A Location Prediction Method and System Based on Bidirectional Inference and Two-Level Association
By employing a location prediction method that performs bidirectional inference and two-layer association from both the user and location perspectives, the problem of low prediction accuracy caused by data sparsity in existing technologies is solved, achieving higher-precision location prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-31
- Publication Date
- 2026-03-10
AI Technical Summary
Existing location prediction methods mainly focus on the user's perspective, which is limited by data sparsity, resulting in low accuracy of prediction results.
We employ a bidirectional inference and two-layer association approach to predict location from both the user and location perspectives. We process data through an embedding layer and an RNN network, optimize the candidate location and visitor ranking matrix, and perform bidirectional inference to obtain the final location prediction result.
It improves the accuracy of location prediction by integrating user and location perspective information, thereby mitigating data sparsity and enhancing the accuracy of prediction results.
Smart Images
Figure CN115357807B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information prediction technology, specifically to a location prediction method and system based on bidirectional inference and two-layer association. Background Technology
[0002] Location prediction is a technique that predicts a user's likely future location based on their historical trajectory data over a period of time. It is commonly used in areas such as intelligent transportation, intelligent advertising, travel recommendations, and crime prediction, making it a hot research topic in current big data analytics and applications.
[0003] In recent years, there has been a large amount of research on location prediction. These studies can be mainly divided into three categories: matrix factorization-based methods (e.g., Xingyi Ren, Meina Song, E Haihong, and Junde Song. Context-aware probabilistic matrix factorization modeling for point-of-interest recommendation. Neurocomputing, 241:38–55, 2017.), Markov chain-based methods (e.g., Jia-Dong Zhang, Chi-Yin Chow, and Yanhua Li. Lore: Exploiting sequential influence for location recommendations. In ACM SIGSPATIAL, pages 103–112, 2014.), and deep network-based methods (e.g., Antonios Karatzoglou, Adrian Jablonski, and Michael Beigl. A seq2seq learning approach for modeling semantictrajectories and predicting the next location. In ACM SIGSPATIAL, pages 528–531, 2018.).
[0004] However, to our knowledge, these methods only study from the user's perspective, that is, focusing on users' location preferences. While these works strive to enrich the modeling of user preferences, they are still limited by data sparsity. Users on social networks typically do not leave real-time tracking information; each user's records are geographically and temporally distant. Therefore, location prediction remains a challenging task. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the above-mentioned background technology and provide a location prediction method and system based on bidirectional inference and two-layer association.
[0006] In a first aspect, this application provides a location prediction method based on bidirectional inference and two-layer association, comprising the following steps:
[0007] Step S1: Obtain the candidate location ranking matrix S of the predicted access locations from the user's perspective. U ;
[0008] Step S2: Obtain the candidate visitor ranking matrix S of the predicted visiting users from the perspective of location. L ;
[0009] Step S3: Optimize and obtain the candidate position sorting matrix S U and candidate visitor ranking matrix S L The predictive effect;
[0010] Step S4: Obtain the new candidate position sorting matrix after reordering.
[0011] Step S5: Obtain the reordered new candidate visitor ranking matrix.
[0012] Step S6, according to and Two-way inference is performed to obtain the final location prediction result.
[0013] According to the first aspect, in a first possible implementation of the first aspect, step S1 specifically includes the following steps:
[0014] Step S11: Using the sequence of access locations for each user as input, the embedding layer processes the data to obtain the embedding representation vector e. l ;
[0015] Step S12: Using the embedded expression vector e l The input is processed by an RNN network to obtain a sorted list of candidate positions, s. U ;
[0016] Step S13: Sort the list s according to each user's candidate location. U The candidate position ranking matrix S of all users is obtained. U .
