regions. The user's dynamic trajectory representation is obtained by inputting the user's spatiotemporal sequence and using the model to infer, and the regional dynamic trajectory representation is directly derived through the regional shared weight matrix. The dynamic trajectory representation of its users and regions can learn the mutual influence between users and regions over time, and construct downstream tasks as an effective feature that integrates spatiotemporal trajectory information. The dynamic representation vectors of its users and regions use daily dynamic representations and are updated and iterated every day.

[0075] Step S20, constructing a user multi-source feature according to the user dynamic trajectory representation, and constructing a regional multi-source feature according to the regional dynamic trajectory representation;

[0076] It should be noted that in addition to the representation of the user's dynamic trajectory, the user's multi-source features also include the user's historical visit vector list of business districts, user basic attributes, user Internet information, and user call status.

[0077] In one embodiment, the steps of constructing user multi-source features according to the user dynamic trajectory representation and constructing regional multi-source features according to the regional dynamic trajectory representation include: using the user dynamic trajectory representation as user trajectory information, and determining business district traffic characteristics according to the regional dynamic trajectory representation; constructing user multi-source features according to the user trajectory information, user basic attributes, user call status, user Internet information, user travel information, and a user's historical business district visit vector list, and constructing regional multi-source features according to the business district traffic characteristics, business district basic attributes, business district energy level characteristics, and business district business format information.

[0078] It should be noted that for the user tower, user characteristics are characterized from different dimensions. The multi-source characteristics of users input by the user tower include user trajectory information, user basic attributes, user call status, user Internet behavior (i.e. user Internet information), user driving behavior (i.e. user travel information) and user historical behavior (i.e. user historical visited business district vector list). Specifically, user basic attributes include user ID, age, gender, marital status and user star rating; user call status includes call duration of the caller, number of days the caller makes a call, call duration of the called party, number of days the called party makes a call, monthly number of callers and monthly number of called parties; user Internet behavior includes the number of times a certain type of application is used and the usage time of a certain type of application And the traffic usage of a certain type of application, where the certain type of application includes the following categories: shopping, video, information applications, social, travel, life services, and games; user driving behavior can be determined by the user's itinerary from the starting point to the end point (Origin-Destination, OD), specifically, the user's driving behavior includes the number of connected base stations, the number of monthly ODs, the total monthly OD distance, and the total monthly OD duration; user trajectory information: use the "user dynamic trajectory representation" based on the output of the spatiotemporal large model as the user trajectory feature; user historical visited business district vector list: a list of regional representations of business district areas that users have visited in the past and have a certain length of stay.

[0079] It should be noted that for the regional tower, the regional vectors matched by each business district are first determined to obtain the regional representation corresponding to the business district. For each business district, the input features of the regional tower include the basic attributes of the business district: business district ID, business district geographical location, business district area, business district type; business district energy level characteristics: business district level, business district evaluation, business district popularity, business district passenger flow; business district business format composition: business district consumption level, business district characteristics, business district positioning; business district passenger flow characteristics: use the "regional dynamic trajectory representation" based on the output of the spatiotemporal large model to match the corresponding business district, and obtain the rules and patterns of the business district based on the passenger flow travel trajectory.

[0080] In a feasible implementation manner, the steps of constructing user multi-source features according to the user trajectory information, user basic attributes, user call status, user Internet information, user travel information, and the user's historical business district visit vector list, and constructing regional multi-source features according to the business district crowd flow characteristics, business district basic attributes, business district energy level characteristics, and business district business format information include: classifying the user multi-source features and the regional multi-source features respectively to obtain a first discrete feature, a first continuous feature, a first vector feature, a second discrete feature, a second continuous feature, and a second vector feature; performing unique hot encoding on the first discrete feature and the second discrete feature respectively to obtain a first user dense vector and a first regional dense vector; performing binning on the first continuous feature and the second continuous feature respectively to obtain a second user dense vector and a second regional dense vector; generating new user multi-source features based on the first user dense vector, the second user dense vector, the user's historical business district visit vector list, and the first vector feature; generating new regional multi-source features based on the first regional dense vector, the second regional dense vector, and the second vector feature.

