Map point of interest recommendation method and system based on deep learning model

Abstract

The present disclosure provides a map interest point recommendation method and system based on a deep learning model, which comprises: obtaining interest point data in a map and order data generated by each interest point; generating user trajectory data and preference label data based on the order data; recommending a target user to a map interest point based on the user trajectory data and preference label data, and a pre-trained RNN network model based on item time series association; wherein the recurrent neural network model based on item time series association introduces the forgetting gate idea in the RNN network model structure, selectively integrates the information of the previous hidden layer, the information of the current hidden layer and the information carried by the current element in the sequence in the information propagation process of each layer, and inputs the integrated information into the next hidden layer for further analysis.

Description

Technical Field

[0001] This disclosure belongs to the field of map point of interest recommendation technology, and in particular relates to a map point of interest recommendation method and system based on a deep learning model. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] In the recommendation field, the most commonly used technique is to mine items of interest based on subjective rating information, including collaborative filtering, matrix factorization, and algorithms based on the former two. The accuracy of the above methods depends on the density of the data and whether the rating information is accurate and objective.

[0004] The inventors discovered that in map-based point-of-interest (POI) recommendation applications, users' POI access trajectories are mainly reflected in historical orders related to POIs (such as orders from fast food restaurants or supermarkets) and subjective reviews. Furthermore, user interests are dispersed, and users with multiple shared reviews are rare. Therefore, any method that directly calculates user similarity or predicts preference scores based on a sparse rating matrix will exhibit poor performance initially, encountering a cold start problem. Generally, user feature modeling is one of the most effective ways to address the cold start problem. However, the effectiveness of user feature modeling depends on the granularity of user features. Without enough reference preference labels, recommendation algorithms in sparse data situations can only make broad recommendations based on the most basic user characteristics, failing to refine recommendations based on user uniqueness. Therefore, user feature modeling is generally used to address the problem of new users. Summary of the Invention

[0005] To address the aforementioned issues, this disclosure provides a map point of interest recommendation method and system based on a deep learning model. The proposed solution uses an RNN network based on the temporal correlation of items to optimize the travel route planning function in a community service platform. By using the user's behavioral trajectory generated in the current time period as a basis, the system predicts the points of interest that the target user may visit in the future, thereby improving prediction accuracy and efficiency.

[0006] According to a first aspect of the embodiments of this disclosure, a map point of interest recommendation method based on a deep learning model is provided, comprising:

[0007] Acquire data on points of interest (POIs) in the map and the order data generated by each POI;

[0008] User trajectory data and preference tag data are generated based on the order data;

[0009] Based on the user trajectory data and preference tag data, as well as the pre-trained RNN network model based on item temporal correlation, map interest points are recommended for the target user.

[0010] The recurrent neural network model based on project temporal correlation introduces the forget gate concept into the RNN network model structure. During the information propagation process of each layer, the information of the previous hidden layer, the information of the current hidden layer, and the information carried by the current element in the sequence are selectively integrated, and the integrated information is input into the next hidden layer for further analysis.

[0011] Furthermore, the training process of the recurrent neural network model based on project time-series correlation specifically involves: constructing a training dataset based on historical data from the community service platform; wherein, the samples in the training dataset include a sequence of interest points in the user's historical trajectory of a preset length and the interest points at the next moment; and training the recurrent neural network model based on project time-series correlation based on the dataset to obtain a trained model.

[0012] Furthermore, the recommendation of map points of interest for target users based on the user trajectory data and preference tag data, as well as the pre-trained RNN network model based on item temporal correlation, specifically involves: for users whose interest point sequence length in historical trajectory meets the preset requirements, the interest point sequence of the preset length is input into the pre-trained network model to obtain recommended interest points.

[0013] Furthermore, for new users whose interest point sequence length does not meet the preset requirements, the similarity of the user's preference tag features with other users is calculated, and interest points visited by neighboring users are randomly recommended.

[0014] Furthermore, the new user's preference tags are obtained based on the initial preset values ​​when they register on the community service platform.

