CN116503588A - POI recommendation method, device and equipment based on multi-element relation space-time network - Google Patents

POI recommendation method, device and equipment based on multi-element relation space-time network Download PDF

Info

Publication number
CN116503588A
CN116503588A CN202310465884.7A CN202310465884A CN116503588A CN 116503588 A CN116503588 A CN 116503588A CN 202310465884 A CN202310465884 A CN 202310465884A CN 116503588 A CN116503588 A CN 116503588A
Authority
CN
China
Prior art keywords
poi
space
track
time
pois
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310465884.7A
Other languages
Chinese (zh)
Inventor
高云君
孙琦晨
陈璐
房子荃
吴东恩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202310465884.7A priority Critical patent/CN116503588A/en
Publication of CN116503588A publication Critical patent/CN116503588A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30241—Trajectory

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a POI recommendation method based on a multi-element relation space-time network, which comprises the following steps: s11, acquiring user information in the LBSN, and historical tracks of POIs and users; s12, constructing a plurality of explicit POI relation graphs according to POIs and historical tracks, and adaptively learning implicit relations of the explicit POIs based on a hypergraph architecture to obtain candidate POI embedded vector sets corresponding to all POIs; s13, acquiring a check-in set of the same space-time position from a historical track according to the space-time position of the last check-in of the user, and adding the check-in set into the current track to obtain an enhanced track; s14, constructing a corresponding space-time correlation matrix according to the enhanced track so as to obtain a track embedded vector; s15, predicting according to the track embedded vector and the candidate POI embedded vector set to obtain a POI recommendation result. The invention also provides a POI device and equipment. The method can effectively improve the accuracy of POI recommendation.

