CN111177565B - Interest point recommendation method based on correlation matrix and word vector model - Google Patents

Interest point recommendation method based on correlation matrix and word vector model Download PDF

Info

Publication number
CN111177565B
CN111177565B CN201911425220.8A CN201911425220A CN111177565B CN 111177565 B CN111177565 B CN 111177565B CN 201911425220 A CN201911425220 A CN 201911425220A CN 111177565 B CN111177565 B CN 111177565B
Authority
CN
China
Prior art keywords
interest
user
point
points
interest point
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.)
Active
Application number
CN201911425220.8A
Other languages
Chinese (zh)
Other versions
CN111177565A (en
Inventor
俞东进
完颜文博
王东京
张新
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.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
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 Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN201911425220.8A priority Critical patent/CN111177565B/en
Publication of CN111177565A publication Critical patent/CN111177565A/en
Application granted granted Critical
Publication of CN111177565B publication Critical patent/CN111177565B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses an interest point recommendation method based on a correlation matrix and a word vector model. The invention carries out vector representation modeling of the user and the interest points by referring to a training word vector model in natural language processing, and the model can capture the geographical influence of the interest points from the sign-on sequence of the user. Suppose that one point of interest is affected by other points of interest in the user's check-in sequence. Consider each point of interest as a "word," consider each user's sequential check-in sequence as a "sentence," and train the points of interest and the user's potential representation vectors using the Skip-gram word vector model in self-speech processing in conjunction. The invention fuses the user and the interest point into a better geographic information model respectively. To this end, the geographic distance between the user and the point of interest and the check-in frequency of the user on neighboring points of interest are used in the model. By considering the influence of adjacent interest points in the model recommendation strategy, the check-in data sparsity problem is solved.