[0015] Furthermore, the point-of-interest data includes the overall rating of the point of interest and the corresponding preference tags, and the order data includes the user's identification information and the identification information of the visited point of interest.

[0016] Furthermore, the trajectory data is obtained by sorting the order data in chronological order; the preference tag data includes the type of interest point and whether there are any promotional activities.

[0017] According to a second aspect of the present disclosure, a map point of interest recommendation system based on a deep learning model is provided, comprising:

[0018] The data acquisition unit is used to acquire data on points of interest in the map and order data generated by each point of interest.

[0019] The data processing unit is used to generate user trajectory data and preference tag data based on the order data;

[0020] The point of interest recommendation unit is used to recommend map points of interest to the target user based on the user trajectory data and preference tag data, as well as a pre-trained RNN network model based on item temporal correlation.

[0021] The recurrent neural network model based on project temporal correlation introduces the forget gate concept into the RNN network model structure. During the information propagation process of each layer, the information of the previous hidden layer, the information of the current hidden layer, and the information carried by the current element in the sequence are selectively integrated, and the integrated information is input into the next hidden layer for further analysis.

[0022] According to a third aspect of the present disclosure, a computer-readable storage medium is provided having a program stored thereon that, when executed by a processor, implements a map point of interest recommendation method based on a deep learning model as described above.

[0023] According to a fourth aspect of the present disclosure, an electronic device is provided, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement a map point of interest recommendation method based on a deep learning model as described above.

[0024] Compared with the prior art, the beneficial effects of this disclosure are:

[0025] (1) This disclosure provides a map point of interest recommendation method and system based on a deep learning model. The scheme proposes an RNN network based on the temporal correlation of items. Based on this network, the tourism route planning function in the community service platform is optimized. The user's behavioral trajectory generated in the current time is used as the basis to predict the points of interest that the target user may visit in the future, thereby improving the prediction accuracy and efficiency.

[0026] (2) Compared with user feature modeling methods, the RNN network based on the temporal correlation of items proposed in this disclosure can analyze user trajectories from a more objective perspective, thereby inferring the next item that the target user may visit based on the user's behavioral trajectory generated in the current time. Compared with other recommendation methods based on cookies and browser access records, ISC-RNN can fully explore the implicit relationships between items and continuously fit the user's future access trajectory based on the user's existing traces, avoiding the impact of inaccuracies caused by subjective factors. The proposed solution applies ISC-RNN to the recommendation of map points of interest for target users, and can plan the next location that the user may be interested in based on the user's current location.

[0027] (3) This disclosure proposes an RNN network based on the temporal correlation of items for the trajectory prediction problem. The network is trained using the historical behavior sequence of users in the community platform as the dataset. During the iteration process of the item sequence, it continuously learns the implicit sequential pattern between items, thereby inferring the user's future behavior trajectory based on the learned pattern. Compared with recommendation methods that mainly rely on explicit information and subjective evaluation, ISC-RNN can objectively summarize the trajectory characteristics of users and recommend the points of interest that the target user may visit in the future based on the different trajectories generated by each user. It can also overcome the problem of decreased recommendation accuracy caused by the dispersion of user interests and cold start.

[0028] Advantages of this disclosure in additional aspects will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this disclosure. Attached Figure Description

[0029] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0030] Figure 1 This is a flowchart of the map interest point recommendation method based on a deep learning model as described in Embodiment 1 of this disclosure;

[0031] Figure 2 This is a schematic diagram of the overall structure of the ISC-RNN model described in Embodiment 1 of this disclosure. Detailed Implementation

[0032] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0034] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0035] Where there is no conflict, the embodiments and features described herein can be combined with each other.

[0036] Example 1:

[0037] The purpose of this embodiment is to provide a map point of interest recommendation method based on a deep learning model.

[0038] A map interest point recommendation method based on a deep learning model includes:

[0039] Acquire data on points of interest (POIs) in the map and the order data generated by each POI;

[0040] User trajectory data and preference tag data are generated based on the order data;

[0041] Based on the user trajectory data and preference tag data, as well as the pre-trained RNN network model based on item temporal correlation, map interest points are recommended for the target user.