[0017] According to the first aspect, in a second possible implementation of the first aspect, step S2 specifically includes the following steps:
[0018] Step S21: Using the visitor sequence for each location as input, the embedding expression vector e is obtained through the embedding layer. u ;
[0019] Step S22: Using the embedded expression vector e u The input is processed by an RNN network to obtain a ranked list of candidate visitors, s. L ;
[0020] Step S23: Based on the candidate visitor ranking list for each location, obtain the candidate visitor ranking matrix S for all locations. L .
[0021] According to the first aspect, in the third possible implementation of the first aspect, step S3 specifically includes the following steps:
[0022] Step S30: Obtain the candidate position sorting matrix S through iterative optimization using a shared loss function system. U And obtain the candidate visitor ranking matrix S L The predictive effect.
[0023] According to the first aspect, in the fourth possible implementation of the first aspect, step S4 specifically includes the following steps:
[0024] Step S41: Calculate the similarity between the candidate position ranking matrices of every two users, and obtain the user similarity matrix C between the candidate position ranking matrices of all users. U ;
[0025] Step S42: Filter and process the user similarity matrix C U Obtain the filtered user similarity matrix
[0026] Step S43: Based on the filtered user similarity matrix Reorder the candidate position sorting matrix for the current user to obtain a new candidate position sorting matrix.
[0027] According to the first aspect, in the fifth possible implementation of the first aspect, in step S42, the user similarity matrix C is... U Elements smaller than the first threshold are set to 0 to obtain the filtered user similarity matrix.
[0028] According to the first aspect, in the sixth possible implementation of the first aspect, step S5 specifically includes the following steps:
[0029] Step S51: Calculate the similarity of the candidate visitor ranking matrix between every two locations, and obtain the location similarity matrix C for all locations.L ;
[0030] Step S52: Filter the location similarity matrix to obtain the filtered location similarity matrix.
[0031] Step S53: Based on the filtered location similarity matrix Reorder the candidate visitor ranking matrix for the current user to obtain a new candidate visitor ranking matrix.
[0032] According to the first aspect, in the seventh possible implementation of the first aspect, in step S52, the location similarity matrix C is... U Elements smaller than the second threshold are set to 0, and the filtered location similarity matrix is obtained.
[0033] Secondly, this application discloses a location prediction system based on bidirectional inference and two-layer association, comprising:
[0034] The candidate location prediction and acquisition module is used to obtain the candidate location ranking matrix S, which is the predicted access location from the user's perspective. U ;
[0035] The candidate visitor prediction and acquisition module is used to obtain the candidate visitor ranking matrix S of visitors predicted from the perspective of location. L ;
[0036] The optimization module, communicatively connected to the candidate location prediction and acquisition module and the candidate visitor prediction and acquisition module, is used to optimize the acquisition of the candidate location sorting matrix S. U and candidate visitor ranking matrix S L The predictive effect;
[0037] A new candidate position acquisition module is used to obtain the new candidate position sorting matrix after reordering.
[0038] A new candidate visitor acquisition module is used to obtain a new, reordered candidate visitor ranking matrix.
[0039] The final location acquisition module is communicatively connected to the new candidate location acquisition module and the new candidate visitor acquisition module, and is used to determine the final location based on the location acquisition module and the visitor acquisition module. and Two-way inference is performed to obtain the final location prediction result.
[0040] According to the second aspect, in a first possible implementation of the first aspect, the candidate location prediction acquisition module includes:
[0041] The location embedding representation vector acquisition unit is used to obtain the embedding representation vector e by taking the sequence of each user's access locations as input and processing it through the embedding layer. l ;
[0042] The candidate location sorting list acquisition unit is communicatively connected to the location embedding expression vector acquisition unit, and is used to obtain the embedding expression vector e. l The input is processed by an RNN network to obtain a sorted list of candidate positions, s. U ;
[0043] The candidate position sorting matrix acquisition unit is communicatively connected to the candidate position sorting list acquisition unit and is used to acquire the candidate position sorting matrix.