Description

POI recommendation method, device and equipment based on multi-element relation space-time network
Technical Field
The invention belongs to the technical field of Internet, and relates to a POI recommendation method, device and equipment based on a multi-element relation space-time network.
Background
With the increasing popularity of Location Based Social Networks (LBSN), such as Foursquare, facebook, the check-in data of users at points of interest (POIs) has increased, resulting in massive amounts of POI track information including user behavior and preferences. Such information may support a variety of location-based personalized services, POI recommendation being one of the important. Its purpose is to predict POIs it is about to visit based on preference information in the user's historical track. The POI recommendation not only can help the user plan travel, but also is beneficial to the POI holder to put advertisements to attract target groups.
In the POI recommendation, not only the sequence mode and the space-time information in the track but also the multi-element relation among different POIs need to be considered, so that designing a POI recommendation method based on the multi-element relation space-time network has become urgent need in academia and industry.
In the process of implementing the present invention, the inventor finds that at least the following problems exist in the prior art:
first, they fail to fully utilize the rich information available in the LBSN, including the multivariate relationship between POIs and personalized movement patterns in the user's trajectory, such that the learned POI and trajectory characterization contain insufficient amounts of information, making the final recommendation inaccurate. Secondly, because of limited computing resources, the existing method only selects the nearest sub-track to compute, and the omission of the long-term history is caused, so that the information about the long-term movement mode of the user is ignored.
Patent document CN115795182a discloses a next POI recommendation method based on a graph roll-up network, in which an initialization embedding layer of user embedding, POI embedding, time embedding, category embedding, and relative position embedding is first generated. Then, based on the graph convolution network, the attention mechanism and the feedforward layer, a convolution layer is built. And finally, carrying out inner product operation through the obtained user characteristics and POI characteristics to obtain the preference of the user to the POI. According to the method, the information of the user is analyzed by constructing the graph convolution network, but the whole historical track is required to be processed in the whole process of the method, and the requirement on the computing capability is high.
Patent document CN115130018A discloses a POI recommendation method and device for distinguishing a exploratory mode and a space-time revisit mode of a user, the method firstly models the preference of the user in the exploratory mode and the preference of the user in the space-time revisit mode, calculates a space-time correlation coefficient by using a time interval and a space interval, and then calculates the scores of all POIs in a POI candidate set in the two modes respectively. And then calculating the transition probabilities of the exploratory mode and the space-time revisit mode, finally calculating the final POI recommendation score by combining the scores of the POIs in the two modes and the transition probabilities of the two modes, and sorting and selecting the final POI recommendation score to recommend the top N POIs with the highest scores. The method needs to analyze the historical data of all users in the area, but does not consider the omission of long-term historical records, so that the information about the long-term movement mode of the users is ignored.
Disclosure of Invention
Aiming at the defects of the prior art, the embodiment of the invention aims to provide a POI recommendation method, device and equipment based on a multi-element relation space-time network, which can fully discover space-time information in LBSN and simultaneously efficiently utilize long-term history records to improve the accuracy of POI recommendation.
According to a first aspect of the object of the embodiment of the present invention, there is provided a POI recommendation method based on a multi-element relationship space-time network, including:
s11, acquiring user information in the LBSN, and historical tracks of the POIs and the user, wherein the historical tracks consist of a plurality of check-ins, and each check-in is a group of triples containing space-time information.
The POIs include names, categories, coordinates, and classifications of points of interest.
S12, constructing a plurality of explicit POI relation graphs according to POIs and historical tracks, and adaptively learning implicit relations of the explicit POIs based on a hypergraph architecture to obtain candidate POI embedded vector sets corresponding to all POIs.
S13, acquiring a check-in set of the same space-time position from the historical track according to the space-time position of the last check-in of the user, and adding the check-in set into the current track to obtain the enhanced track.
S14, constructing a corresponding space-time correlation matrix according to the enhanced track so as to obtain a track embedded vector.
S15, predicting according to the track embedded vector and the candidate POI embedded vector set to obtain a POI recommendation result.
Specifically, in S12, the construction process of the candidate POI embedded vector set is as follows:
s21, initializing a learnable embedded vector for each POI;
s22, judging the distance between the geographic positions in every two POIs, and if the distance is smaller than a geographic distance threshold value, using the distance as a neighbor node to construct a distance relation graph;
s23, based on two continuous check-ins in the history track, constructing a relation graph as neighbor nodes, and replacing the nodes with corresponding POIs according to the time-space information of the check-ins to obtain a conversion relation graph;
and S24, carrying out weighted fusion based on the distance relation diagram and the conversion relation diagram to obtain a candidate POI embedded vector set.
Specifically, in S13, the procedure for obtaining the enhancement track is as follows:
s31, creating a corresponding space-time record table according to the historical track;
s32, mapping corresponding time and geographic information to high-dimensional embedded representations respectively according to the space-time position of the last sign-in of the user;
s33, performing aggregation search in the space-time record table by adopting a KNN clustering method according to the obtained embedded representation so as to obtain similar historical check-in;