[0081] It should be noted that the first discrete feature, the first continuous feature and the first vector feature refer to user features, and the second discrete feature, the second continuous feature and the second vector feature refer to region features.

[0082] It should be noted that for user multi-source features, it is necessary to distinguish between discrete features, continuous features and vector features. Discrete features refer to features that can be accurately counted and have no intermediate values, including user ID, age, gender, marital status, user star rating, number of days for callers, number of days for called parties, number of monthly callers, number of monthly called parties, number of times a certain type of application is used, number of connected base stations, and number of monthly ODs; continuous features refer to data that can take any value within a range, including call duration for callers, call duration for called parties, time for using a certain type of application, traffic usage for a certain type of application, total monthly OD distance, and total monthly OD duration; vector features include user representations derived from the spatiotemporal large model (i.e. user trajectory information).

[0083] It should be noted that for discrete features, the features are converted from high-dimensional sparse features to low-dimensional dense vectors after one-hot encoding; for continuous features, the features are converted from high-dimensional sparse features to low-dimensional dense vectors after binning, and the binning algorithm here uses chi-square binning; since the vector features are dense vectors, there is no need to process the vector features; for the user's historical visited business district vector list, firstly, the user's historical behavior characteristics are used to filter the areas where the user has visited and has a certain length of stay as the user's historical visited areas, and then the business district global positioning system is used to determine the area to which the business district belongs, and the associated business district areas are arranged in chronological order from far to near to form a user's historical visited business district area list, and the dynamic trajectory representation of all areas is associated according to the identification list, and finally the user's historical visited business district vector list is obtained.

[0084] It should be noted that for user multi-source features, discrete features, continuous features and vector features are distinguished; for discrete features, different data dictionaries are constructed, ID mapping is performed according to the data dictionary to complete one-hot encoding, and further converted into dense vectors using a fully connected layer; for continuous features, buckets are constructed according to the numerical distribution of each feature, one-hot encoding is completed based on the buckets, and further converted into dense vectors using a fully connected layer; finally, the dense vectors converted from different source data are connected to complete the construction of user multi-source features.

[0085] It should be noted that for regional multi-source features, discrete features, continuous features and vector features are distinguished by referring to the construction method of user multi-source features. Discrete features include business district ID, business district geographical location, business district type, business district level, business district evaluation, business district popularity, business district characteristics, and business district positioning; continuous features include business district area, business district passenger flow, and business district consumption level; vector features include regional dynamic trajectory representation output by the spatiotemporal large model. For discrete features, the features are converted from high-dimensional sparse features to low-dimensional dense vectors after one-hot encoding; for continuous features, the features are converted from high-dimensional sparse features to low-dimensional dense vectors after binning. The binning algorithm here uses chi-square binning.

[0086] It should be noted that for regional multi-source features, discrete features, continuous features and vector features are distinguished; for discrete features, different data dictionaries are constructed, ID mapping is performed according to the data dictionary to complete the one-hot encoding, and further converted into dense vectors using a fully connected layer; for continuous features, buckets are constructed according to the numerical distribution of each feature, one-hot encoding is completed based on the buckets, and further converted into dense vectors using a fully connected layer; finally, the dense vectors converted from different source data are connected to complete the construction of regional multi-source features.

[0087] Step S30, inputting the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and inputting the region multi-source features into the region tower of the dual-tower recommendation model to obtain a region representation vector;

[0088] It should be noted that the dual-tower recommendation model is composed of a user tower and a region tower. It makes personalized recommendations by learning vector representations of users and regions. The core is to map users and regions to a common vector space and make recommendations by calculating the similarity between user vectors and region vectors.

[0089] It should be noted that the dual-tower recommendation model can achieve extremely high online reasoning effects, and its corresponding update mechanism is as follows: for feature input, the latest features of users and regions are obtained in real time, including but not limited to real-time distributed publish-subscribe message system (kafka), feature snapshots, etc.; for the dynamic trajectory representation of the spatiotemporal model output, it can be refined to the hour and minute level. For the vector feature, the vector database is used to distinguish time and model version for real-time update to ensure that the online system can smoothly obtain the feature; for the user's historical visit to the business district feature, the user's visit information needs to be aggregated in real time according to the user's historical trajectory, stored and updated. For model updates, the end time node of the upstream feature engineering is used as the model start training time node. After the model is trained, the user model and regional model are exported to the target path for online loading or switching models. The regional model is used to predict all regional vectors and store them in the vector database for real-time calculation by the online system.