[0044] Compared with the prior art, the advantages of the present invention are as follows:
[0045] This application effectively improves the accuracy of location prediction by obtaining the final location prediction result through bidirectional inference from the user's perspective and from the location's perspective, as well as by reordering the candidate locations and candidate visitor ranking matrix. Attached Figure Description
[0046] Figure 1 This is a flowchart of the location prediction method based on bidirectional inference and two-layer association according to an embodiment of the present invention;
[0047] Figure 2 This is another method flowchart of the location prediction method based on bidirectional inference and two-layer association according to an embodiment of the present invention;
[0048] Figure 3 This is a functional block diagram of the location prediction system based on bidirectional inference and two-layer association according to an embodiment of the present invention;
[0049] Figure 4 This is another functional module block diagram of the location prediction system based on bidirectional inference and two-layer association in an embodiment of the present invention. Detailed Implementation
[0050] Referring now to specific embodiments of the invention, examples of which are illustrated in the accompanying drawings. Although the invention will be described in conjunction with specific embodiments, it will be understood that it is not intended to limit the invention to the described embodiments. Rather, it is intended to cover variations, modifications, and equivalents included within the spirit and scope of the invention as defined by the appended claims. It should be noted that the method steps described herein can be implemented by any functional block or functional arrangement, and any functional block or functional arrangement can be implemented as a physical entity or a logical entity, or a combination of both.
[0051] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0052] Note: The examples described below are merely specific examples and are not intended to limit the embodiments of the present invention to the specific steps, values, conditions, data, order, etc. Those skilled in the art can utilize the concept of the present invention to construct more embodiments not mentioned herein by reading this specification.
[0053] There are three main categories of research methods in existing location prediction technologies: matrix factorization-based methods, Markov chain-based methods, and deep network-based methods. However, these methods only study from the user's perspective, focusing on user preferences for location. They are limited by data sparsity and have low prediction accuracy.
[0054] In view of this, in order to improve the prediction accuracy of location prediction results, see [reference needed]. Figure 1 As shown, this embodiment of the invention provides a location prediction method based on bidirectional inference and two-layer association, including the following steps:
[0055] Step S1: Obtain the candidate location ranking matrix S of the predicted access locations from the user's perspective. U ;
[0056] Step S2: Obtain the candidate visitor ranking matrix S of the predicted visiting users from the perspective of location. L ;
[0057] Step S3: Optimize and obtain the candidate position sorting matrix S U and candidate visitor ranking matrix S L The predictive effect;
[0058] Step S4: Obtain the new candidate position sorting matrix after reordering.
[0059] Step S5: Obtain the reordered new candidate visitor ranking matrix.
[0060] Step S6, according to and Two-way inference is performed to obtain the final location prediction result.
[0061] This application improves the accuracy of location prediction by inferring from both the user's and location's perspectives and obtaining the final location prediction result based on the reordered candidate locations and candidate visitor ranking matrix.
[0062] In one embodiment, please refer to Figure 2 Step S1 specifically includes the following steps:
[0063] Step S11: Sequence of each user's access locations As input, the embedded representation vector e is obtained after processing by the embedding layer. l The role of the embedding layer is to represent sparse data (one-hot encoding) as a denser low-dimensional vector. The role of the embedding layer is to map high-dimensional sparse data to low-dimensional dense data. It is a type of neural network and can be directly implemented using PyTorch or TensorFlow.
[0064] Step S12: Using the embedded expression vector e l The input is processed by an RNN network to obtain a sorted list of candidate positions, s. U ;
[0065] Step S13: Sort the list s according to each user's candidate location. U The candidate position ranking matrix S of all users is obtained. U .
[0066] In one embodiment, step S2 specifically includes the following steps:
[0067] Step S21: Using the visitor sequence for each location as input, the embedding expression vector e is obtained through the embedding layer. u ;
[0068] Step S22: Using the embedded expression vector e u The input is processed by an RNN network to obtain a ranked list of candidate visitors, s. L ;
[0069] Step S23: Based on the candidate visitor ranking list for each location, obtain the candidate visitor ranking matrix S for all locations. L .