Description

Interest point recommendation method based on correlation matrix and word vector model
Technical Field
The invention belongs to the technical field of data mining and recommendation systems, and particularly relates to an interest point recommendation method based on a correlation matrix and a word vector model.
Background
With the popularity of smartphones and other mobile devices, Location Based Social Networks (LBSNs) have become very popular. More and more mobile phone users record their own positions and upload the positions to social media application software (such as WeChat, Gowalla, Foursqare and the like), which generates massive user position check-in data which contains some preferences, behavior characteristics and activity rules of some users. The interest point recommendation is to recommend the interest points to be visited by the user later by analyzing the historical visit records of the user, so that the location-based services (such as recommending playing points, restaurants and the like) can be recommended for the user in a personalized manner, the use experience of the user can be improved, a merchant can be assisted to do targeted services by analyzing the preference information of the user, and the marketing effect and the market benefit of the merchant can be improved. Therefore, analyzing and recommending check-in data for users has been under considerable research.
The existing interest point recommendation method mainly comprises two algorithms, including a memory-based collaborative filtering recommendation algorithm and a model-based collaborative filtering recommendation algorithm. A memory-based algorithm uses the user's check-in data in point of interest recommendations to predict the user's preferences. One of the most important issues with these methods is data sparseness when a user checks-in when a large number of elements in the data are empty (i.e., they do not provide any information). On the other hand, some model-based methods, such as matrix decomposition, can be used to improve the accuracy and scalability of point of interest recommendation. However, since there are many points of interest available and a single user can only access a few of them, the collaborative filtering based method often has sparse data. This has the result that the collaborative filtered user-interest point matrix becomes very sparse, resulting in a reduced recommendation without significant correlation between the user and the interest point. Meanwhile, with the development of deep learning and natural language processing technologies, related research achievements begin to show unique advantages of the traditional recommendation method at first, and some implicit data features which cannot be captured by the traditional machine learning method can be mined through the deep learning, so that a research object can be more comprehensively modeled, and the combination of the traditional machine learning algorithm and the deep learning algorithm becomes a new research direction.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an interest point recommendation method based on a correlation matrix and a word vector model.
The invention carries out vector representation modeling of the user and the interest points by referring to a training word vector model in natural language processing, and the model can capture the geographical influence of the interest points from the sign-on sequence of the user. Suppose that one point of interest is affected by other points of interest in the user's check-in sequence. Consider each point of interest as a "word," consider each user's sequential check-in sequence as a "sentence," and train the points of interest and the user's potential representation vectors using the Skip-gram word vector model in self-speech processing in conjunction.
The method comprises the following specific steps:
step (1) inputting user sign-in data and interest point longitude and latitude data, wherein the sign-in data comprises user ID, interest point ID and sign-in time; the longitude and latitude data of the interest points comprise the ID, longitude and latitude of the interest points;
reading the check-in data, counting the access times of the user to each interest point, and constructing a user-interest point access frequency matrix F;
reading the longitude and latitude data of the interest point, and constructing an interest point-longitude and latitude dictionary L in the form of { interest point ID (longitude, latitude) };
calculating the most frequently visited interest point HAL of each useru＝argmax(fu) Wherein f isuRepresenting the frequency of accessing each interest point by the user u, and constructing a most frequently accessed interest point dictionary HAL of the user in the form of { user ID: interest point ID };
setting distance thresholds alpha and gamma, wherein alpha represents the radius of a high-activity area of a user, gamma represents the radius of a neighbor of an interest point, initializing a user-interest point correlation matrix M, and assigning each value of the matrix as 0;
step (4), constructing a user-interest point correlation matrix M: for each user u, calculating the cumulative sum Σ f of the frequencies of all the points of interest visited by the useru(ii) a Traversing all the interest points p, if the interest points p are not visited by the user u, obtaining the interest point dictionary HAL which is visited by the user most frequently by inquiring the longitude and latitude dictionary L of the interest pointsuThe longitude and latitude of the point of interest p, and based on this, the most frequently visited point of interest HAL is calculateduAnd the distance between the point of interest p and the point of interest p, if the distance is greater than alpha, continuously traversing the next point of interest; if the distance is equal to alpha, initializing the access frequency of the user u around the point of interest p
Figure BDA0002353362480000021