[0090] In the specific implementation, Figure 2 As shown in the figure, the user tower includes a user feature embedding layer, a multi-head self-attention mechanism layer (Self-Attention), a user feature cross layer (CROSS_LAYER), a user deep neural network layer (Deep Neural Networks, DNN) and a user normalization layer (L2 normalization); the regional tower includes a regional feature embedding layer (feature embedding), a regional feature cross layer (CROSS_LAYER), a regional deep neural network layer (DNN) and a regional normalization layer (L2 normalization).

[0091] Step S40: determining a business district recommendation result according to the user characterization vector and the region characterization vector.

[0092] It should be noted that the user tower and regional tower in the dual-tower recommendation model can be exported and saved separately as models loaded during actual recommendation. For the user tower and the regional tower, during actual prediction, since the regional representation vector output by the regional tower is very small compared to the magnitude of the user representation vector, the regional representation vector is inferred when the regional tower is exported, and the regional representation vector is stored in the vector database for use at any time.

[0093] In the specific implementation, Figure 3 As shown in the figure, the spatiotemporal model generates user dynamic trajectory representation and regional dynamic trajectory representation based on mobile phone signaling data, and uses the user's historical visit vector list of business districts as a strong feature input recommendation model. The recommendation model adopts a dual-tower recommendation model, which is divided into a user tower and a regional tower. In addition to the user dense vector and the user's historical visit features, the user tower also contains features such as user basic attributes, user Internet behavior, and user call behavior. Except for the dense vector features, the remaining features are all uniquely encoded, and the fully connected layer is concatenated to reduce the dimension to the specified dimension to form a dense vector of each feature; for the user's historical visit area dense vector list, the self-attention mechanism is used to extract the regional features of interest to the user; then these dense vectors are concatenated together to use feature cross-learning to learn the high-dimensional information between different features; finally, the cross feature and the dense vectors of all features are concatenated together, and the fully connected layer is connected to output the final user module representation. The processing method of regional tower is basically similar to that of user tower. User tower and regional tower together constitute a dual-tower model, aligning the feature encodings of the two fields to the same vector space, and finally using the vector product of user tower representation and regional tower representation to indicate whether the user has been to the area, excavating and understanding the hidden deep relationship between users and areas, and learning users' preferences for specific areas. When applying, build a real-time feature engineering storage and query mechanism, a model timing training and update system, and use the dense vectors output by user tower and regional tower to calculate similarity, so as to make real-time recommendations for user travel.

[0094] In a feasible implementation, the step of determining the business district recommendation result based on the user characterization vector and the region characterization vector includes: determining the dot product calculation result of the user characterization vector and the region characterization vector; using the region characterization vector with the maximum dot product calculation result as the target region characterization vector; determining the target business district corresponding to the target region characterization vector, and using the target business district as the business district recommendation result.

[0095] It should be noted that when making real-time recommendations to users online, the user's dynamic representation, the list of historically visited business district vectors and other features are obtained in real time, and the user vector is obtained after the corresponding user tower is input, and the user representation vector and the candidate region representation vector in the vector database are batch-multiplied. For different users, the similarity between them and the region representation vector is obtained, and the final recommendation result is obtained by sorting according to the score (for example, the region representation vector with the largest dot product calculation result is used as the target region representation vector, and then the target business district corresponding to this target region representation vector is recommended to the user).

[0096] This embodiment generates user dynamic trajectory representation and regional dynamic trajectory representation according to mobile phone signaling data based on a spatiotemporal large model; constructs user multi-source features according to the user dynamic trajectory representation, and constructs regional multi-source features according to the regional dynamic trajectory representation; inputs the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and inputs the regional multi-source features into the regional tower of the dual-tower recommendation model to obtain a regional representation vector; determines the business district recommendation result according to the user representation vector and the regional representation vector, thereby solving the technical problem that the traditional business district recommendation scheme lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation result being not accurate enough. Compared with the prior art, this application uses a spatiotemporal large model to extract hidden information between users and regions, and learns users' regular daily behaviors and preference behaviors from massive signaling trajectory data to form user dynamic trajectory representation and regional dynamic trajectory representation, and then uses a separable dual-tower deep model to learn the correlation information between users and business districts, thereby providing users with more accurate business district recommendations.

