CN109918571B - POI recommendation algorithm combining expert trust in recommendation system - Google Patents
POI recommendation algorithm combining expert trust in recommendation system Download PDFInfo
- Publication number
- CN109918571B CN109918571B CN201910206690.9A CN201910206690A CN109918571B CN 109918571 B CN109918571 B CN 109918571B CN 201910206690 A CN201910206690 A CN 201910206690A CN 109918571 B CN109918571 B CN 109918571B
- Authority
- CN
- China
- Prior art keywords
- user
- check
- expert
- range
- recommendation
- 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
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a POI recommendation algorithm combined with expert trust in a recommendation system, which belongs to the field of recommendation algorithms, and mainly infers POIs (Point of Interest, namely points of interest) of users according to historical sign-in data of the users. That is, the POI of the user is predicted by mining sign-in information of the user and combining the influence of time and the influence of geographic position, so that the use experience and satisfaction of the user are improved.
Description
Technical Field
The invention relates to a POI recommendation algorithm combining expert trust in a recommendation system, and belongs to the field of recommendation algorithms.
Background
Recently, online Social Networking (OSN) service providers integrate paradigms by adding geographic locations as new dimensions to their frameworks. In conjunction with geographic information, online social networks have evolved into location-based social networks (LBSNs). Thus, through the LBSN, users can not only publish, reply and forward messages, but also declare their presence at a particular location (such activity is commonly referred to as "check-in"). Thus, users can establish social relationships and share their experience of accessing certain points of interest (e.g., museums, mountains, coffee shops).
A recent survey classified the recommendation systems in the LBSN. Of the categories they mention, our interest is in location (i.e., restaurant, bar, tourist attraction, etc.) recommendations. Most location recommendation systems are based on the concept of collaborative filtering and apply analysis algorithms to determine similarity between objects in the LBSN. In addition to these concepts, LBSN has also been studied to find related elements and their impact suggestions in terms of manufacturing. Some previous research efforts checked the user's personal static registration preferences through geographic location registration. Other people find social friends prone to similar enrollment behavior based on social impact theory, and researchers have also investigated obvious social friendships on LBSN. These findings indicate that user preferences for places, social influences and geographical influences play an important role in making suggestions.
Before the invention proposes, in the field of label recommendation algorithm of recommendation system, there are disadvantages of considering label weight and adding time factor, and cache replacement by using these methods:
(1) The influence of time and geographic position on POI recommendation is not fully considered, so that the recommendation result of a user is not ideal;
(2) When a new user wants to be recommended, there is no good solution in the face of the cold start problem, since it has no past check-in data and cannot be recommended according to his habit and sociality.
Disclosure of Invention
The invention aims to overcome the defects and develop a recommendation algorithm for recommending the POI by combining expert trust in a recommendation body system.
The technical scheme of the invention is as follows:
the POI recommendation algorithm combined with expert trust in the recommendation system is characterized by comprising the following steps:
(1) Acquiring sign-in information of a recommended user;
(2) Screening the rest sign-in information in the time period t within the distance d of the recommended user, and finishing preliminary expert user screening;
(3) Performing secondary screening on the check-in times and check-in ranges;
(4) Selecting POIs of expert users to obtain a recommendation list;
(5) And performing personalized optimization on the recommendation result by using a kernel function to obtain a better recommendation list.
Preferably, the step (2) performs preliminary screening according to the check-in data record of the user in combination with time and space.
Preferably, the step (3) fully combines the influence of the number of check-ins and the check-in range to select the most accurate expert user.
Preferably, the step (4) is used for acquiring POIs of expert users, sorting according to the number of check-in times and returning to the recommendation list.
Preferably, the step (5) is to perform secondary optimization of the recommendation result for the data sparse user, and return to the item recommendation list.
Preferably, the specific way to choose the most accurate expert user is as follows:
c for check-in times n The sign-in number is the number of times, and the sign-in number is the number of timesExpert user selection plays a critical role; c (C) n Is obtained directly through the information acquisition of the check-in times of the user, user u i Within a given time t and a given distance d, according to the number of check-ins C n Called expert user u e The probability of (2) is:
in the formula (1), the components are as follows,is user u i Is checked in times,/->Is the sum of all check-in times in range d;
c for check-in range r The method is characterized in that the wider the signing range is, the higher the borrowability of the user is; the smaller the check-in range, the lower the borrowability of the user, then user u i Within a given time t and a given distance d, according to the check-in range C r Called expert user u e The probability of (2) is:
in the formula (2), if user u i If the sign-in ranges of the user u and the user u are not intersected, alpha is 1, otherwise, 0 is obtained;for user u i Is checked in range of->A check-in range for user u;
comprehensively considering the influence of the number of check-in times and the check-in range, the probability of being an expert user is obtained as follows: p (u) e |u i ,t,d)=βP(u e |u i ,t,d,C n )+(1-β)P(u e |u i ,t,d,C r )(3);
In the formula (3), beta is an adjusting parameter (beta epsilon [0,1 ]), and the larger the beta is, the larger the influence of the sign-in times is; and (3) obtaining the user with the highest probability of becoming the expert user as the expert user through the formula, and arranging POIs (point of interest) in the range d from high to low according to the signing times at the time t to form a recommendation list.