[0070] In one embodiment, step S3 specifically includes the following steps:
[0071] Step S30: Obtain the candidate position sorting matrix S through iterative optimization using a shared loss function system. U And obtain the candidate visitor ranking matrix S L The predictive effect.
[0072] In a more specific embodiment, the candidate position sorting matrix S U And obtain the candidate visitor ranking matrix S L The corresponding element values in the two matrices are subtracted, and the sum of the differences is used as the loss function. Training ends when the loss function no longer decreases during the iteration process, and the candidate position sorting matrix S is obtained through optimization. U And obtain the candidate visitor ranking matrix S L The predictive effect.
[0073] In one embodiment, step S4 specifically includes the following steps:
[0074] Step S41: Calculate the similarity between the candidate position ranking matrices of every two users, and obtain the user similarity matrix C between the candidate position ranking matrices of all users. U ;
[0075] Step S42: Filter and process the user similarity matrix C U Obtain the filtered user similarity matrix To filter out useless or duplicate user information;
[0076] Step S43: Based on the filtered user similarity matrix Reorder the candidate position sorting matrix for the current user to obtain a new candidate position sorting matrix. The specific formula is as follows:
[0077] In one embodiment, in step S42, the user similarity matrix C is analyzed. U The value is less than the first threshold θ u By setting the elements to 0, the filtered user similarity matrix can be obtained. Users corresponding to the remaining non-zero elements after filtering are selected as the subset of users most similar to the target user.
[0078] In one embodiment, step S5 specifically includes the following steps:
[0079] Step S51: Calculate the similarity of the candidate visitor ranking matrix between every two locations, and obtain the location similarity matrix C for all locations. L ;
[0080] Step S52: Filter the location similarity matrix to obtain the filtered location similarity matrix. To filter out useless location information;
[0081] Step S53: Based on the filtered location similarity matrix Reorder the candidate visitor ranking matrix for the current user to obtain a new candidate visitor ranking matrix. The specific formula is as follows:
[0082] In one embodiment, in step S52, the location similarity matrix C is... U Elements smaller than the second threshold are set to 0, and the filtered location similarity matrix is obtained.
[0083] In one embodiment, step S6 specifically includes the following steps:
[0084] Based on the prediction results from the two angles, bidirectional inference is performed to obtain the final position prediction result S. F This yields the final location prediction result. Bidirectional inference is achieved through maxpooling, with the specific formula being... The maxpooling operation takes the maximum value at corresponding positions in two matrices. When performing bidirectional inference, if we only infer the locations that user will visit in the future from the perspective of user 1, the result is a list of location probabilities, where the probability of visiting location A is very small and the probability of visiting location B is relatively large. However, when inferring the possible users visiting a location from the perspective of the location, we find that user 1 has a high probability in the list of user probabilities visiting location A. Therefore, in bidirectional inference, considering the inference results from both perspectives will correct the list of locations that user 1 may visit.
[0085] Based on the same inventive concept, please refer to Figure 3 This application discloses a location prediction system based on bidirectional inference and two-layer association, comprising:
[0086] The candidate location prediction acquisition module 100 is used to acquire the candidate location ranking matrix S of the access locations predicted from the user's perspective. U ;
[0087] The candidate visitor prediction and acquisition module 200 is used to obtain the candidate visitor ranking matrix S of visiting users predicted from the perspective of location. L ;
[0088] The optimization module 300, which is communicatively connected to the candidate location prediction and acquisition module 100 and the candidate visitor prediction and acquisition module 200, is used to optimize the acquisition of the candidate location sorting matrix S. U and candidate visitor ranking matrix S L The predictive effect;
[0089] The new candidate position acquisition module 400 is used to acquire the reordered new candidate position sorting matrix.
[0090] The new candidate visitor acquisition module 500 is used to obtain the reordered new candidate visitor ranking matrix.