[0097] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 4 , the user tower includes a user feature embedding layer, a multi-head self-attention mechanism layer, a user feature cross layer, a user deep neural network layer and a user normalization layer, and step S30 also includes steps S301 to S305:

[0098] Step S301, inputting the multi-source features of the user into the user tower, processing the user's historical visited business district vector list through the multi-head self-attention mechanism layer, and obtaining the user history vector;

[0099] In the specific implementation, for the user's historical visited business district vector list, the multi-head self-attention mechanism layer in the user tower is used to learn and extract the associations between the user's historical visited business districts, and obtain the user's historical vector after learning by the multi-head self-attention mechanism layer. The formula of the multi-head self-attention mechanism is as follows:

[0100] MultiHeadAtten(Q,K,V)=Concat(head1,head2,...,head n )W h

[0101] Among them, each head i The calculation process is:

[0102]

[0103] Among them, Q, K, V represent input matrices, which are all input vectors in the self-attention mechanism layer, and W h Represents the matrix of linear change after splicing, W i Q , W i K , W i V Denotes the linear transformation matrix corresponding to each matrix, d k Indicates the dimension of the input query vector and plays a role in numerical adjustment to avoid the inner product being too large.

[0104] Step S302, processing the user multi-source features and the user history vector through the user feature embedding layer to obtain a user splicing vector feature;

[0105] It should be noted that all dense vectors through the user feature embedding layer are concatenated together to form a new vector representation (i.e., user concatenated vector feature), as shown below:

[0106]

[0107] in, Represents the embedding of the real-valued feature after processing. It represents a dense vector feature, which includes the user representation vector and the user history vector after the self-attention mechanism layer. x0 represents the concatenated vector feature (i.e., the user concatenated vector feature). The real-valued feature means that the value of the feature is a real number.

[0108] Step S303, processing the user concatenated vector feature through the user feature cross layer to obtain a user high-order representation after crossover of different features;

[0109] It should be noted that in the user tower, we use the user feature cross layer (i.e., cross layer) in the Deep & Cross Network (DCN) to automatically cross-learn the user concatenated vector features to obtain excellent feature combinations (to obtain user high-order representations). The cross formula is as follows:

[0110]

[0111] in, represents the output of the l,l+1th layer of the cross structure, are the parameters and bias items of this layer respectively.

[0112] Step S304, processing the user concatenated vector features and the user high-order representation through the user deep neural network layer to obtain an initial user vector;

[0113] Step S305 , processing the initial user vector through the user normalization layer to obtain a user representation vector.

[0114] It can be understood that, on the one hand, the user concatenated vector features are learned through the cross structure to obtain the nonlinear features and high-order information after the crossover of different features, thereby obtaining the user high-order representation. On the other hand, the user concatenated vector features and the user high-order representation are concatenated together to enter the DNN network structure (i.e., the user deep neural network layer), output the initial user vector, and then pass through the user normalization layer to output the final user representation vector:

[0115] h l+1 =f(h l W d +b d )

[0116]

[0117] Among them, h l ,h l+1 represents the output of the l,l+1th layer of the fully connected layer, h0 represents the DNN input vector (i.e., the input vector of the user deep neural network layer) concatenated with the user concatenated vector feature and the user high-order representation, the number of DNN layers is L2, and the final user representation vector is:

[0118] user emb =L2normalization(h L2 )

[0119] In a feasible implementation, the regional tower includes a regional feature embedding layer, a regional feature crossing layer, a regional deep neural network layer and a regional normalization layer; the step of inputting the regional multi-source features into the regional tower of the dual-tower model to obtain a user representation vector includes: inputting the regional multi-source features into the regional tower, processing the regional multi-source features through the regional feature embedding layer to obtain regional splicing vector features; processing the regional splicing vector features through the regional feature crossing layer to obtain a regional high-order representation after crossing different features; processing the regional splicing vector features and the regional high-order representation through the regional deep neural network layer to obtain an initial regional vector; processing the initial regional vector through the regional normalization layer to obtain a regional representation vector.