[0042] The recurrent neural network model based on project temporal correlation introduces the forget gate concept into the RNN network model structure. During the information propagation process of each layer, the information of the previous hidden layer, the information of the current hidden layer, and the information carried by the current element in the sequence are selectively integrated, and the integrated information is input into the next hidden layer for further analysis.

[0043] Furthermore, the training process of the recurrent neural network model based on project time-series correlation specifically involves: constructing a training dataset based on historical data from the community service platform; wherein, the samples in the training dataset include a sequence of interest points in the user's historical trajectory of a preset length and the interest points at the next moment; and training the recurrent neural network model based on project time-series correlation based on the dataset to obtain a trained model.

[0044] Furthermore, the recommendation of map points of interest for target users based on the user trajectory data and preference tag data, as well as the pre-trained RNN network model based on item temporal correlation, specifically involves: for users whose interest point sequence length in historical trajectory meets the preset requirements, the interest point sequence of the preset length is input into the pre-trained network model to obtain recommended interest points.

[0045] Furthermore, for new users whose interest point sequence length does not meet the preset requirements, the similarity of the user's preference tag features with other users is calculated, and interest points visited by neighboring users are randomly recommended.

[0046] Furthermore, the new user's preference tags are obtained based on the initial preset values ​​when they register on the community service platform.

[0047] Furthermore, the point-of-interest data includes the overall rating of the point of interest and the corresponding preference tags, and the order data includes the user's identification information and the identification information of the visited point of interest.

[0048] Furthermore, the trajectory data is obtained by sorting the order data in chronological order; the preference tag data includes the type of interest point and whether there are any promotional activities.

[0049] Specifically, for ease of understanding, the solution described in this disclosure will be explained in detail below with reference to the accompanying drawings:

[0050] like Figure 1 As shown, this disclosure provides a map interest point recommendation method based on a deep learning model, which specifically includes the following steps:

[0051] Step 1: Collect interest point data and order data generated by each interest point from the community service platform, and preprocess the order data to generate user trajectory data and preference tag data.

[0052] Step 2: Design the ISC-RNN network structure. To address the issue of long-term dependency, the information transmission process in the hidden layers is optimized. The optimized ISC-RNN can selectively retain old trajectory feature information while accepting new trajectory feature information, thereby achieving more accurate and comprehensive prediction of the target user's future trajectory.

[0053] Step 3: Iteratively select the user trajectory data from Step 1 using a preset sliding window, and apply it in a special way to the training of ISC-RNN. ISC-RNN learns the user's point of interest access trajectory by taking the previous sequence as input and the future sequence as output.

[0054] Step 4: To address the issue of new users (i.e., users without any historical trajectory) encountered during the recommendation process, a collaborative filtering algorithm based on the similarity of preference tag features is designed to calculate neighboring users with high similarity and recommend the points of interest visited by neighboring users.

[0055] Step 5: If the target user is a new user, the algorithm from Step 4 is used to recommend points of interest visited by nearby users. If the target user already has enough historical data, the trained ISC-RNN model is used to predict the target user's future trajectory. The predicted points of interest are the points of interest the target user is most likely to visit next.

[0056] Furthermore, in step 1, since the application scenario of this disclosure is a community service platform, interest point data and order data generated by each interest point are collected from the community service travel application to analyze the user's trajectory and preference characteristics. The interest point data includes the overall rating of the interest point and its corresponding preference tags (such as location type, availability of promotional activities, etc.), while the order data includes the user's identification information and the identification information of the visited interest points. Sorting the order data chronologically yields the user's trajectory data, which reflects the implicit temporal correlation between map interest points. Additionally, the number of visits to interest points and their corresponding preference tags is counted in the user's trajectory data. The target user's preference tag access frequency is represented as a vector, where each dimension represents a preference tag, and the value corresponding to each dimension represents the number of times the preference tag is accessed.