Preferably, the preferred recommendation list in step (5) is as follows:
in equation (4), given user u and suggested location L, where n represents the total number of locations visited by the user l= { L 1 ,l 2 ...l n },d i Is l and l i Distance between them. The sum of f (d) is calculated:
equation (5) this is to calculate the probability that the user accesses the proposed location based on the unique check-in geographical distribution given by set D. d, d i Is l and l i The distance between D' is the distance from the set D of users, n D Is the total number of elements in the set D. Where f (d) is a non-parametric density estimation method (kernel density estimator), thus using the bandwidth and the gaussian distribution of kernels h and K (x), respectively:
h=1.06σn -1/5 (6) In formula (6), σ is the standard deviation of the samples in D:
according to the method, the POI (Point of Interest, namely the interest point) of the user is deduced according to the historical sign-in data of the user, namely, the POI of the user is predicted by mining sign-in information of the user and combining the influence of time and the influence of geographic position, so that the use experience and satisfaction of the user are improved.
Drawings
FIG. 1-schematic diagram of check-in times and check-in ranges in step (3) of the present invention;
FIG. 2-the effect of β value on recall in step (3) of the present invention;
FIG. 3-the effect of β value on accuracy in step (3) of the present invention;
FIG. 4-comparison of accuracy of the present invention with other algorithms;
FIG. 5-comparison of recall of the present invention with other algorithms.
Detailed Description
The POI recommendation algorithm combined with expert trust in the recommendation system comprises the following steps:
(1) Acquiring sign-in information of a recommended user;
(2) Screening the rest sign-in information in the time period t within the distance d of the recommended user, and finishing preliminary expert user screening;
(3) Performing secondary screening on the check-in times and check-in ranges;
c for check-in times n And representing the liveness of the user. The more the number of check-ins, the higher the user liveness; the fewer check-ins, the lower the user liveness. The number of check-ins plays a critical role in the selection of expert users, C n Is directly obtained through the information of the check-in times of the user.
Formula (1) represents user u i Within a given time t and a given distance d, according to the number of check-ins C n Called expert user u e Is a probability of (2).Is user u i Is checked in times,/->Is the range dThe sum of all check-in times in the network.
C for check-in range r Representing the range of motion of the user. The wider the sign-in range is, the higher the borrowability of the user is; the smaller the check-in range, the lower the borrowability of the user. The scope of check-in is therefore also of critical importance to the influence of the selection of expert users.
The invention obtains a minimum circumcircle covering all check-in places through check-in position information, and the check-in range is represented by calculating the area of the circle.The recommended user u takes the position of the recommended user u as the circle center, and d is the circle area of the radius. The latitude and longitude of the check-in position need to be converted into x and y coordinates by mapsis software before calculation is performed. And screening x min ,y;x max ,y;x,y max ;x,y min The four points are recorded once, and the number is C m And (3) representing.
If: (1) c (C) m ≤0,②0<C m ≤2,/>The center of the connecting line of the two points is used as the circle center, and the distance between the connecting lines of the two points is used as the circular area of the diameter; (3) c (C) m =3 and 3 points on a straight line, +.>The area of the circle taking the midpoint of the straight line as the center and the straight line as the diameter is adopted; (4) c (C) m =3 and 3 points are not on a straight line, +.>The area of the circumcircle of the three points is defined; (5) c (C) m =4,/>The area of the circumscribed circle for the four points; (6) if->And->The intersection only needs to calculate the area of the intersection part.
Then user u i Within a given time t and a given distance d, according to the check-in range C r Called expert user u e The probability of (2) is:
in the formula (2), if user u i If the check-in ranges of the user u are not intersected, alpha is 1, and otherwise, 0.For user u i Is checked in range of->Is the check-in range for user u.
To comprehensively consider the influence of the number of check-ins and the check-in range, equations (1) and (2) can be combined, and then user u selects user u i The probability for an expert user is: p (u) e |u i ,t,d)=βP(u e |u i ,t,d,C n )+(1-β)P(u e |u i ,t,d,C r )(3)。
In the formula (3), beta is the adjustment parameter (beta epsilon [0,1 ]), and the larger the beta is, the larger the influence of the check-in times is. And (3) obtaining the user with the highest probability of becoming the expert user as the expert user through the formula, and arranging POIs (point of interest) in the range d from high to low according to the signing times at the time t to form a recommendation list.
(4) Selecting POIs of expert users to obtain a recommendation list;
(5) Personalized optimization of recommended results using kernel functionsAnd obtaining a better recommendation list.h=1.06σn -1/5 ,
Step (1) acquiring a user ID, a POI ID, time and longitude and latitude according to sign-in data of a user;
extracting the acquired user information through the step (2);
weighing the influence of the number of check-in times and the check-in range through the step (3), and selecting expert users;
sorting POIs of expert users according to the signing times from high to low through the step (4) to generate a TOP-N list, recommending the list if a new user is available, and otherwise, entering the step (5);
and (5) for the sparse data users, performing personalized updating on the recommendation list by using a kernel function.
A good recommendation system is to achieve the effect of multi-party win-win, and therefore has multiple judgment standards. The invention adopts recall rate and accuracy rate to evaluate the accuracy of the recommendation algorithm. The recall rate calculation formula is:the accuracy rate calculation formula is: />
Fig. 4 and 5 show the accuracy and recall of the TOP-N, N values of 5,10,15,20, respectively, and the recommendation effect of each algorithm is analyzed accordingly. From the analysis of the experimental results, it is possible to obtain: the recommendation effect of the algorithm U is general as a basic recommendation algorithm; the algorithm P considers the change of the user interest by combining with the time factor, so that the recommendation effect is improved; and the algorithm T also considers the influence of the geographic distance on the basis of the time factor, so that a relatively better recommendation effect is obtained. However, the algorithm selects expert users by comprehensively considering time, geographic factors, check-in times and check-in ranges, and performs secondary optimization on recommended results by using a kernel function, so that the accuracy and recall rate are superior to those of the first three algorithms. This is also illustrating the effectiveness of the algorithm, and can more accurately make recommendations to the user.
The invention has the advantages that the expert user is selected by comprehensively considering time, geographical factors, check-in times and check-in range, and the recommended result is secondarily optimized by using the kernel function, so that the recall rate and the accuracy are improved.
Claims (1)
1. The POI recommendation algorithm combined with expert trust in the recommendation system is characterized by comprising the following steps:
(1) Acquiring sign-in information of a recommended user;
(2) Screening the rest sign-in information in the time period t within the distance d of the recommended user, and finishing preliminary expert user screening;
(3) Performing secondary screening on the check-in times and check-in ranges;
(4) Selecting POIs of expert users to obtain a recommendation list;
(5) Personalized optimization is carried out on the recommendation result by using a kernel function, so that a better recommendation list is obtained;
the step (2) performs preliminary screening according to the sign-in data record of the user and the combination time and space;
the step (3) fully combines the influence of the check-in times and the check-in range to select the most accurate expert user;
the step (4) is used for acquiring POIs of expert users, sorting according to the signing times and returning to a recommendation list;
the step (5) is used for carrying out secondary optimization on the recommendation result of the data sparse user and returning to the article recommendation list;
the specific method for selecting the most accurate expert user is as follows:
c for check-in times n The representation represents the activity of the user, and the more the number of check-ins is, the higher the activity of the user is, and the number of check-ins isThe fewer the number, the lower the user liveness, so the number of check-ins plays a critical role in the selection of expert users; c (C) n Is obtained directly through the information acquisition of the check-in times of the user, user u i Within a given time t and a given distance d, according to the number of check-ins C n Called expert user u e The probability of (2) is:
in the formula (1), the components are as follows,is user u i Is checked in times,/->Is the sum of all check-in times in range d;
c for check-in range r The method is characterized in that the wider the signing range is, the higher the borrowability of the user is; the smaller the check-in range, the lower the borrowability of the user, then user u i Within a given time t and a given distance d, according to the check-in range C r Called expert user u e The probability of (2) is:
in the formula (2), if user u i If the sign-in ranges of the user u and the user u are not intersected, alpha is 1, otherwise, 0 is obtained;for user u i Is checked in range of->A check-in range for user u;
comprehensively considering the influence of the number of check-in times and the check-in range, the probability of being an expert user is obtained as follows:
P(u e |u i ,t,d)=βP(u e |u i ,t,d,C n )+(1-β)P(u e |u i ,t,d,C r ) (3);
in the formula (3), beta is an adjusting parameter (beta epsilon [0,1 ]), and the larger the beta is, the larger the influence of the sign-in times is; obtaining a user with the highest probability of becoming an expert user, namely the expert user, through a formula (3), and arranging POIs (point of interest) in a range d from high to low according to the signing times at time t to form a recommendation list;
the preferred recommendation list in step (5) is as follows:
in equation (4), given user u and suggested location L, where n represents the total number of locations visited by the user l= { L 1 ,l 2 ...l n },d i Is l and l i A distance therebetween; the sum of f (d) is calculated:
equation (5) this is to calculate the probability that the user accesses the proposed location based on the unique check-in geographic distribution given by set D; d, d i Is l and l i The distance between D' is the distance from the set D of users, n D Is the total number of elements in the set D; where f (d) is a non-parametric density estimation method, thus using bandwidth and gaussian distributions for kernels h and K (x), respectively:
h=1.06σn -1/5 (6) In formula (6), σ is the standard deviation of the samples in D;
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910206690.9A CN109918571B (en) | 2019-03-19 | 2019-03-19 | POI recommendation algorithm combining expert trust in recommendation system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910206690.9A CN109918571B (en) | 2019-03-19 | 2019-03-19 | POI recommendation algorithm combining expert trust in recommendation system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109918571A CN109918571A (en) | 2019-06-21 |
CN109918571B true CN109918571B (en) | 2023-08-22 |
Family
ID=66965560
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910206690.9A Active CN109918571B (en) | 2019-03-19 | 2019-03-19 | POI recommendation algorithm combining expert trust in recommendation system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109918571B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102479202A (en) * | 2010-11-26 | 2012-05-30 | 卓望数码技术(深圳)有限公司 | Recommendation system based on domain expert |
CN105260795A (en) * | 2015-10-13 | 2016-01-20 | 广西师范学院 | Time-space prediction method for position of key personnel based on conditional random field |
CN106886921A (en) * | 2017-02-17 | 2017-06-23 | 正源信用(北京)科技有限公司 | Personalized recommendation method based on user interest |
CN109460520A (en) * | 2019-01-21 | 2019-03-12 | 宿州学院 | Point of interest recommended method based on geography-social relationships and deep implicit interest digging |
-
2019
- 2019-03-19 CN CN201910206690.9A patent/CN109918571B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102479202A (en) * | 2010-11-26 | 2012-05-30 | 卓望数码技术(深圳)有限公司 | Recommendation system based on domain expert |
CN105260795A (en) * | 2015-10-13 | 2016-01-20 | 广西师范学院 | Time-space prediction method for position of key personnel based on conditional random field |
CN106886921A (en) * | 2017-02-17 | 2017-06-23 | 正源信用(北京)科技有限公司 | Personalized recommendation method based on user interest |
CN109460520A (en) * | 2019-01-21 | 2019-03-12 | 宿州学院 | Point of interest recommended method based on geography-social relationships and deep implicit interest digging |
Non-Patent Citations (1)
Title |
---|
基于地理位置和社交关系的兴趣点推荐算法研究;陈艳;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;20190115(第01期);正文第33-44页 * |
Also Published As
Publication number | Publication date |
---|---|
CN109918571A (en) | 2019-06-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11470443B2 (en) | Harvesting labels for significant locations based on candidate points of interest and contextual data | |
CN108460101B (en) | Point of interest recommended method of the facing position social networks based on geographical location regularization | |
JP5536485B2 (en) | Portable terminal, server, program, and method for estimating address / location as user moves | |
CN109376313B (en) | Information recommendation method and device, terminal and storage medium | |
KR101979401B1 (en) | Ranking nearby destinations based on visit likelihoods and predicting future visits to places from location history | |
JP2018528505A (en) | Provide, organize, and manage location history records for mobile devices | |
CN107341261B (en) | Interest point recommendation method oriented to location social network | |
US20150058345A1 (en) | Realtime activity suggestion from social and event data | |
Liu et al. | Points of interest recommendation from GPS trajectories | |
CN110334293B (en) | Position social network-oriented position recommendation method with time perception based on fuzzy clustering | |
CN106960044B (en) | Time perception personalized POI recommendation method based on tensor decomposition and weighted HITS | |
CN106875205B (en) | Object selection method and device | |
CN107220312B (en) | Interest point recommendation method and system based on co-occurrence graph | |
CN106605418A (en) | Power management of mobile clients using location-based services | |
Marakkalage et al. | Understanding the lifestyle of older population: Mobile crowdsensing approach | |
Chu et al. | Friend recommendation for location-based mobile social networks | |
CN110134879B (en) | Interest point recommendation algorithm based on differential privacy protection | |
CN108369586A (en) | System and method for detecting interaction and influence in network | |
US9609485B2 (en) | Predicting companion data types associated with a traveler at a geographic region including lodging | |
KR101690620B1 (en) | Pattern labeling | |
CN110874437B (en) | Personalized interest point recommendation method based on multiple interest point pair ordering | |
KR101693429B1 (en) | System for identifying human relationships around users and coaching based on identified human relationships | |
CN109684561B (en) | Interest point recommendation method based on deep semantic analysis of user sign-in behavior change | |
CN111538916A (en) | Interest point recommendation method based on neural network and geographic influence | |
CN109918571B (en) | POI recommendation algorithm combining expert trust in recommendation system |
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 |