[0120] In the specific implementation, all dense vectors (i.e., regional multi-source features) are spliced ​​together through the regional feature embedding layer to form a new vector representation (i.e., regional splicing vector features), and then the regional splicing vector features are automatically cross-learned through the user feature cross layer to obtain excellent feature combinations to obtain regional high-order representations. On the other hand, the regional splicing vector features and the regional high-order representations are spliced ​​together to enter the DNN network structure (i.e., regional deep neural network layer), output the initial regional vector, and then pass through the regional normalization layer to output the final regional representation vector.

[0121] In a feasible implementation manner, before the step of inputting the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and the step of inputting the regional multi-source features into the regional tower of the dual-tower recommendation model to obtain a regional representation vector, it also includes: inputting the training samples into the initial dual-tower recommendation model to obtain the predicted probability; determining the cross entropy loss value according to the cross entropy loss function based on the true value label of the training samples and the predicted probability; and updating the initial dual-tower recommendation model according to the cross entropy loss value to obtain the dual-tower recommendation model.

[0122] It should be noted that the initial dual-tower recommendation model refers to an untrained model; training samples refer to samples used to train the model, including positive samples and negative samples. In the selection of negative samples, global random negative sampling and mixed sampling of negative sampling in the batch are used to ensure efficient training of sampling in the batch, and to consider global long-tail samples as negative samples in the training process. The training task of the model belongs to a binary classification task, and the cross entropy loss function can be used for training until the model converges. Specifically, the cross entropy loss function is as follows:

[0123]

[0124] Where N represents the number of training samples, y irepresents the true value label of the i-th training sample, y i ' represents the predicted probability of the i-th sample.

[0125] It should be noted that an optimization algorithm (such as gradient descent) can be used to adjust the parameters of the initial dual-tower recommendation model according to the cross-entropy loss value, with the aim of minimizing the loss value, thereby obtaining a trained dual-tower recommendation model and improving the prediction accuracy of the model.

[0126] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the business district recommendation method based on spatiotemporal trajectory features of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0127] This application also provides a business district recommendation device based on spatiotemporal trajectory features, please refer to Figure 5 The device for recommending a business district based on spatiotemporal trajectory features includes:

[0128] A generation module 10, for generating a user dynamic trajectory representation and a regional dynamic trajectory representation according to mobile phone signaling data based on a spatiotemporal large model;

[0129] A construction module 20, configured to construct a user multi-source feature according to the user dynamic trajectory representation, and to construct a region multi-source feature according to the region dynamic trajectory representation;

[0130] An input module 30, configured to input the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and to input the region multi-source features into the region tower of the dual-tower recommendation model to obtain a region representation vector;

[0131] The determination module 40 is used to determine a business district recommendation result according to the user characterization vector and the area characterization vector.

[0132] The device for recommending a business district based on spatiotemporal trajectory features provided by the present application adopts the method for recommending a business district based on spatiotemporal trajectory features in the above-mentioned embodiment, which can solve the technical problem that the traditional business district recommendation scheme lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation result being not accurate enough. Compared with the prior art, the beneficial effects of the device for recommending a business district based on spatiotemporal trajectory features provided by the present application are the same as the beneficial effects of the method for recommending a business district based on spatiotemporal trajectory features provided by the above-mentioned embodiment, and the other technical features of the device for recommending a business district based on spatiotemporal trajectory features are the same as the features disclosed in the above-mentioned embodiment method, which will not be described in detail here.

[0133] In one embodiment, the generation module 10 is also used to obtain mobile phone signaling data; eliminate abnormal data in the mobile phone signaling data to obtain target mobile phone signaling data; based on the grid division result of the spatial area, generate a user space-time triplet sequence according to the target mobile phone signaling data; input the user space-time triplet sequence into the space-time large model to obtain the user dynamic trajectory representation and the regional dynamic trajectory representation, wherein the space-time large model is constructed using a generative pre-trained transformation model.