[0057] Furthermore, in step 2, the ISC-RNN design is based on the RNN structure. However, the network depth of an RNN is determined by the length of the input sequence during each training iteration. During the backpropagation optimization process of a multi-layer network, the RNN model is prone to gradient vanishing or gradient exploding problems due to the addition or multiplication of gradient calculation functions between layers. Additionally, if the learning sequence length of the RNN is too long, the model may not be able to memorize all the information contained in the long sequence. In solving long-term dependency problems, it is necessary to memorize and analyze older information from the input sequence to output more comprehensive and accurate prediction results. Therefore, the ISC-RNN model references the forget gate concept in LSTM, selectively integrating information from the previous hidden layer, the current hidden layer, and the information carried by the current element in the sequence during the information propagation process of each layer, and inputting the integrated information into the next hidden layer for further analysis. Finally, the ISC-RNN model integrates information from all hidden layers and all elements in the sequence and outputs relatively accurate prediction values. Compared to other models based on RNN or LSTM structures, ISC-RNN can selectively integrate information from various network layers and elements while maintaining accuracy, avoiding excessive computational complexity due to information overload, and its structure is relatively simple. To address the vanishing and exploding gradient problems, the ReLU function is used as the activation function for information propagation in each layer. The conceptual formula for ISC-RNN is given in the specific implementation.

[0058] Further, in step 3, the ISC-RNN model is trained using a user history sequence of a certain length as input and the subsequent interest point information as output. In the scenario applied in this disclosure, the input length of the user sequence is at least 5, while the interest point to be predicted is only 1. Therefore, in each iteration, a sliding window of length 6 is used to select 6 consecutive interest points from the user history sequence to train the network, where the first 5 interest points are the network input and the 6th interest point is the network output. Furthermore, the sliding window moves by 1 unit each time. After multiple iterations, the ISC-RNN model learns the implicit relationships between interest points from multiple consecutive sequences. Based on the learned patterns, the ISC-RNN model can predict the 1st interest point following a random interest point sequence of length 5, and the predicted result is the interest point that the target user is most likely to visit in the future. The representation of interest point information in the ISC-RNN model will be described in detail below.

[0059] Furthermore, in step 4, the "new user" problem refers to users who are using the community service platform for the first time and have not left any historical records. In this case, the ISC-RNN model is not applicable to this type of target user. For the new user problem, this disclosure designs a collaborative filtering algorithm based on the similarity of preference tag features to calculate neighboring users with high similarity and recommends points of interest visited by neighboring users to the target user. For newly registered users, their preference tag features are generally obtained through initial user settings, and the value of each dimension is set to 1 by default. The preference tag features are represented as described in step 1, in the form of vectors, with the value of each dimension corresponding to the possible access frequency of the user under that tag. The similarity between vectors is mainly calculated using cosine similarity:

[0060]

[0061] Here, La and Lb are the vectors representing the preference label features of users a and b, respectively, and their similarity can be calculated using cosine similarity. Finally, the similarity of the target user's preference labels with other users is calculated to select neighboring users with high similarity, and the points of interest visited by these neighboring users are recommended to the target user. The collaborative filtering algorithm based on preference label feature similarity is only suitable for solving the problem of new users. Once the length of a user's historical trajectory reaches the input requirement of the ISC-RNN model, the community service platform switches to the ISC-RNN model to recommend points of interest that the target user may visit in the future. According to the training method of the ISC-RNN model in step 3, the length of the input user's historical trajectory is at least 5, which means that the user only needs to leave 5 points of interest access records in the community service platform to obtain the points of interest prediction results that match their behavioral trajectory through the ISC-RNN model.

[0062] Further, in step 5, the historical trajectory length of the target user is first analyzed. If the user's historical trajectory length does not meet the input requirements of the ISC-RNN model, the user is considered a new user, and neighboring users with high similarity are calculated based on the user's preference label features. This allows for further recommendations of points of interest visited by these neighboring users. Additionally, if the user has a sufficiently long trajectory of points of interest, the nearest subsequence of this trajectory is input into the ISC-RNN model. The ISC-RNN then predicts the points of interest the target user might visit in the future based on the learned sequential relationships between these points of interest. Finally, the prediction result from the ISC-RNN is the recommendation result that matches the user's trajectory characteristics.