and 3-4, combining the historical sign-in with the current track to obtain an enhanced track.
Specifically, in S14, the space-time correlation matrix is constructed and obtained by using a linear interpolation method based on the time and space intervals between check-ins in the enhanced track.
Specifically, in S14, the track embedded vector is obtained by calculation using a self-attention mechanism that adds temporal-spatial information expansion.
Specifically, in S15, the procedure for obtaining the POI recommendation result is as follows:
s51, calculating the obtained track embedded vectors and all candidate POI embedded vectors by adopting a score calculation function, and obtaining probability scores corresponding to the candidate POI embedded vectors;
s52, sequencing the candidate POI embedded vectors according to the probability score to obtain the TOP-K POIs as POI recommendation results and outputting the POIs.
According to a second aspect of the object of the embodiments of the present invention, there is provided a POI recommendation device, which is implemented by the above POI recommendation method based on a multi-element relationship space-time network, including:
and the acquisition module is used for acquiring the historical tracks of the user, the POI and the user in the LBSN.
And the construction module is used for generating a plurality of corresponding explicit POI relation diagrams according to the POIs and the historical tracks.
And the POI embedding module is used for fusing and obtaining a candidate POI embedding vector set based on the generated multiple explicit POI relation diagrams.
And the space-time enhancement module is used for generating an enhancement track and a corresponding space-time incidence matrix.
And the track embedding module is used for learning the track representation by utilizing the space-time correlation matrix and the enhanced track to obtain a track embedding vector.
And the recommendation module is used for calculating and ranking probability scores by utilizing the track embedded vectors and all the candidate POI embedded vectors to obtain a final POI recommendation result.
According to a third aspect of the object of an embodiment of the present invention, there is provided an electronic apparatus including:
one or more processors.
And a memory for storing one or more programs.
The one or more programs, when executed by the one or more processors, cause the one or more processors to implement the POI recommendation method based on the multi-relation spatiotemporal network described above.
Compared with the prior art, the invention has the beneficial effects that:
(1) A data-driven hypergraph structure is constructed to capture implicit relationships between POIs, providing a more accurate representation and a priori knowledge for sequential learning.
(2) A motion perception space-time track embedding module is designed to fully extract the movement modes and preferences of users contained in the track.
(3) And the related history retrieval mode is utilized, the movement modes of the user are enriched by searching related history records, meanwhile, noise interference is reduced, calculation cost is reduced to the maximum extent, and the accuracy of the recommendation result is improved.
Drawings
FIG. 1 is a flow chart illustrating a POI recommendation method based on a multi-relationship spatiotemporal network, according to an exemplary embodiment;
FIG. 2 is a block diagram illustrating a POI recommendation device based on a multi-element spatio-temporal network, according to an exemplary embodiment;
fig. 3 is a schematic diagram of an electronic apparatus shown according to an example embodiment.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present application.
The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in this application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.
It should be understood that although the terms first, second, third, etc. may be used herein to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of check-in from another. For example, a first check-in may also be referred to as a second check-in, and similarly, a second check-in may also be referred to as a first check-in, without departing from the scope of the present application. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.
As shown in fig. 1, the method provided in this embodiment includes the following steps:
s11, acquiring historical tracks of the user, the POI and each user in the LBSN, wherein the historical tracks of the userEach sign-in r is a group of triples containing space-time information;
s12, constructing a plurality of explicit POI relation graphs according to POIs and historical tracks, and adaptively learning implicit relations of the explicit POIs based on a hypergraph architecture to obtain candidate POI embedded vector sets corresponding to all POIs;
s13, acquiring a check-in set of the same space-time position from a historical track according to the space-time position of the last check-in of the user, and adding the check-in set into the current track to obtain an enhanced track;
s14, constructing a corresponding space-time correlation matrix according to the enhanced track so as to obtain a track embedded vector;
s15, predicting according to the track embedded vector and the candidate POI embedded vector set to obtain a POI recommendation result.
As can be seen from the above embodiments, the present application proposes a multi-element relationship spatiotemporal attention network for POI recommendation task. In one aspect, a data-driven hypergraph structure is constructed to capture implicit relationships between POIs, providing more accurate representation and a priori knowledge for sequential learning. On the other hand, a motion perception space-time track embedding module is designed to fully extract the movement modes and preferences of users contained in the track. Finally, the relevant history searching mode is utilized, the moving modes of the user are enriched by searching relevant history records, meanwhile, noise interference is reduced, calculation cost is reduced to the maximum extent, and accuracy of the recommendation result is improved.