Traversing all interest points p visited by user uuIf it is likeInterest points p and puIf the distance between the two is less than the threshold value gamma, the access frequency F [ u, p ] of the user u to the interest point p is inquired through the access frequency matrix Fu]And is accumulated to
Figure BDA0002353362480000022
Otherwise, traversing the next interest point; after the traversal of the user u accesses all the interest points, calculating the correlation coefficient between the user u and the interest point p
Figure BDA0002353362480000023
After the traversal is completed, the construction of a user-interest point correlation matrix M is completed; f. ofuRepresenting the access frequency of the query user u to all the interest points;
step (5) creating a user access sequence file, wherein each behavior is an interest point access sequence of a certain user, the ID of the user is arranged at the head of the line, all the accessed interest point IDs of the user are arranged behind the ID of the user according to the access sequence, and the user ID and the interest point ID are separated by using a blank space;
and (6) constructing a word vector training model, simulating a Skip-gram model for training word vectors in natural language processing, taking the interest point access sequence of each user as a sentence, taking each interest point in the interest point access sequence of each user as a word, and simultaneously combining user ID information to obtain a target function for model training:
Figure BDA0002353362480000031
where U represents a set of users, siFor user uiAccessing interest point sequences, wherein S represents a set of all user accessing interest point sequences, and c represents the size of a context window;
probability function
Figure BDA0002353362480000032
Wherein
Figure BDA0002353362480000033
Figure BDA0002353362480000034
As a fusion function, viRepresenting user uiInput vector, | P | represents the total number of points of interest, | P |, andjrepresenting a point of interest pjInput vector of l'jRepresenting a point of interest pjOutputting the vector;
and (7) carrying out model training: firstly, setting model parameters, and respectively setting the data size batch _ size of each batch of training, the dimension embedding _ dim of a trained user vector and an interest point vector, the learning rate learning _ rate, the negative sample sampling number num _ samples, the maximum iteration round number epochs and the interest point context window size context _ size; secondly, training by adopting a negative sampling method, and defining a loss function as follows:
Figure BDA0002353362480000035
wherein P isnegWhich represents a set of m negative examples,
Figure BDA0002353362480000036
the vector is input for the negative samples and,
Figure BDA0002353362480000037
outputting a vector for the negative sample, wherein sigma (-) is a sigmoid function, and then starting iterative training;
step (8) according to the user-interest point correlation matrix M obtained by training in step (4) and the user u output by the word vector modeliIs input vector viPoint of interest pjInput vector l ofjThereby calculating the Preference of the user to each interest point and outputting a recommendation result, wherein the calculation formula is referenceup＝(vi·lj)×M[ui,pj]。
The invention has the beneficial effects that the invention is mainly divided into two parts:
the first is to construct a user-interest point correlation matrix, which considers the geographic environment from the perspective of the user and the interest point, respectively, to provide an effective geographic model. For this reason, the active region of the user is considered as the user perspective of the model. To model the point of interest perspective, it is assumed that the more sign-ins around an unvisited POI, the less relevance the POI should be recommended. And constructing a correlation matrix by a calculation method of the correlation between the user and the interest points.
Secondly, the POIs visited by each user, especially the consecutive points of interest in the user check-in data, are affected by each other geographically. In order to capture the geographic influence among the interest points, modeling is carried out on the user and the interest points by referring to a Skip-gram word vector model in a natural language processing technology, and word vectors of the user and the interest points are respectively trained for interest point recommendation.
The invention fuses the user and the interest point into a better geographic information model respectively. To this end, the geographic distance between the user and the point of interest (from the user's perspective) and the check-in frequency of the user on neighboring points of interest (from the point of interest's perspective) are used in the model. By considering the influence of adjacent interest points in the model recommendation strategy, the check-in data sparsity problem is solved.
Experiments of a real data set show that the interest point recommendation method based on the correlation matrix and the word vector model has better recommendation capability compared with the traditional model, and exceeds the traditional interest point recommendation algorithm in multiple indexes (Precision, Recall and the like).
Drawings
FIG. 1 is a diagram of a word vector model.
Detailed Description
The interest point recommendation method based on the correlation matrix and the word vector model provided by the invention is specifically described as follows:
step (1) inputting user sign-in data and interest point longitude and latitude data, wherein the sign-in data comprises user ID, interest point ID and sign-in time; the longitude and latitude data of the interest points comprise ID, longitude and latitude reading data of the interest points, and a user-interest point access frequency matrix F, a longitude and latitude dictionary L of the interest points and a user high activity area dictionary HAL are constructed;