[0063] Specifically, the main purpose of the scheme described in this disclosure is to model the user's historical trajectory, analyze the potential temporal relationships between points of interest in the user's trajectory, and recommend points of interest that the target user may visit in the future based on this. Compared with other recommendation algorithms based on subjective rating information, the Recurrent Neural Networks based on Item Sequence Correlations (ISC-RNN) proposed in this disclosure can learn the temporal correlation features of the user's trajectory, thereby accurately predicting the next point of interest that the user will visit. Since ISC-RNN analyzes user trajectories, even if the user's interests are relatively scattered, it can provide the most suitable candidate points of interest for the target user based on the trajectory features of each user. In addition, to solve the problem of new users, this disclosure models the user in the form of preference label features and proposes a collaborative filtering algorithm based on the similarity of preference label features to calculate the target user's nearest neighbors, thereby further recommending points of interest visited by the nearest neighbors to the target user. Furthermore, collaborative filtering algorithms based on the similarity of preference tag features are only suitable for solving the problem of new users. In the application scenarios involved in this disclosure, the ISC-RNN model has low requirements for the length of user trajectories. Users only need to leave a small number of records in the community service platform to obtain recommended interest points through the ISC-RNN model.

[0064] Through the analysis and processing of the above data, user trajectory features and preference tag features are extracted, preparing for the next step of training the ISC-RNN model. Before model training, this disclosure optimizes the model to address potential problems. Generally, the network depth of an RNN is determined by the length of the input sequence in each training iteration. During the backpropagation optimization process of multi-layer networks, the RNN model is prone to gradient vanishing or gradient exploding problems due to the addition or multiplication of gradient calculation functions between layers. Furthermore, if the learning sequence length of the RNN is too long, the model may not be able to memorize all the information contained in the long sequence. In solving long-term dependent problems, it is necessary to memorize and analyze relatively old information in the input sequence to output more comprehensive and accurate prediction results. Therefore, the ISC-RNN model references the forget gate idea in LSTM, selectively integrating the information from the previous hidden layer, the current hidden layer, and the information carried by the current element in the sequence during the information propagation process of each layer, and inputting the integrated information into the next hidden layer for further analysis. Finally, the ISC-RNN model integrates the information from all hidden layers and all elements in the sequence and outputs relatively accurate prediction values. Compared to other models based on RNN or LSTM structures, ISC-RNN can selectively integrate information from various network layers and elements while maintaining accuracy, avoiding excessive computational complexity due to information overload, and its structure is relatively simple. To address the vanishing and exploding gradient problems, the ReLU function is used as the activation function for information propagation in each layer. The conceptual formula of ISC-RNN is briefly described below:

[0065] T k =φ(T k-1 )+w fr(k-1) T k-1 +w curr (k)X k

[0066] Among them, T k This represents the information of the k-th layer, while T k-1φ represents the information from the previous layer (i.e., layer k-1), and φ represents the activation function used in each hidden layer during information propagation. Because the sigmoid or tanh functions are prone to causing excessively high gradient values ​​(gradient explosion) or excessively low gradient values ​​(gradient vanishing) due to continuous multiplication of derivatives during gradient calculation, ReLU is used as the activation function in each layer's propagation. The derivative of the ReLU function is 0 or 1, and it does not decay to 0 during backpropagation, thus it is suitable for solving the gradient vanishing problem. Furthermore, to solve the gradient explosion problem, a threshold is usually set for the model's gradient. When the gradient value exceeds the pre-set threshold during optimization, the gradient value is set to the threshold. From the above, it can be seen that if w... fr(k-1) When the value of w is 1, its function is equivalent to a network model that completely preserves information. fr(k-1) If the value is 0, the model is equivalent to a regular RNN network. Therefore, the information retention parameter for each layer needs to be learned through an appropriate objective function so that the ISC-RNN model does not become too computationally complex due to information redundancy while achieving the required prediction accuracy.