In the implementation of step S11: acquiring historical tracks of a user, POI and each user in LBSN, wherein user data is expressed as a user set U= { U 1 ,u 2 ,…,u |U| POI is represented as a set of points of interest l= { L 1 ,l 2 ,…,l |L| A point of interest generally refers to a location of various entities, such as a restaurant, a store. The user history trackConsists of a plurality of check-ins, which indicate that the user u is at the time t i The previous history trace is a time-ordered check-in sequence. Wherein each sign-in r is a set of triples containing spatiotemporal information, each of said triples r= (u, l) i ,t i ) Where u represents a user, l i Is POI with unique geographic longitude and latitude coordinates, t i Representing a discrete time frame. Colloquially, the triplet represents user u at time t i Visit POI l i . After definition of POI, user and representation of user trajectory, input of historical trajectory of user to be predicted +.>Candidate POI set l= { L 1 ,l 2 ,…,l |L| And the current time t i . The POI recommending task is that the user is at t i+1 Recommending a list of POIs, each POI being associated with a score, the higher the score is, the more likely the score isThe higher the probability that the user wants to access the corresponding POI.
In the implementation of S12:
s21, initializing a learnable embedded vector for each POI.
S22, judging the distance between the geographical positions in every two POIs, and if the distance is smaller than a geographical distance threshold value, using the distance as a neighbor node to construct a distance relation graph.
Specifically, a graph G is initialized d ＝(V d ,E d ) Wherein V is d Representing all POI nodes, E d Representing edges with geographical neighbour relations G d The correlation adjacency matrix of which is denoted as A d ∈R N×N Setting a threshold delta d =2 km, if POI l i And POI l j The geographic distance between them is less than delta d Then A d [i,j]=1, otherwise a d [i,j]＝0。
S23, based on two continuous check-ins in the history track, the nodes are used as neighbor nodes to construct a relation graph, and the nodes are replaced by corresponding POIs according to the time-space information of the check-ins, so that a conversion relation graph is obtained.
Specifically, a graph G is initialized t ＝(V t ,E t ) Wherein V is t Representing all POI nodes, E t Representing edges with geographical neighbour relations G t The correlation adjacency matrix of which is denoted as A t ∈R N×N Wherein A is t [i,j]＝freq(l i ,l j )，freq(l i ,l j ) Representing statistics of POI l in all trajectories i And POI l j The switching frequency between.
And S24, carrying out weighted fusion based on the distance relation diagram and the conversion relation diagram to obtain a candidate POI embedded vector set.
Specifically, first, for two kinds of explicit relation diagrams G which have been generated t And G d First, each node is mapped by using the initialized embedded vector in the S21, and the POI node is converted into a dense vector representationAt the same timeThe M dimension is split into K parts, denoted +.>A vector representation representing the K-th part. Next, using a method of graph convolution based on decoupling learning, at G t And G d The graph-embedding learning is performed separately, and in particular, the embedding process can be expressed asWherein D is a degree matrix, I is an identity matrix,represents a normalized adjacency matrix, Θ k Representing a weight matrix for calculating the latitude of the K part in graph convolution, and finally obtaining two different POI embeddings and embedding H based on geographic distance d And embedding H based on conversion relation t . To obtain the implicit multivariate graph, we first maintain a low latitude transformation matrix +.>H embedded by using the two POIs d And H t Adjacent matrix A for guided generation of hypergraph h ＝(H·W h ) T W here h Representing the weight matrix in hypergraph learning.
The information propagation process on the hypergraph is divided into two steps, aggregation and propagation. Firstly, aggregating attributes of different POIs according to influence weights of the different POIs in each hyperedge by utilizing an aggregation function to generate a representation of the hyperedge, so as to capture internal association in each implicit relation, wherein the internal association can be specifically expressed as the following functions:next, H (*) Represents H d And H t In order to obtain the embedding update of the POI under the influence of different polynary relations, the influence caused by different relations is fed back to the participated POI nodes to obtain POsI embedding->Finally, to obtain POI embeddings under the influence of different relationships, we fuse the POI embeddings obtained under the different relationship graphs by means of weighted addition to obtain the final POI embeddings X, specifically, the process expressed as x=w 3 [w 1 H d +(1-w 1 )H t ]+(1-w 3 )[w 2 H ′ d +(1-w 2 )H t ′ ]W is herein 1 ，w 2 ，w 3 Representing three weighting parameters.
In the implementation of S13:
s31: and creating a space-time record table according to the complete historical track of the user.
Specifically, for complete recording of the spatiotemporal information of all tracks, a spatiotemporal table (STT) of user history tracks is created, with which the movement patterns of the user are enriched by filling in history check-in records, each record being stored as a quintuple:the five-tuple represents user u i POI i at position d is visited at time t, where i is the id of the POI, d= (lon, lat) represents a geographic latitude and longitude position, j represents this is the jth visit in his entire track, and t is a timestamp. To prevent similar but not identical check-in records from losing information, we do not save the check-in records in their original form. Instead, we use the POI and track embedding described above to transform the l, t, d projections into the embedding space, resulting inFinally, a space-time table is generated by using the mapped recordHere->Is->Abbreviations representing users u i Is a sign-in record after projection conversion.
S32: mapping time and position information of the user to high-dimensional embedded representations respectively according to the current space-time position of the user;
specifically, we approximate the current state of the user using his/her most recent check-in record. In this way, the STT allows us to search for user history check-in records that are similar to the user's current state.
S33, performing aggregation search in the space-time record table by using a KNN clustering method according to the obtained embedded representation of the time space to obtain related historical sign-in records;
specifically, a recent history track of a user is givenWe will check in nearest->Mapping into vector space to initialize a query q n ＝(E n ,E t (t n ),E d (d n ) Using this query to search in STT to obtain user u i Is a similar history of the set of check-ins Mem (u i )＝Search(STT(u i ),q n ) Where the Search function can be replaced using different clustering algorithms, e.g. DBSCAN and KNN, here we use multi-condition KNN, where E n And E is d (d n ) As two conditions, a threshold β is set, and if the similarity between the history and the most recent record is higher than β, we add the history to the temporary set. After clustering we selected top-k similar records from the temporary set as the result, finally obtaining +.>