[0091] The final location acquisition module 600 is communicatively connected to the new candidate location acquisition module 400 and the new candidate visitor acquisition module 500, and is used to determine the final location based on the location of the visitor. and Two-way inference is performed to obtain the final location prediction result.
[0092] In one embodiment, please refer to Figure 4The candidate location prediction and acquisition module includes:
[0093] Location embedding vector acquisition unit 110 is used to obtain the embedding vector e by taking the sequence of access locations of each user as input and processing it through the embedding layer. l ;
[0094] The candidate location sorting list acquisition unit 120 is communicatively connected to the location embedding expression vector acquisition unit, and is used to obtain the embedding expression vector e. l The input is processed by an RNN network to obtain a sorted list of candidate positions, s. U ;
[0095] The candidate position sorting matrix acquisition unit 130 is communicatively connected to the candidate position sorting list acquisition unit and is used to acquire the candidate position sorting matrix.
[0096] Based on the same inventive concept, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements all or part of the method steps of the above method.
[0097] The present invention can implement all or part of the processes in the above methods, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.
[0098] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a memory and a processor. The memory stores a computer program that runs on the processor. When the processor executes the computer program, it implements all or part of the method steps described above.
[0099] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the computer device, connecting all parts of the computer device through various interfaces and lines.
[0100] Memory can be used to store computer programs and / or modules. The processor performs various functions of the computer device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can primarily include a program storage area and a data storage area. The program storage area can store the operating system and at least one application program required for a function (e.g., sound playback, image playback, etc.); the data storage area can store data created based on the use of the mobile phone (e.g., audio data, video data, etc.). Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMedia Cards (SMC), Secure Digital (SD) cards, Flash Cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.
[0101] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, servers, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.
[0102] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), servers, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0103] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0104] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0105] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A position prediction method based on bidirectional inference and double-layer association, characterized in that, The method comprises the following steps: Step S1, acquiring a candidate position ranking matrix of a predicted access position from a user perspective ; Step S2, acquiring a candidate visitor ranking matrix of the visiting user predicted from a place perspective ; Step S3, optimizing the candidate position ranking matrix and the prediction effect of the candidate visitor ranking matrix Step S4, obtaining a new candidate position ordering matrix reordered ; Step S5, obtaining a new candidate visitor ordering matrix reordered ; Step S6, according to and bidirectional inference, and obtain the final position prediction result; The step S3 specifically comprises the following steps: Step S30, obtaining a candidate position ranking matrix by mutual iteration optimization through a shared loss function and obtaining a candidate visitor ranking matrix with a prediction effect; Wherein, the candidate position ranking matrix and the candidate visitor ranking matrix The corresponding element values in the two matrices are subtracted, the total sum of the difference values of all elements is taken as the loss function, the training is ended when the loss function no longer becomes smaller in the iteration process, and the candidate position ranking matrix and the prediction effect of the candidate visitor ranking matrix are obtained by optimization. The step S4 specifically comprises the following steps: Step S41, calculate the similarity between the candidate position ranking matrix of each two users, obtain the user similarity matrix between the candidate position ranking matrix of all users ; Step S42, filtering the user similarity matrix , obtaining the filtered user similarity matrix ; Step S43, reorder the candidate location ranking matrix of the current user according to the filtered user similarity matrix reorder the candidate location ranking matrix of the current user, and obtain a new candidate location ranking matrix ; In the step S42, the user similarity matrix is obtained by setting the elements smaller than the first threshold value to 0. ; The step S5 specifically comprises the following steps: Step S51, calculate the candidate visitor ranking matrix similarity between each two places, obtain the place similarity matrix of all places ; Step S52, filtering the place similarity matrix to obtain a filtered place similarity matrix ; Step S53, reorder the filtered place similarity matrix reorder the candidate visitor ordering matrix of the current user to obtain a new candidate visitor ordering matrix ; In the step S52, the place similarity matrix is obtained by setting the elements smaller than the second threshold value to 0. .