[0067] Furthermore, the ISC-RNN model is trained by taking a user history sequence of a certain length as input and the subsequent interest point information as output. In the scenario applied in this disclosure, the input length of the user sequence is at least 5, while the interest point to be predicted is only 1. Therefore, in each iteration, a sliding window of length 6 is used to select 6 consecutive interest points in the user history sequence to train the network, where the first 5 interest points are the network input and the 6th interest point is the network output. In addition, the sliding window moves by 1 unit each time. After multiple iterations, the ISC-RNN model learns the implicit relationships between interest points from multiple consecutive sequences. Based on the learned patterns, the ISC-RNN model can predict the 1st interest point following a random interest point sequence of length 5, and the predicted result is the interest point that the target user is most likely to visit in the future. The structure and training process of the ISC-RNN model are as follows: Figure 2 As shown. In each analysis process, the interest point information is compressed into vectors and input into the ISC-RNN model, and the final output is also in vector form, showing the predicted interest point information. To measure the model's accuracy, the model optimizes its parameters through backpropagation based on the error between the predicted and true interest points. The loss function is the sum of the squared distances between all true and predicted interest point vectors. The definition of the loss function is as follows:

[0068]

[0069] Where X is the set of input sequences for the model, Y is the set of true interest point vectors, and f(X) represents the set of predicted interest point vectors. d(x, y) represents the distance function that measures the difference between vectors. In the scenario applied in this disclosure, the Euclidean distance function is used to calculate the distance between the predicted and true values.

[0070] Because the ISC-RNN model requires a certain length of user trajectory information (which can be set according to actual needs) as a basis to predict the most likely points of interest that a target user will visit in the future, its applicability is poor when encountering target users with little or no trajectory information or almost no trace of interest point visits. Therefore, to solve the problem of such target users, this disclosure designs a collaborative filtering algorithm based on the similarity of preference label features to calculate neighboring users with high similarity and recommend the points of interest visited by neighboring users to the target user. For newly registered users, their preference label features are generally obtained through initial user settings, and the value of each dimension is set to 1 by default. Since the preference label features are represented as in step 1, in the form of vectors, the value of each dimension corresponds to the possible access frequency of the user under that label. The calculation of the similarity between vectors is mainly based on cosine similarity:

[0071]

[0072] Among them, L a With L b The similarity between the vectors representing the preference label features of users a and b can be calculated using cosine similarity. Finally, the similarity of the target user's preference labels with other users is calculated to select neighboring users with high similarity, and the points of interest visited by these neighboring users are recommended to the target user. The collaborative filtering algorithm based on preference label feature similarity is only suitable for solving the problem of new users. Once the length of a user's historical trajectory reaches the input requirement of the ISC-RNN model, the community service platform switches to the ISC-RNN model to recommend points of interest that the target user may visit in the future. According to the training method of the ISC-RNN model mentioned earlier, the length of the input user's historical trajectory is at least 5, which means that the user only needs to leave 5 points of interest access records in the community service platform to obtain points of interest prediction results that match their behavioral trajectory through the ISC-RNN model.

[0073] In summary, when recommending points of interest (POIs) to target users, this disclosure first assesses the length of the target user's historical trajectory. If the target user has insufficient information on the community service platform, it calculates highly similar neighboring users based on the user's own preference tags and recommends POIs visited by these neighboring users. If the target user has sufficient historical trajectory data for analysis, it predicts potential future map POIs using an ISC-RNN model and recommends these POIs to the target user.

[0074] Example 2:

[0075] The purpose of this embodiment is to provide a map point of interest recommendation system based on a deep learning model.

[0076] A map point of interest recommendation system based on a deep learning model, comprising:

[0077] The data acquisition unit is used to acquire data on points of interest in the map and order data generated by each point of interest.