S34, combining the related historical sign-in record with the current track to obtain the track with the enhanced memory.
Specifically, we combine the obtained correlation history check-ins into the original sequence to compensate for long period pattern loss, the process is expressed as Finally obtain user u i Trajectory H after memory enhancement ′ (u i )。
In the implementation of S14:
s41, calculating time and space intervals among different check-ins according to the geographic information and the time information of each check-in the enhanced track.
Specifically, an enhanced trajectory is inputFirst, each sign-in is mapped to a vector space by the final POI-embedded X obtained in S13 to obtain +.>Wherein->Representative sign-in->The mapped vector representation.
The relative time interval for each check-in the track is then calculated,representing the relative time length between the ith and jth accesses in the track, wherein +.>Representing the track S u Except for 0).
On the other hand, the relative space interval between longitude and latitude between each check-in is calculated respectively by using haverine algorithm,representing the relative distance of the spatial separation between li and lj in the longitude coordinate transformation direction; />The calculation method is similar, and the relative distance in the coordinate conversion direction of the latitudide is represented, wherein +.>Represent S u Minimum space interval in the longitude and latitude direction (except 0).
S42, constructing a space-time correlation matrix based on a linear interpolation method by utilizing the time and space intervals.
Specifically, using the above generation t i,j Andthree space-time interval matrixes are formed, and a time threshold delta is set t And two spatial thresholds>And shearing the three space-time interval matrixes by using the three thresholds to prevent the convergence from being influenced by the excessive interval. Next, a learnable time-embedded dictionary is maintained>And two learnable spatially embedded dictionaries +.>And->And mapping the three sheared space-time matrixes, so that sparse discrete interval values are converted into a dense continuous low-dimensional space, the sparse representation problem is relieved, and the space interval matrixes under two coordinates are added to finally obtain a space interval matrix and a time interval matrix which can express any direction.
S43, inputting the space-time correlation matrix and the enhanced track sequence into a track embedding module, and calculating to obtain track embedding by using a self-attention mechanism added with space-time information expansion.
In particular, self-attention aggregated spatiotemporal information is utilized to update representations of each POI in a trajectory based on the obtained spatial and temporal interval matrices and the enhanced trajectory sequence Wherein W is V Representing model parameter weight matrix, p j Position embedding representing jth position to capture interactions between POI accesses and obtain spatiotemporal dependencies between user preference accesses taking into account different relationships between POIs and POIs, each +.> Representing impact weights between different POIs, whereAnd->Respectively representing the elements in the time and space interval matrixes constructed as above, M represents the dimension of the element, W Q And W is K Two model weight matrices.
In the implementation of S15Inputting the candidate POI embedded vector and the obtained track vector, and calculating the association degree between the track vector and different candidate POI embedded vectors Wherein x is t ′ Vector representation X representing POI accessed at time t l And finally ranking according to the obtained association degree, and obtaining a final POI recommendation list from large to small according to the association degree.
The embodiment provides a POI recommending device, which is realized by the POI recommending method provided by the embodiment.
As shown in fig. 2, the following modules are included:
the acquisition module is used for: for obtaining a user, POI data and a historical track of each user in an LBSN (location based social network), the track comprising a number of check-ins, wherein each check-in consists of a user, a POI and a time;
the construction module comprises: the method comprises the steps of constructing a plurality of explicit POI relation diagrams according to POI data and sign-in information of historical tracks, and designing a hypergraph architecture to adaptively learn implicit POI multivariate relation diagrams;
POI embedding module: the method is used for aggregating information in different types of graphs, and final POI embedding is obtained by means of weighted fusion by using two different graph message transmission methods.
A space-time enhancement module: the method is used for searching relevant historic record enhancement tracks by utilizing the existing space-time positions of the users and constructing a space-time correlation matrix.
Track embedding module: the method is used for learning the track characterization by utilizing the space-time correlation matrix and the enhanced track to obtain the track embedded vector.
And a recommendation module: and the method is used for calculating and ranking probability scores by utilizing the track embedded vector and all candidate POI embedded vectors to obtain a final POI recommendation result.
For the device embodiments, reference is made to the description of the method embodiments for the relevant points, since they essentially correspond to the method embodiments. The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purposes of the present application. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
Correspondingly, the application also provides electronic equipment, which comprises: one or more processors; a memory for storing one or more programs; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the POI recommender method based on a multi-relational spatiotemporal network as described above.
As shown in fig. 3, a hardware structure diagram of any device with data processing capability, where the POI recommendation method based on the multiple-relationship space-time network is provided in this embodiment, except for the processor, the memory and the network interface shown in fig. 3, any device with data processing capability in the embodiment generally includes other hardware according to the actual function of the any device with data processing capability, which is not described herein again.
Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains.
It is to be understood that the present application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof.