2. The method for location prediction based on bidirectional inference and double-layer association according to claim 1, wherein, The step S1 specifically comprises the following steps: Step S11, taking the access location sequence of each user as input, obtaining an embedding expression vector through embedding layer processing ; Step S12, embedding the expression vector For input, the RNN network is processed to obtain a candidate position ranking list ; Step S13, ranking list of candidate locations for each user , obtaining a candidate location ranking matrix for all users .
3. The method for location prediction based on bidirectional inference and double-layer association according to claim 1, wherein, The step S2 specifically comprises the following steps: Step S21, taking the visitor sequence of each place as input, obtaining an embedding expression vector through embedding layer processing ; Step S22, embedding the expression vector For input, the RNN network is processed to obtain a candidate visitor ranking list ; Step S23, according to the candidate visitor ranking list of each site, get the candidate visitor ranking matrix of all sites .
4. A location prediction system based on bidirectional inference and double-layer association, characterized in that, Comprise: a candidate position prediction obtaining module, configured to obtain a candidate position ranking matrix of the predicted access position from the user perspective ; a candidate visitor prediction obtaining module, configured to obtain a candidate visitor ranking matrix of the predicted visiting user from a location perspective ; An optimization module, in communication with the candidate location prediction obtaining module and the candidate visitor prediction obtaining module, is configured to optimize the candidate location ranking matrix and the candidate visitor ranking matrix. and the prediction effect of the candidate visitor ranking matrix . A new candidate position obtaining module, in communication connection with the optimization module, is configured to obtain a new candidate position ranking matrix reordered ; A new candidate visitor obtaining module, in communication with the optimization module, is configured to obtain a new candidate visitor ranking matrix after reordering ; The final position obtaining module is in communication connection with the new candidate position obtaining module and the new candidate visitor obtaining module, and is used for obtaining a final position prediction result according to and carrying out bidirectional inference. The optimization module is specifically used for: The candidate position ranking matrix is obtained by mutual iteration optimization of shared loss function coefficients and the prediction effect of the candidate visitor ranking matrix is obtained. Wherein, the candidate position ranking matrix and the candidate visitor ranking matrix The corresponding element values in the two matrices are subtracted, the total sum of the difference values of all elements is taken as the loss function, the training is ended when the loss function no longer becomes smaller in the iteration process, and the candidate position ranking matrix and the prediction effect of the candidate visitor ranking matrix are obtained by optimization. The new candidate position acquisition module is specifically used for: computing the similarity between the candidate location ranking matrices of each two users, obtaining a user similarity matrix between the candidate location ranking matrices of all users ; Filtering a user similarity matrix , obtaining the filtered user similarity matrix ; reordering the candidate location ranking matrix for the current user reordering the candidate location ranking matrix for the current user ; wherein the user similarity matrix is obtained by setting to zero the elements of the user similarity matrix that are smaller than a first threshold The new candidate visitor acquisition module is specifically used for: computing a candidate visitor rank matrix similarity between each two locations, obtaining a location similarity matrix for all locations ; filter the location similarity matrix to obtain a filtered location similarity matrix ; reordering the candidate visitor ordering matrix for the current user based on the filtered location similarity matrix reordering the candidate visitor ordering matrix for the current user based on the filtered location similarity matrix ; wherein the place similarity matrix is obtained by setting to 0 the elements of the place similarity matrix which are smaller than a second threshold value .
5. The bidirectional inference and bilayer association based location prediction system of claim 4, wherein, The candidate position prediction acquisition module comprises: The location embedding expression vector acquisition unit is configured to obtain an embedding expression vector by processing a visit location sequence of each user through an embedding layer. ; The candidate position ranking list obtaining unit is connected in communication with the place embedding expression vector obtaining unit, and is configured to obtain the candidate position ranking list by processing the embedding expression vector as input, to obtain the candidate position ranking list by processing the embedding expression vector ; A candidate position ranking matrix acquisition unit, which is communicatively connected with the candidate position ranking list acquisition unit and is configured to acquire a candidate position ranking matrix.