[0134] In one embodiment, the construction module 20 is further used to use the user dynamic trajectory representation as user trajectory information, and determine the business district traffic characteristics according to the regional dynamic trajectory representation; construct user multi-source characteristics according to the user trajectory information, user basic attributes, user call status, user Internet information, user travel information and the user's historical visit business district vector list, and construct regional multi-source characteristics according to the business district traffic characteristics, business district basic attributes, business district energy level characteristics and business district business format information.

[0135] In one embodiment, the construction module 20 is also used to classify the user multi-source features and the regional multi-source features respectively to obtain a first discrete feature, a first continuous feature, a first vector feature, a second discrete feature, a second continuous feature and a second vector feature; perform one-hot encoding on the first discrete feature and the second discrete feature respectively to obtain a first user dense vector and a first regional dense vector; perform binning on the first continuous feature and the second continuous feature respectively to obtain a second user dense vector and a second regional dense vector; generate a new user multi-source feature based on the first user dense vector, the second user dense vector, the user's historical visited business district vector list and the first vector feature; generate a new regional multi-source feature based on the first regional dense vector, the second regional dense vector and the second vector feature.

[0136] In one embodiment, the user tower includes a user feature embedding layer, a multi-head self-attention mechanism layer, a user feature crossing layer, a user deep neural network layer and a user normalization layer; the input module 30 is also used to input the user multi-source features into the user tower, process the user's historical business district visit vector list through the multi-head self-attention mechanism layer to obtain a user history vector; process the user multi-source features and the user history vector through the user feature embedding layer to obtain a user splicing vector feature; process the user splicing vector feature through the user feature crossing layer to obtain a user high-order representation after crossing different features; process the user splicing vector feature and the user high-order representation through the user deep neural network layer to obtain an initial user vector; process the initial user vector through the user normalization layer to obtain a user representation vector.

[0137] In one embodiment, the regional tower includes a regional feature embedding layer, a regional feature crossing layer, a regional deep neural network layer and a regional normalization layer; the input module 30 is also used to input the regional multi-source features into the regional tower, process the regional multi-source features through the regional feature embedding layer to obtain regional splicing vector features; process the regional splicing vector features through the regional feature crossing layer to obtain a regional high-order representation after crossing different features; process the regional splicing vector features and the regional high-order representation through the regional deep neural network layer to obtain an initial regional vector; process the initial regional vector through the regional normalization layer to obtain a regional representation vector.

[0138] In one embodiment, the input module 30 is also used to input the training samples into the initial dual-tower recommendation model to obtain the predicted probability; based on the true value label of the training samples and the predicted probability, determine the cross entropy loss value according to the cross entropy loss function; update the initial dual-tower recommendation model according to the cross entropy loss value to obtain the dual-tower recommendation model.

[0139] In one embodiment, the determination module 40 is also used to determine the dot product calculation result of the user characterization vector and the region characterization vector; use the region characterization vector with the maximum dot product calculation result as the target region characterization vector; determine the target business district corresponding to the target region characterization vector, and use the target business district as the business district recommendation result.

[0140] The present application provides a business district recommendation device based on spatiotemporal trajectory features, and the business district recommendation device based on spatiotemporal trajectory features includes: 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 the at least one processor can execute the business district recommendation method based on spatiotemporal trajectory features in the above-mentioned embodiment one.

[0141] Reference below Figure 6, which shows a schematic diagram of the structure of a device for recommending a business district based on spatiotemporal trajectory features suitable for implementing an embodiment of the present application. The device for recommending a business district based on spatiotemporal trajectory features in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The business district recommendation device based on spatiotemporal trajectory features shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0142] like Figure 6 As shown, the business district recommendation device based on spatiotemporal trajectory features may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the business district recommendation device based on spatiotemporal trajectory features are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the business district recommendation device based on spatiotemporal trajectory features to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a business district recommendation device based on spatiotemporal trajectory features with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0143] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0144] The business district recommendation device based on spatiotemporal trajectory features provided by the present application adopts the business district recommendation method based on spatiotemporal trajectory features in the above-mentioned embodiment, which can solve the technical problem that the traditional business district recommendation scheme lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation result being not accurate enough. Compared with the prior art, the beneficial effects of the business district recommendation device based on spatiotemporal trajectory features provided by the present application are the same as the beneficial effects of the business district recommendation method based on spatiotemporal trajectory features provided by the above-mentioned embodiment, and the other technical features of the business district recommendation device based on spatiotemporal trajectory features are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0145] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0146] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0147] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the business district recommendation method based on spatiotemporal trajectory features in the above-mentioned embodiment.