reading the check-in data, counting the access times of the user to each interest point, and constructing a user-interest point access frequency matrix F; reading the longitude and latitude data of the interest point, and constructing an interest point-longitude and latitude dictionary L in the form of { interest point ID (longitude, latitude) }; constructing a user most frequent access interest point dictionary HAL by calculating the most frequent access interest points of each user, wherein the form is { user ID: interest point ID };
the user's activity area is represented using the points of interest most frequently visited by the user:
HALu＝argmax(fu)
wherein:
·furepresenting the frequency with which user u visits the point of interest.
Step (3) setting model parameters, and calculating a correlation matrix M of each user to each interest point;
the correlation calculation formula of the user and the interest points is as follows:
Figure BDA0002353362480000041
wherein:
·
Figure BDA0002353362480000051
representing a user u and a point of interest piThe correlation of (c);
·
Figure BDA0002353362480000052
representing the frequency with which the user u visits neighboring points of interest of the point of interest p;
the correlation matrix M is calculated as follows:
1) setting distance thresholds alpha and gamma;
2) initializing a correlation matrix M and assigning an initial value to be 0;
3) for each user u, the cumulative sum Σ f of the frequencies of all the points of interest visited by the user is calculatedu；
4) Traversing all interest points p;
5) if the point of interest p is not visited by the user u, the point of interest HAL visited by the user most frequently is searched respectively through the latitude and longitude dictionary L of the point of interestuThe longitude and latitude information of the interest point p is calculated, the distance between the interest point p and the longitude and latitude information is calculated according to the longitude and latitude information, if the distance is smaller than alpha, the next step is carried out, and if not, the next interest point is traversed;
6) initialization
Figure BDA0002353362480000053
Traversing all interest points p visited by user uu；
7) If points of interest p and puThe geographic distance therebetween is less than the threshold value gamma, then
Figure BDA0002353362480000054
Figure BDA0002353362480000055
8) After all the interest points are visited by the user u after the traversal is completed, calculating the correlation coefficient between the user u and the interest point p
Figure BDA0002353362480000056
The user-interest point correlation matrix geographic model can capture the geographic influence of the user and the interest point angle respectively. From the user's perspective, the geographic information may be modeled by considering the user's active area. On the other hand, from the point of interest perspective, geographic information may be modeled as check-in frequency of neighboring points of interest to the selected point of interest, and thus may indicate user preferences of one point of interest relative to its neighboring points of interest.
And (4) creating a user access sequence file, wherein each behavior is an interest point access sequence of a certain user, the ID of the user is arranged at the head of the line, all the accessed interest point IDs of the user are arranged behind the ID of the user according to the access sequence, and the user ID and the interest point ID are separated by using a blank space. We mark the kth interest point in the sequence of user u as
Figure BDA0002353362480000057
Each line in the file is as follows:
Figure BDA0002353362480000058
and (5) constructing a word vector training model.
FIG. 1 illustrates a user-point of interest word vector training model. For the convenience of discussion, the definitions of the relevant symbols appearing in steps (4), (5) and (6) are defined uniformly:
u represents a set of users, UiRepresenting a user;
p represents a set of points of interest, PiRepresenting points of interest;
| P | represents the number of points of interest in the data;
·virepresentative user uiInputting a vector;
·ljrepresenting a point of interest pjInput vector of l'jRepresenting a point of interest pjOutputting the vector;
s represents the set of all users visiting the sequence of points of interest, SiRepresenting user uiA sequence of points of interest accessed;
c represents the contextual window size of the point of interest;
·Pnegrepresenting m negative sample sets;
σ (·) is a sigmoid function;
·
Figure BDA0002353362480000061
is a fusion function.
The model uses the access sequence of each user as a sentence, each interest point in each sequence is a word, and simultaneously combines the ID information of the user to obtain the target function of model training as follows:
Figure BDA0002353362480000062
wherein the probability function P (P)j+k|pi,ui) In order to realize the purpose,
Figure BDA0002353362480000063
Figure BDA0002353362480000064
and (6) training a word vector model.
1) Setting model parameters:
batch _ size ═ 64, representing the amount of data per batch of training;
imbedding _ dim ═ 30, the dimensionality of the trained user vector and the point of interest vector;
learning _ rate is 0.001, learning rate;
num _ samples ═ 16, number of samples sampled negatively;
epochs 2000, maximum number of iterations;
context _ size ═ 2, the point of interest context window size.
2) The loss function is set to be:
Figure BDA0002353362480000065
wherein P isneg represents a set of m negative samples,
Figure BDA0002353362480000066
the vector is input for the negative samples and,
Figure BDA0002353362480000067
outputting a vector for the negative sample, wherein sigma (-) is a sigmoid function, and then starting iterative training;
3) and performing iterative training.
Step (7) according to the user-interest point correlation matrix M obtained by training in step (4) and the user vector v output by the word vector modeliInterest point vector ljTherefore, the preference of the user for each interest point is calculated, and a recommendation result is output, wherein the calculation formula is as follows:
Preferenceup＝(vi·lj)×M[ui,pj]。