[0078] The data processing unit is used to generate user trajectory data and preference tag data based on the order data;

[0079] The point of interest recommendation unit is used to recommend map points of interest to the target user based on the user trajectory data and preference tag data, as well as a pre-trained RNN network model based on item temporal correlation.

[0080] The recurrent neural network model based on project temporal correlation introduces the forget gate concept into the RNN network model structure. During the information propagation process of each layer, the information of the previous hidden layer, the information of the current hidden layer, and the information carried by the current element in the sequence are selectively integrated, and the integrated information is input into the next hidden layer for further analysis.

[0081] In further embodiments, the following is also provided:

[0082] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0083] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0084] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0085] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0086] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0087] Those skilled in the art will recognize that the units, i.e., algorithm steps, of the various examples described in connection with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. 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 disclosure.

[0088] The map point of interest recommendation method and system based on a deep learning model provided in the above embodiments can be implemented and has broad application prospects.

[0089] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A map point of interest recommendation method based on a deep learning model, characterized in that, include: Acquire data on points of interest (POIs) in the map and the order data generated by each POI; User trajectory data and preference tag data are generated based on the order data; Based on the user trajectory data and preference tag data, as well as the pre-trained RNN network model based on item temporal correlation, map interest points are recommended for the target user. The recurrent neural network model based on project temporal correlation introduces the forget gate concept into the RNN network model structure. During the information propagation process of each layer, the information of the previous hidden layer, the information of the current hidden layer, and the information carried by the current element in the sequence are selectively integrated, and the integrated information is input into the next hidden layer for further analysis.

2. The map interest point recommendation method based on a deep learning model as described in claim 1, characterized in that, The training process of the recurrent neural network model based on project time-series correlation is as follows: a training dataset is constructed based on historical data in the community service platform; wherein, the samples in the training dataset include a sequence of interest points in the user's historical trajectory of a preset length and the interest points at the next moment; the recurrent neural network model based on project time-series correlation is trained based on the dataset to obtain a trained model.

3. The map interest point recommendation method based on a deep learning model as described in claim 1, characterized in that, The method of recommending map points of interest for target users based on the user trajectory data and preference tag data, as well as the pre-trained RNN network model based on item temporal correlation, specifically involves: for users whose interest point sequence length in historical trajectory meets the preset requirements, the interest point sequence of the preset length is input into the pre-trained network model to obtain recommended interest points.

4. The map interest point recommendation method based on a deep learning model as described in claim 3, characterized in that, For new users whose interest point sequence length does not meet the preset requirements, calculate the similarity of the user's preference tag features with other users, and randomly recommend interest points visited by neighboring users.

5. The map interest point recommendation method based on a deep learning model as described in claim 4, characterized in that, The new user's preference tags are obtained based on the initial preset values ​​when they register on the community service platform.

6. The map interest point recommendation method based on a deep learning model as described in claim 1, characterized in that, The point of interest data includes the overall rating of the point of interest and the corresponding preference tags, and the order data includes the user's identification information and the identification information of the point of interest visited.

7. The map interest point recommendation method based on a deep learning model as described in claim 1, characterized in that, The trajectory data is obtained by sorting the order data in chronological order; the preference tag data includes the type of interest point and whether there are any promotional activities.

8. A map point of interest recommendation system based on a deep learning model, characterized in that, include: The data acquisition unit is used to acquire data on points of interest in the map and order data generated by each point of interest. The data processing unit is used to generate user trajectory data and preference tag data based on the order data; The point of interest recommendation unit is used to recommend map points of interest to the target user based on the user trajectory data and preference tag data, as well as a pre-trained RNN network model based on item temporal correlation. The recurrent neural network model based on project temporal correlation introduces the forget gate concept into the RNN network model structure. During the information propagation process of each layer, the information of the previous hidden layer, the information of the current hidden layer, and the information carried by the current element in the sequence are selectively integrated, and the integrated information is input into the next hidden layer for further analysis.

9. A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements a map point of interest recommendation method based on a deep learning model as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a map interest point recommendation method based on a deep learning model as described in any one of claims 1-7.

Images