[0148] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, 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 above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0149] The computer-readable storage medium may be included in the device for recommending a business district based on spatiotemporal trajectory features; or may exist independently without being assembled into the device for recommending a business district based on spatiotemporal trajectory features.

[0150] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by a business district recommendation device based on spatiotemporal trajectory features, the business district recommendation device based on spatiotemporal trajectory features: generates user dynamic trajectory representation and regional dynamic trajectory representation based on mobile phone signaling data based on a spatiotemporal large model; constructs user multi-source features based on the user dynamic trajectory representation, and constructs regional multi-source features based on the regional dynamic trajectory representation; inputs the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and inputs the regional multi-source features into the regional tower of the dual-tower recommendation model to obtain a regional representation vector; and determines a business district recommendation result based on the user representation vector and the regional representation vector.

[0151] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0152] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0153] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0154] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned business district recommendation method based on spatiotemporal trajectory features, and can solve the technical problem that the traditional business district recommendation scheme lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation results being not accurate enough. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the business district recommendation method based on spatiotemporal trajectory features provided by the above-mentioned embodiment, and will not be repeated here.

[0155] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for recommending a business district based on spatiotemporal trajectory features.

[0156] The computer program product provided by this application can solve the technical problem that the traditional business district recommendation solution lacks the linkage and integration of online and offline information between users and regions, resulting in the final business district recommendation result being inaccurate. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the business district recommendation method based on spatiotemporal trajectory features provided in the above embodiment, and will not be elaborated here.

[0157] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for recommending business districts based on spatiotemporal trajectory features, characterized in that: The method comprises: Based on the spatiotemporal big model, the user dynamic trajectory representation and regional dynamic trajectory representation are generated according to the mobile phone signaling data; constructing a user multi-source feature according to the user dynamic trajectory representation, and constructing a regional multi-source feature according to the regional dynamic trajectory representation; Inputting the user multi-source features into the user tower of the dual-tower recommendation model to obtain a user representation vector, and inputting the region multi-source features into the region tower of the dual-tower recommendation model to obtain a region representation vector; A business district recommendation result is determined according to the user characterization vector and the region characterization vector.

2. The method according to claim 1, characterized in that The step of generating a user dynamic trajectory representation and a regional dynamic trajectory representation based on the spatiotemporal large model according to the mobile phone signaling data comprises: Get mobile phone signaling data; Eliminate abnormal data in the mobile phone signaling data to obtain target mobile phone signaling data; Based on the grid division result of the spatial region, generating a user spatiotemporal triple sequence according to the target mobile phone signaling data; The user spatiotemporal triple sequence is input into a spatiotemporal large model to obtain a user dynamic trajectory representation and a region dynamic trajectory representation, wherein the spatiotemporal large model is constructed using a generative pre-trained transformation model.

3. The method according to claim 1, characterized in that The steps of constructing a user multi-source feature according to the user dynamic trajectory representation and constructing a regional multi-source feature according to the regional dynamic trajectory representation include: Using the user dynamic trajectory representation as user trajectory information, and determining the pedestrian flow characteristics of the business district according to the regional dynamic trajectory representation; The user multi-source features are constructed based on the user trajectory information, user basic attributes, user call status, user Internet information, user travel information and the user's historical visited business district vector list, and the regional multi-source features are constructed based on the business district crowd flow characteristics, business district basic attributes, business district energy level characteristics and business district business format information.