Claims (1)

1. an interest point recommendation method based on a correlation matrix and a word vector model is characterized by comprising the following specific steps:
step (1) inputting user sign-in data and interest point longitude and latitude data, wherein the sign-in data comprises user ID, interest point ID and sign-in time; the longitude and latitude data of the interest points comprise the ID, longitude and latitude of the interest points;
reading the check-in data, counting the access times of the user to each interest point, and constructing a user-interest point access frequency matrix F;
reading the longitude and latitude data of the interest points, and constructing an interest point-longitude and latitude dictionary L in the form of { interest point ID: (longitude, latitude) };
calculating the most frequently visited interest point HAL of each useru＝argmax(fu) Wherein f isuRepresenting the frequency of accessing each point of interest by the user u, and constructing a user most frequently accessing point of interest dictionary HAL in the form of { user ID: point of interest ID };
setting distance thresholds alpha and gamma, wherein alpha represents the radius of a high-activity area of a user, gamma represents the radius of a neighbor of an interest point, initializing a user-interest point correlation matrix M, and assigning each value of the matrix as 0;
step (4), constructing a user-interest point correlation matrix M: for each user u, the cumulative sum sigma f of the frequencies of all the points of interest visited by the user is calculatedu(ii) a Traversing all the interest points p, if the interest points p are not visited by the user u, obtaining the interest point dictionary HAL which is visited by the user most frequently by inquiring the longitude and latitude dictionary L of the interest pointsuThe longitude and latitude of the point of interest p, and based on this, the most frequently visited point of interest HAL is calculateduAnd the distance between the point of interest p and the point of interest p, if the distance is greater than alpha, continuously traversing the next point of interest; if the distance is equal to alpha, initializing the access frequency of the user u around the point of interest p
Figure FDA0003166743530000011
Traversing all interest points p visited by user uuIf points of interest p and puIf the distance between the two is less than the threshold value gamma, the access frequency F [ u, p ] of the user u to the interest point p is inquired through the access frequency matrix Fu]And is accumulated to
Figure FDA0003166743530000012
Otherwise, traversing the next interest point; after the traversal of the user u accesses all the interest points, calculating the correlation coefficient between the user u and the interest point p
Figure FDA0003166743530000013
After the traversal is completed, the construction of a user-interest point correlation matrix M is completed; f. ofuRepresenting the access frequency of the query user u to all the interest points;
step (5) creating a user access sequence file, wherein each behavior is an interest point access sequence of a certain user, the ID of the user is arranged at the head of the line, all the accessed interest point IDs of the user are arranged behind the ID of the user according to the access sequence, and the user ID and the interest point ID are separated by using a blank space;
and (6) constructing a word vector training model, simulating a Skip-gram model for training word vectors in natural language processing, taking the interest point access sequence of each user as a sentence, taking each interest point in the interest point access sequence of each user as a word, and simultaneously combining user ID information to obtain a target function for model training:
Figure FDA0003166743530000021
where U represents a set of users, siFor user uiAccessing interest point sequences, wherein S represents a set of all user accessing interest point sequences, and c represents the size of a context window;
probability function
Figure FDA0003166743530000022
Wherein
Figure FDA0003166743530000023
Figure FDA0003166743530000024
As a fusion function, viRepresenting user uiInput vector, | P | represents the total number of points of interest, | P |, andjrepresenting a point of interest pjInput vector of l'jRepresenting a point of interest pjOutputting the vector;
and (7) carrying out model training: firstly, setting model parameters, and respectively setting the data size batch _ size of each batch of training, the dimension embedding _ dim of a trained user vector and an interest point vector, the learning rate learning _ rate, the negative sample sampling number num _ samples, the maximum iteration round number epochs and the interest point context window size context _ size; secondly, training by adopting a negative sampling method, and defining a loss function as follows:
Figure FDA0003166743530000025
wherein P isnegWhich represents a set of m negative examples,
Figure FDA0003166743530000026
the vector is input for the negative samples and,
Figure FDA0003166743530000027
outputting a vector for the negative sample, wherein sigma (-) is a sigmoid function, and then starting iterative training;
step (8) according to the user-interest point correlation matrix M obtained by training in step (4) and the word directionUser u of quantity model outputiIs input vector viPoint of interest pjInput vector l ofjThereby calculating the Preference of the user to each interest point and outputting a recommendation result, wherein the calculation formula is referenceup＝(vi·lj)×M[ui，pj]。
CN201911425220.8A 2019-12-31 2019-12-31 Interest point recommendation method based on correlation matrix and word vector model Active CN111177565B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911425220.8A CN111177565B (en) 2019-12-31 2019-12-31 Interest point recommendation method based on correlation matrix and word vector model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911425220.8A CN111177565B (en) 2019-12-31 2019-12-31 Interest point recommendation method based on correlation matrix and word vector model

Publications (2)

Publication Number Publication Date
CN111177565A CN111177565A (en) 2020-05-19
CN111177565B true CN111177565B (en) 2021-08-31