Claims (8)

1. A POI recommendation method based on a multi-element relationship space-time network, comprising:
s11, acquiring user information in the LBSN, and historical tracks of POIs and users, wherein the historical tracks consist of a plurality of check-ins, and each check-in is a group of triples containing space-time information;
s12, constructing a plurality of explicit POI relation graphs according to POIs and historical tracks, and adaptively learning implicit relations of the explicit POIs based on a hypergraph architecture to obtain candidate POI embedded vector sets corresponding to all POIs;
s13, acquiring a check-in set of the same space-time position from a historical track according to the space-time position of the last check-in of the user, and adding the check-in set into the current track to obtain an enhanced track;
s14, constructing a corresponding space-time correlation matrix according to the enhanced track so as to obtain a track embedded vector;
s15, predicting according to the track embedded vector and the candidate POI embedded vector set to obtain a POI recommendation result.
2. The POI recommendation method based on the multi-relation space-time network according to claim 1, wherein in S12, the construction process of the candidate POI embedding vector set is as follows:
s21, initializing a learnable embedded vector for each POI;
s22, judging the distance between the geographic positions in every two POIs, and if the distance is smaller than a geographic distance threshold value, using the distance as a neighbor node to construct a distance relation graph;
s23, based on two continuous check-ins in the history track, constructing a relation graph as neighbor nodes, and replacing the nodes with corresponding POIs according to the time-space information of the check-ins to obtain a conversion relation graph;
and S24, carrying out weighted fusion based on the distance relation diagram and the conversion relation diagram to obtain a candidate POI embedded vector set.
3. The POI recommendation method based on the multivariate relation space-time network according to claim 1, wherein in S13, the enhancement track is obtained as follows:
s31, creating a corresponding space-time record table according to the historical track;
s32, mapping corresponding time and geographic information to high-dimensional embedded representations respectively according to the space-time position of the last sign-in of the user;
s33, performing aggregation search in the space-time record table by adopting a KNN clustering method according to the obtained embedded representation so as to obtain similar historical check-in;
s34, combining the historical sign-in with the current track to obtain an enhanced track.
4. The POI recommendation method based on the multivariate relation space-time network according to claim 1, wherein in S14, the space-time correlation matrix is constructed and obtained by adopting a linear interpolation method based on time and space intervals between check-ins in the enhanced track.
5. The POI recommendation method based on a multi-element relationship space-time network according to claim 1, wherein in S14, the trajectory embedding vector is obtained by calculation using a self-attention mechanism adding a space-time information expansion.
6. The POI recommendation method based on the multi-relation space-time network according to claim 1, wherein in S15, the POI recommendation result is obtained as follows:
s51, calculating the obtained track embedded vectors and all candidate POI embedded vectors by adopting a score calculation function, and obtaining probability scores corresponding to the candidate POI embedded vectors;
s52, sequencing the candidate POI embedded vectors according to the probability score to obtain the TOP-K POIs as POI recommendation results and outputting the POIs.
7. A POI recommending apparatus, characterized in that it is realized by the POI recommending method based on the multi-element relation space-time network according to any one of claims 1 to 6, comprising:
the acquisition module is used for acquiring historical tracks of the user, the POI and the user in the LBSN;
the construction module is used for generating a plurality of corresponding explicit POI relation diagrams according to the POIs and the historical tracks;
the POI embedding module is used for obtaining a candidate POI embedding vector set in a fusion mode based on the generated multiple explicit POI relation diagrams;
the space-time enhancement module is used for generating an enhancement track and a corresponding space-time incidence matrix;
the track embedding module is used for learning track characterization by utilizing the space-time correlation matrix and the enhanced track to obtain a track embedding vector;
and the recommendation module is used for calculating and ranking probability scores by utilizing the track embedded vectors and all the candidate POI embedded vectors to obtain a final POI recommendation result.
8. An electronic device, comprising:
one or more processors;
a memory for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the POI recommendation method based on a multi-relation spatiotemporal network of any of claims 1 to 6.
CN202310465884.7A 2023-04-24 2023-04-24 POI recommendation method, device and equipment based on multi-element relation space-time network Pending CN116503588A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310465884.7A CN116503588A (en) 2023-04-24 2023-04-24 POI recommendation method, device and equipment based on multi-element relation space-time network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310465884.7A CN116503588A (en) 2023-04-24 2023-04-24 POI recommendation method, device and equipment based on multi-element relation space-time network