4. The method according to claim 3, characterized in that The steps of constructing user multi-source features according to the user trajectory information, user basic attributes, user call status, user Internet access information, user travel information, and a vector list of user historically visited business districts, and constructing regional multi-source features according to the business district flow characteristics, business district basic attributes, business district energy level characteristics, and business district business format information include: Classifying the user multi-source feature and the region multi-source feature respectively to obtain a first discrete feature, a first continuous feature, a first vector feature, a second discrete feature, a second continuous feature, and a second vector feature; Performing one-hot encoding on the first discrete feature and the second discrete feature respectively to obtain a first user dense vector and a first region dense vector; Binning the first continuous feature and the second continuous feature respectively to obtain a second user dense vector and a second region dense vector; Generate a new user multi-source feature based on the first user dense vector, the second user dense vector, the user's historical visited business district vector list, and the first vector feature; A new regional multi-source feature is generated based on the first regional dense vector, the second regional dense vector, and the second vector feature.

5. The method according to claim 3, characterized in that The user tower includes a user feature embedding layer, a multi-head self-attention mechanism layer, a user feature cross layer, a user deep neural network layer and a user normalization layer; wherein the step of inputting the user multi-source features into the user tower of the dual-tower model to obtain the user representation vector includes: Input the multi-source features of the user into the user tower, and process the user's historical visited business district vector list through the multi-head self-attention mechanism layer to obtain a user history vector; Processing the user multi-source features and the user history vector through the user feature embedding layer to obtain a user splicing vector feature; Processing the user concatenated vector features through the user feature cross layer to obtain a high-order representation of the user after crossover of different features; Processing the user concatenated vector features and the user high-order representation through the user deep neural network layer to obtain an initial user vector; The initial user vector is processed by the user normalization layer to obtain a user representation vector.

6. The method according to claim 3, characterized in that The regional tower includes a regional feature embedding layer, a regional feature cross layer, a regional deep neural network layer and a regional normalization layer; the step of inputting the regional multi-source features into the regional tower of the dual-tower model to obtain the user representation vector includes: Inputting the regional multi-source features into the regional tower, processing the regional multi-source features through the regional feature embedding layer to obtain regional splicing vector features; Processing the regional splicing vector features through the regional feature cross layer to obtain a high-order representation of the region after crossover of different features; Processing the regional splicing vector features and the regional high-order representation through the regional deep neural network layer to obtain an initial regional vector; The initial region vector is processed by the region normalization layer to obtain a region representation vector.

7. The method according to claim 1, characterized in that Before the step of inputting the user multi-source features into the user tower of the dual-tower recommendation model to obtain the user representation vector, and the step of inputting the region multi-source features into the region tower of the dual-tower recommendation model to obtain the region representation vector, the method further includes: Input the training samples into the initial dual-tower recommendation model to obtain the predicted probability; Based on the true value label of the training sample and the predicted probability, determining a cross entropy loss value according to a cross entropy loss function; The initial dual-tower recommendation model is updated according to the cross entropy loss value to obtain a dual-tower recommendation model.

8. The method according to claim 1, characterized in that The step of determining a business district recommendation result according to the user representation vector and the region representation vector comprises: Determine a dot product calculation result of the user characterization vector and the region characterization vector; The region representation vector with the maximum dot product calculation result is used as the target region representation vector; A target business district corresponding to the target area representation vector is determined, and the target business district is used as a business district recommendation result.

9. A device for recommending business districts based on spatiotemporal trajectory features, characterized in that: The device comprises: A generation module, used to generate user dynamic trajectory representation and regional dynamic trajectory representation based on the spatiotemporal large model and mobile phone signaling data; A construction module, configured to construct a user multi-source feature according to the user dynamic trajectory representation, and to construct a region multi-source feature according to the region dynamic trajectory representation; An input module, configured to input the user multi-source features into a user tower of a dual-tower recommendation model to obtain a user representation vector, and to input the region multi-source features into a region tower of a dual-tower recommendation model to obtain a region representation vector; The determination module is used to determine the business district recommendation result according to the user characterization vector and the area characterization vector.

10. A device for recommending business districts based on spatiotemporal trajectory features, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the method for recommending a business district based on spatiotemporal trajectory features as described in any one of claims 1 to 8.

11. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the business district recommendation method based on spatiotemporal trajectory features as described in any one of claims 1 to 8 are implemented.

12. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method for recommending a business district based on spatiotemporal trajectory features as claimed in any one of claims 1 to 8 are implemented.

Images