Family

ID=70656083

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911425220.8A Active CN111177565B (en) 2019-12-31 2019-12-31 Interest point recommendation method based on correlation matrix and word vector model

Country Status (1)

Country Link
CN (1) CN111177565B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111782955A (en) * 2020-07-01 2020-10-16 支付宝(杭州)信息技术有限公司 Interest point representing and pushing method and device, electronic equipment and storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105653637A (en) * 2015-12-28 2016-06-08 苏州大学 Interest point recommendation method based on hierarchical structure
CN106126615A (en) * 2016-06-22 2016-11-16 苏州大学 The method and system that a kind of point of interest is recommended
CN108874877A (en) * 2018-05-03 2018-11-23 浙江大学城市学院 A kind of point of interest recommended method based on figure link analysis of facing position social networks
CN109190053A (en) * 2018-07-04 2019-01-11 南京邮电大学 One kind being based on point of interest importance and the authoritative point of interest recommended method of user

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105653637A (en) * 2015-12-28 2016-06-08 苏州大学 Interest point recommendation method based on hierarchical structure
CN106126615A (en) * 2016-06-22 2016-11-16 苏州大学 The method and system that a kind of point of interest is recommended
CN108874877A (en) * 2018-05-03 2018-11-23 浙江大学城市学院 A kind of point of interest recommended method based on figure link analysis of facing position social networks
CN109190053A (en) * 2018-07-04 2019-01-11 南京邮电大学 One kind being based on point of interest importance and the authoritative point of interest recommended method of user

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
"Point-of-Interest Recommendation Based on User Contextual Behavior Semantics";Dongjin Yu等;《International Journal of Software Engineering and Knowledge Engineering》;20191130;第29卷(第11期);全文 *
"基于用户签到行为的兴趣点推荐方法研究";王嘉春;《中国优秀硕士学位论文全文数据库 信息科技辑》;20181015;全文 *

Also Published As

Publication number Publication date
CN111177565A (en) 2020-05-19

Similar Documents

Publication Publication Date Title
CN110929164B (en) Point-of-interest recommendation method based on user dynamic preference and attention mechanism
Xu et al. Quality assessment of building footprint data using a deep autoencoder network
CN109460520B (en) Interest point recommendation method based on geographic-social relationship and deep implicit interest mining
CN110119475B (en) POI recommendation method and system
CN110134885B (en) Interest point recommendation method, device, equipment and computer storage medium
CN108288067A (en) Training method, bidirectional research method and the relevant apparatus of image text Matching Model
CN109948066B (en) Interest point recommendation method based on heterogeneous information network
CN107341261B (en) Interest point recommendation method oriented to location social network
CN113505204B (en) Recall model training method, search recall device and computer equipment
Li et al. Where you instagram? associating your instagram photos with points of interest
CN102222098A (en) Method and system for pre-fetching webpage
CN111695046B (en) User portrait inference method and device based on spatio-temporal mobile data representation learning
CN102135983A (en) Group dividing method and device based on network user behavior
CN111538916B (en) Interest point recommendation method based on neural network and geographic influence
Ding et al. Spatial-temporal distance metric embedding for time-specific POI recommendation
CN111104607A (en) Location recommendation method and device based on sign-in data
CN112055038B (en) Method for generating click rate estimation model and method for predicting click probability
CN115408618A (en) Interest point recommendation method based on social relationship fusion position dynamic popularity and geographic features
CN111177565B (en) Interest point recommendation method based on correlation matrix and word vector model
WO2016037346A1 (en) Measuring and diagnosing noise in urban environment
Guo et al. Cosolorec: Joint factor model with content, social, location for heterogeneous point-of-interest recommendation
CN114461943B (en) Deep learning-based multi-source POI semantic matching method and device and storage medium thereof
Lyu et al. Movement-aware map construction
Hadiji et al. Computer science on the move: Inferring migration regularities from the web via compressed label propagation
Abdel-Fatao et al. Unifying spatial, temporal and semantic features for an effective GPS trajectory-based location recommendation

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
GR01 Patent grant
GR01 Patent grant