Publications (1)

Publication Number Publication Date
CN116503588A true CN116503588A (en) 2023-07-28

Family

ID=87317796

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310465884.7A Pending CN116503588A (en) 2023-04-24 2023-04-24 POI recommendation method, device and equipment based on multi-element relation space-time network

Country Status (1)

Country Link
CN (1) CN116503588A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117591751A (en) * 2024-01-19 2024-02-23 国网湖北省电力有限公司信息通信公司 Picture embedding-based up-down Wen Zhongcheng-degree fusion interest point recommendation method and system

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117591751A (en) * 2024-01-19 2024-02-23 国网湖北省电力有限公司信息通信公司 Picture embedding-based up-down Wen Zhongcheng-degree fusion interest point recommendation method and system
CN117591751B (en) * 2024-01-19 2024-04-26 国网湖北省电力有限公司信息通信公司 Picture embedding-based interest point recommendation method and system based on upper-lower Wen Zhongcheng-degree fusion

Similar Documents

Publication Publication Date Title
Yin et al. Joint modeling of user check-in behaviors for real-time point-of-interest recommendation
Feng et al. Poi2vec: Geographical latent representation for predicting future visitors
Chen et al. Curriculum meta-learning for next POI recommendation
CN110928993B (en) User position prediction method and system based on deep cyclic neural network
Salehi et al. Hybrid recommendation approach for learning material based on sequential pattern of the accessed material and the learner’s preference tree
CN110119475B (en) POI recommendation method and system
CN109460520B (en) Interest point recommendation method based on geographic-social relationship and deep implicit interest mining
CN110334293B (en) Position social network-oriented position recommendation method with time perception based on fuzzy clustering
CN111008332A (en) Content item recommendation method, device, server and storage medium
CN110309434B (en) Track data processing method and device and related equipment
Vellaichamy et al. Hybrid Collaborative Movie Recommender System Using Clustering and Bat Optimization.
CN113158038B (en) Interest point recommendation method and system based on STA-TCN neural network framework
CN116244513B (en) Random group POI recommendation method, system, equipment and storage medium
CN112199609A (en) POI recommendation system and method for sensing space-time semantic interval under self-attention
CN111475744B (en) Personalized position recommendation method based on ensemble learning
CN111695046A (en) User portrait inference method and device based on spatio-temporal mobile data representation learning
CN113536105A (en) Recommendation model training method and device
CN116503588A (en) POI recommendation method, device and equipment based on multi-element relation space-time network
US20210239479A1 (en) Predicted Destination by User Behavior Learning
US20100161380A1 (en) Rating-based interests in computing environments and systems
CN107909498B (en) Recommendation method based on area below maximized receiver operation characteristic curve
Li et al. Indoor mobility semantics annotation using coupled conditional Markov networks
CN117435819A (en) Method, system and storage medium for recommending interest points through space-time dynamic perception
Oh et al. Explorative navigation of mobile sensor networks using sparse Gaussian processes
CN113935251B (en) User behavior prediction model generation method and device and user behavior prediction method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination