CN109101642B - Method for reducing group recommendation list based on subgroup and social behavior - Google Patents

Method for reducing group recommendation list based on subgroup and social behavior Download PDF

Info

Publication number
CN109101642B
CN109101642B CN201810951568.XA CN201810951568A CN109101642B CN 109101642 B CN109101642 B CN 109101642B CN 201810951568 A CN201810951568 A CN 201810951568A CN 109101642 B CN109101642 B CN 109101642B
Authority
CN
China
Prior art keywords
subgroup
group
preference
user
representing
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
CN201810951568.XA
Other languages
Chinese (zh)
Other versions
CN109101642A (en
Inventor
毛宇佳
刘学军
何瑾琳
张军强
陆淑娟
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.)
Nanjing Tech University
Original Assignee
Nanjing Tech 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 Nanjing Tech University filed Critical Nanjing Tech University
Priority to CN201810951568.XA priority Critical patent/CN109101642B/en
Publication of CN109101642A publication Critical patent/CN109101642A/en
Application granted granted Critical
Publication of CN109101642B publication Critical patent/CN109101642B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/01—Social networking

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a method for reducing a group recommendation list based on subgroups and social behaviors, and belongs to the technical field of data processing. Dividing a data set into a plurality of groups to obtain a project theme characteristic model and a user theme preference model, and dividing each group into a plurality of subgroups; acquiring an initial group recommendation list, subgroup preferences and subgroup weights according to the subgroups; obtaining group preference through a weighting model according to the subgroup weight and the subgroup preference; and performing similarity matching on the group preference and the initial group recommendation list to obtain a final group recommendation list. The invention can maximally reduce the group recommendation list on the premise of meeting the recommendation accuracy and fairness, so that the group members can make a selection more conveniently.

Description

Method for reducing group recommendation list based on subgroup and social behavior
Technical Field
The invention belongs to the technical field of data processing, and particularly relates to a method for reducing a group recommendation list based on subgroups and social behaviors.
Background
With the rise of the internet era, people's clothes and food habits have changed greatly. Users purchase, read and browse on the internet, enjoy the rapidness brought by the internet and provide a lot of personal information for the internet, and the users not only receive information on the internet, but also manufacture information, which causes the explosive growth of data. In the face of increasingly dramatic data, sufficient analysis and utilization are needed to better serve users, and how to quickly and accurately obtain contents desired by users from massive data is a problem to be solved at present. The recommendation system aims to provide the user with items of interest in an overloaded search space. Currently, recommendation systems have been successfully applied in the fields of education, electronic commerce, financial investment, etc.
With the growing development of recommendation systems, the acquisition channels of user preferences are more and more diversified, and different additional information is also used for analyzing the requirement preferences of users, such as social relations, check-in records, comment information, pictures uploaded by the users and the like. The purpose of the recommendation system is also gradually expanded from the traditional accuracy of recommendation improvement to the improvement of the real-time performance and diversity of recommendation, so that the use experience of the user is improved, and better service is provided for the user more humanizedly. However, the conventional recommendation system provides recommendations for a single user, and many recommendations are often provided for a group of people, so the recommendation system needs to consider the needs of each user in the group and make recommendations satisfying the group members, thereby generating a group recommendation system. At present, group recommendation researches are mainly aimed at improving accuracy and fairness of results, and researches between scale size of a research recommendation list and recommendation satisfaction are not much. The final selection of the group members can be influenced by too large and too small a recommendation list generated by a group, and the larger the list is, the greater the diversity is, but the more difficult the member selection is; the list is too small to guarantee that as many items in the recommendation list as possible meet the member's preferences.
Disclosure of Invention
The invention aims to provide a method for reducing a group recommendation list based on subgroups and social behaviors, which can solve the problem of overlarge scale of the recommendation list in group recommendation and can ensure the fairness of recommendation.
Specifically, the invention is realized by adopting the following technical scheme, comprising the following steps:
dividing the data set into a plurality of groups to obtain a project theme characteristic model and a user theme preference model, and dividing each group into a plurality of subgroups;
acquiring an initial group recommendation list, subgroup preferences and subgroup weights according to the subgroups;
obtaining group preference through a weighting model according to the subgroup weight and the subgroup preference;
and performing similarity matching on the group preference and the initial group recommendation list to obtain a final group recommendation list.
Further, the step of obtaining the item topic feature model and the user topic preference model includes:
acquiring a project theme characteristic model by adopting an LDA theme model;
dividing users into active users and inactive users according to historical scoring records; an active user obtains a user theme preference model through TF-IDF and time factors; an inactive user obtains a theme preference model through an external expert;
members with high similarity are divided into a subgroup by utilizing similarity of preference of attributes among the members.
Further, the step of obtaining the initial group recommendation list comprises:
obtaining a similar subgroup of the target subgroup from two angles of scoring information and cross-item attribute information through a cosine similarity formula;
and obtaining similar subgroups of subgroups in the current group according to the similar subgroups, and obtaining recommendation lists of the subgroups in the current group, wherein the set of the recommendation lists is used as an initial group recommendation list. Further, the calculating of the subgroup preference refers to calculating the similarity between the members, and taking the preference with high similarity of the subgroup members as the current subgroup preference.
Further, the calculating of the subgroup weight means that the data is further processed, and the weight occupied by the subgroup in the current group is calculated according to the obtained tolerance and the beneficial behavior index of the members in the group.
The invention has the following beneficial effects: the method for reducing the group recommendation list based on the subgroups and the social behaviors maximally reduces the group recommendation list on the premise of meeting the recommendation accuracy and fairness, so that group members can make selections more conveniently.
Drawings
Fig. 1 is a system framework diagram of embodiment 1 of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following examples and the accompanying drawings.
Example 1:
an embodiment of the present invention, taking the movie list as an example, introduces a method for narrowing down the group recommendation list based on subgroups and social behaviors, and the implementation process is shown in fig. 1.
The method comprises the following steps: the group is divided into subgroups.
The method comprises the steps of filtering useless data of an obtained original data set, preprocessing (randomly dividing the data set into a plurality of groups) the data set to obtain a project theme feature (namely, a movie type) model and a user theme preference model, and dividing each group into a plurality of subgroups. The method for obtaining the project theme feature model and the user theme preference model comprises the following steps:
1-1) adopting an LDA (Latent Dirichlet Allocation) topic model to obtain the following project topic feature models:
Figure BDA0001770546400000021
wherein, use
Figure BDA0001770546400000031
Representing the s-th film msWhether or not the subject feature g is contained ini. When in use
Figure BDA0001770546400000032
Representing a movie msContains subject feature giOn the contrary, the number of the first and second,
Figure BDA0001770546400000033
representing a movie msDoes not contain the subject feature gi。
1-2) dividing users into active and inactive users according to historical scoring records.
Active users can communicate with each other through TF-IDF (Term Frequency-Inverse file Frequency)The time factor obtains a user topic preference model. Active user ujFor subject feature giPreference of
Figure BDA0001770546400000034
The calculation method is as follows:
Figure BDA0001770546400000035
wherein the content of the first and second substances,
Figure BDA0001770546400000036
representing a movie msWhether or not the subject feature g is contained iniWhen it comes to
Figure BDA0001770546400000037
Representing a movie msContaining the subject feature giOn the contrary, in the case of a high-frequency,
Figure BDA0001770546400000038
representing a movie msDoes not contain the subject feature gi(ii) a n is the number of movies in the current group; k is the number of the current movie theme features;
Figure BDA0001770546400000039
representing a user u merged into a forgetting functioniFor movie msScoring of (4);
Figure BDA00017705464000000310
the calculation formula of (2) is as follows:
Figure BDA00017705464000000311
wherein the content of the first and second substances,
Figure BDA00017705464000000312
representative user ujFor movie msF (Δ t) represents a forgetting function, and f (Δ t) is calculated by the formula:
Figure BDA00017705464000000313
wherein, Δ T represents the time difference from the scoring behavior of the user to the current time point, T0Is a decay coefficient, controls the speed of interest decay, T0The larger the interest decay rate. For temporary preferences, i.e. the preferences just generated, the forgetting speed of the user is faster due to the smaller time difference Δ t, and for fixed preferences, the forgetting speed is slower.
Thus, active user ujSubject preference model
Figure BDA00017705464000000314
Comprises the following steps:
Figure BDA00017705464000000315
where m is the total number of users, i.e. the sum of the number of active users and the number of inactive users.
The inactive user obtains the theme preference model through an external expert. Based on the observation of users in the sweepy microblog, the following two types of users are collectively called as external experts:
● social celebrities, have a large number of concerns and are recognized by the general public in real life.
● experts, are in widespread use in a particular area.
External expert etFor subject feature giPreference of
Figure BDA00017705464000000316
The calculation method of (A) is as follows:
Figure BDA0001770546400000041
wherein the content of the first and second substances,
Figure BDA0001770546400000042
external expert e representing the integration of a forgetting functiontSubject feature g in microblogiThe number of the forgetting function, where the calculation formula of the forgetting function is the same as the forgetting function f (Δ t);
Figure BDA0001770546400000043
representing external experts etThe total number of microblogs; k is the number of movie theme features;
Figure BDA0001770546400000044
representing user ujA set of external experts of interest.
Therefore, an external expert etSubject preference model
Figure BDA0001770546400000045
Comprises the following steps:
Figure BDA0001770546400000046
the inactive user obtains the theme preference model through an external expert. Topic preference model for inactive users
Figure BDA0001770546400000047
The calculation method is as follows:
Figure BDA0001770546400000048
wherein the content of the first and second substances,
Figure BDA0001770546400000049
is a subject preference model for an external expert,
Figure BDA00017705464000000410
representing user ujWith external experts etThe common behavior ratio is calculated by the following method:
Figure BDA00017705464000000411
wherein the content of the first and second substances,
Figure BDA00017705464000000412
representing user ujForwarding or commenting by etThe number of micro-blogs to send,
Figure BDA00017705464000000413
representing user ujA set of external experts of interest.
1-3) members with high similarity are divided into a subgroup by utilizing similarity of preference of attributes among the members.
Member ujAnd ulInter-pair attribute giSimilarity of preference
Figure BDA00017705464000000414
The calculation method is as follows:
Figure BDA00017705464000000415
wherein the content of the first and second substances,
Figure BDA00017705464000000416
representative user ujFor giThe degree of preference of (a) is,
Figure BDA00017705464000000417
representative user u1For giK is the number of movie theme features, calculated from the above equation (2).
Step two: obtaining an initial group recommendation list, calculating subgroup preference and calculating subgroup weight.
2a) An initial group recommendation list is obtained.
2a-1) obtaining a similar subgroup Sim (SG) of the target subgroup from two angles of the scoring information and the cross-item attribute information through the following cosine similarity formulax，SGy)。sim(SGx，SGy) The calculation formula of (2) is as follows:
sim(SGx，SGy)＝λsimR(SGx，SGy)+(1-λ)simp(SGx，SGy) (11)
wherein simR(SGx，SGy) Representing subgroup SG based on scoring informationxAnd SGySimilarity between, simp(SGx，SGy) Representing sub-group SG based on cross-item attributexAnd SGyAnd the similarity between the two groups is that lambda is a weight factor and the value range is 0-1.
In formula (11), subgroup SG based on score informationxAnd SGySimilarity sim betweenR(SGx，SGy) The calculation formula of (c) is:
Figure BDA0001770546400000051
wherein the content of the first and second substances,
Figure BDA0001770546400000052
representing subgroup SGxAnd SGyThe collection of movies viewed by the middle members together is calculated by calculating the number of the sub-group members to the movie msIs scored to obtain subgroup SGxFor movie msIs scored
Figure BDA0001770546400000053
And subgroup SGyFor movie msIs scored
Figure BDA0001770546400000054
In equation (11), subgroup SG based on cross-item attributesxAnd SGySimilarity sim betweenp(SGx，SGy) The calculation formula of (2) is as follows:
Figure BDA0001770546400000055
wherein the content of the first and second substances,
Figure BDA0001770546400000056
representing subgroup SGxAnd SGyA collection of movies that the members in the group have viewed together,
Figure BDA0001770546400000057
representative subgroup SGxFor giThe value of the preference of (c) is,
Figure BDA0001770546400000058
representative subgroup SGyFor giA preference value of (c).
2a-2) according to similar subgroup Sim (SG)x，SGy) Obtaining subgroup SG in current group GxAnd obtaining the recommendation lists of all subgroups in the current group G, and collecting the recommendation lists as an initial group recommendation list top-N. the top-N is obtained as follows:
Figure BDA0001770546400000059
wherein the content of the first and second substances,
Figure BDA00017705464000000510
representative subgroup SGxIs a subgroup SGxFor movie msIs scored
Figure BDA00017705464000000511
The top k movies.
2b) Using the similarity between the members calculated in 1-3), the common preference (i.e. the attribute preference with high similarity) of the subgroup members is used as the current subgroup preference
Figure BDA00017705464000000512
2c) The subgroup weights are calculated.
Further processing the data according to the resulting tolerance of members of the groupDegree and Litta behavior index, calculating the subgroup SG in the current groupxThe occupied weight specifically comprises the following steps:
2c-1) analyzing the reactions of different users in the conflict situation when the users face strangers to obtain the tolerance index of the users
Figure BDA0001770546400000061
The calculation formula is as follows:
Figure BDA0001770546400000062
wherein, X1And X2Representing the social activity and social influence of the user, alpha, beta and gamma are parameters, the value ranges of alpha and beta are-1-0, the value ranges of gamma are 0-1, and X is1And X2The calculation methods of (a) are respectively as follows:
Figure BDA0001770546400000063
Figure BDA0001770546400000064
wherein the content of the first and second substances,
Figure BDA0001770546400000065
representing user ujIs the set of people of interest of (a),
Figure BDA0001770546400000066
representing user ujIs focused on a person's preferred subject matter feature set,
Figure BDA0001770546400000067
indicating the length of time for which the user is registered,
Figure BDA0001770546400000068
a set of fans representing a user is presented,
Figure BDA0001770546400000069
representing user ujThe mutual powder aggregation of (1).
2c-2) consider when user ujFace friend u1Then, the user u is obtained according to the social relationshipjRival behavior index of
Figure BDA00017705464000000610
The calculation formula is as follows:
Figure BDA00017705464000000611
wherein the content of the first and second substances,
Figure BDA00017705464000000612
representing user ujThe mutual powder aggregation of (A) and (B),
Figure BDA00017705464000000617
representing user u1The mutual powder aggregation of (1).
2c-3) obtaining the subgroup SG in the current group according to the obtained user tolerance and the Litta behavior indexxOccupied weight
Figure BDA00017705464000000613
Figure BDA00017705464000000614
The calculation formula of (2) is as follows:
Figure BDA00017705464000000615
wherein, | SGx| represents subgroup SGxThe number of people in (1) is,
Figure BDA00017705464000000618
representative user ujThe tolerance of the pressure sensor is indicated by the tolerance index,
Figure BDA00017705464000000616
representative user ujFace friend u1Users u in timejRival behavior index of (SG)x-{ujRepresents subgroup SGxUser u is dividedjOther users than the others.
Step three: a group preference is obtained.
Obtaining group preference through a weighting model according to the subgroup weight and the subgroup preference, wherein the calculation formula is as follows:
Figure BDA0001770546400000071
wherein, SGxFor the purpose of the current sub-group,
Figure BDA0001770546400000072
is subgroup SG in current subgroup GxThe weight that is taken up by the user,
Figure BDA0001770546400000073
preference for subgroups, i.e. subgroup SGxThe theme preferences of (3), have been obtained by step 2 b).
Step four: and obtaining a final group recommendation list.
And performing similarity matching on the group preference and the initial group recommendation list to obtain a final group recommendation list. Similarity matching is carried out by utilizing a cross-item attribute similarity calculation formula, and the calculation mode is as follows:
Figure BDA0001770546400000074
wherein the content of the first and second substances,
Figure BDA0001770546400000075
representing the target group G pair GiThe value of the preference of (c) is,
Figure BDA0001770546400000076
representative movie msWhether or not the subject feature g is contained iniMovie, filmmsIs the set of movies contained in the initial group recommendation list top-N, k being the number of movie theme features.
In order to evaluate the performance of the algorithm, a data set can be divided into a training set and a testing set, and Precision (Precision), Recall (Recall) and an F-measure (F-measure) are used as evaluation indexes of the performance of the algorithm. The calculation method of each evaluation index is as follows:
Figure BDA0001770546400000077
Figure BDA0001770546400000078
Figure BDA0001770546400000079
wherein E is*Representing items that are present in the final recommended list of the training set and the test set at the same time, i.e. items that are predicted to be correct, EDRepresenting items in the final recommendation list in the test set, ErAn item representing the final recommendation list in the training set. The larger the Precision, Recall, F values are, the better the algorithm performance is.
And for each group, calculating the precision rate, the recall rate and the F value of the group according to a final recommendation list obtained by the group in the training set and the test set, and finally taking the average value of the precision rate, the recall rate and the F value of all the groups as the final result of the algorithm.
Different group partitions can be performed on the data set again, the final recommendation list is obtained according to the method, and the average values of the accuracy rate, the recall rate and the F value of all the groups are calculated. And comparing the result with the result obtained by the previous group division mode, and when the average values of the precision rate, the recall rate and the F value of the two results tend to be converged, the current group division of the data set and the obtained final recommendation result can be considered to be appropriate.
Although the present invention has been described in terms of the preferred embodiment, it is not intended that the invention be limited to the embodiment. Any equivalent changes or modifications made without departing from the spirit and scope of the present invention also belong to the protection scope of the present invention. The scope of the invention should therefore be determined with reference to the appended claims.

Claims (2)

1. The method for reducing the group recommendation list based on the subgroup and the social behavior is characterized by comprising the following steps:
dividing the data set into a plurality of groups to obtain a project theme characteristic model and a user theme preference model, and dividing each group into a plurality of subgroups;
acquiring an initial group recommendation list, subgroup preferences and subgroup weights according to the subgroups;
obtaining group preference through a weighting model according to the subgroup weight and the subgroup preference;
performing similarity matching on the group preference and the initial group recommendation list to obtain a final group recommendation list;
the step of obtaining the project theme feature model and the user theme preference model and dividing each group into a plurality of subgroups comprises the following steps:
acquiring a project theme characteristic model by adopting an LDA theme model;
dividing users into active users and inactive users according to historical scoring records; the active user obtains a user theme preference model through TF-IDF and a time factor; an inactive user obtains a theme preference model through an external expert;
dividing the data of the members with high similarity into a subgroup by utilizing the similarity of the preference of the members to the attributes;
the active user obtains the user theme preference model through the TF-IDF and the time factor, and the method comprises the following steps:
computing active user ujFor subject feature giPreference of
Figure FDA0003614548140000011
The calculation method is as follows:
Figure FDA0003614548140000012
wherein the content of the first and second substances,
Figure FDA0003614548140000013
representing a movie msWhether or not the subject feature g is contained iniWhen it comes to
Figure FDA0003614548140000014
Representing a movie msContains subject feature giOn the contrary, the number of the first and second,
Figure FDA0003614548140000015
representing a movie msDoes not contain the subject feature gi(ii) a n is the number of movies in the current group; k is the number of the current movie theme features;
Figure FDA0003614548140000016
representing a user u who has merged into a forgetting functionjFor movie msScoring of (4);
Figure FDA0003614548140000017
the calculation formula of (2) is as follows:
Figure FDA0003614548140000018
wherein the content of the first and second substances,
Figure FDA0003614548140000019
representative user ujFor movie msF (Δ t) represents a forgetting function, and f (Δ t) is calculated by the formula:
Figure FDA00036145481400000110
wherein, Δ T represents the time difference from the scoring behavior of the user to the current time point, T0Is a decay coefficient, controls the speed of interest decay, T0The larger, the slower the interest decay rate; for temporary preference, namely preference just generated, the forgetting speed of the user is higher due to smaller time difference delta t, and for fixed preference, the forgetting speed is lower;
computing active user ujSubject preference model
Figure FDA0003614548140000021
Comprises the following steps:
Figure FDA0003614548140000022
wherein m is the total number of users, i.e. the sum of the number of active users and the number of inactive users;
the method for acquiring the theme preference model by the inactive user through the external expert comprises the following steps:
computing external expert etFor subject feature giPreference of
Figure FDA0003614548140000023
The calculation method is as follows:
Figure FDA0003614548140000024
wherein the content of the first and second substances,
Figure FDA0003614548140000025
external expert e representing the integration of a forgetting functiontSubject feature g in microblogiThe number of forgetting functions, where the calculation formula of the forgetting function is the same as the forgetting function f (Δ t);
Figure FDA0003614548140000026
representing an external expert etTotal micro-blog ofCounting; k is the number of movie theme features;
Figure FDA0003614548140000027
representing user ujA set of external experts focused on;
define external expert etSubject preference model
Figure FDA0003614548140000028
Comprises the following steps:
Figure FDA0003614548140000029
computing topic preference models for inactive users
Figure FDA00036145481400000210
The calculation method is as follows:
Figure FDA00036145481400000211
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA00036145481400000212
is a subject preference model for an external expert,
Figure FDA00036145481400000213
representing user ujWith external experts etThe common behavior ratio is calculated by the following method:
Figure FDA00036145481400000214
wherein the content of the first and second substances,
Figure FDA00036145481400000215
representing user ujForwarding or commenting by etThe number of micro-blogs to send,
Figure FDA00036145481400000216
representing user ujA set of external experts focused on;
the step of obtaining an initial group recommendation list comprises:
obtaining a similar subgroup Sim (SG) of the target subgroup from two angles of the scoring information and the cross-item attribute information through a cosine similarity formulax,SGy)；
According to similar subgroup Sim (SG)x,SGy) Obtaining subgroup SG in current group GxObtaining the recommendation lists of all subgroups in the current group G, and taking the set of the recommendation lists as an initial group recommendation list top-N;
similar subgroup Sim (SG) of the target subgroupx，SGy) The calculation formula of (2) is as follows:
sim(SGx，SGy)＝λsimR(SGx，SGy)+(1-λ)simp(SGx，SGy) (11)
wherein, simR(SGx，SGy) Representing subgroup SG based on scoring informationxAnd SGySimilarity between, simp(SGx，SGy) Representing sub-group SG based on cross-item attributexAnd SGyThe similarity between the two groups is that lambda is a weight factor and the value range is 0-1;
in the formula (11), the subgroup SG based on the score informationxAnd SGySimilarity between them simR(SGx，SGy) The calculation formula of (c) is:
Figure FDA0003614548140000031
wherein the content of the first and second substances,
Figure FDA0003614548140000032
representing subgroup SGxAnd SGyThe collection of movies viewed by the middle members together is calculated by calculating the number of the sub-group members to the movie msThe average score of (3) results in subgroup SGxFor movie msIs scored by
Figure FDA0003614548140000033
And subgroup SGyFor movie msIs scored
Figure FDA0003614548140000034
In formula (11), the subgroup SG based on the cross-item attributexAnd SGySimilarity sim betweenp(SGx，SGy) The calculation formula of (2) is as follows:
Figure FDA0003614548140000035
wherein the content of the first and second substances,
Figure FDA0003614548140000036
representing subgroup SGxAnd SGyA collection of movies that the members in the group have viewed together,
Figure FDA0003614548140000037
representative subgroup SGxFor giThe value of the preference of (c) is,
Figure FDA0003614548140000038
representative subgroup SGyFor giA preference value of;
the method for acquiring the initial group recommendation list top-N comprises the following steps:
Figure FDA0003614548140000039
wherein the content of the first and second substances,
Figure FDA00036145481400000310
representative subgroup SGxIs a subgroup SGxFor movie msIs scored
Figure FDA00036145481400000311
Highest height
Figure FDA00036145481400000312
A movie;
the subgroup preference refers to a preference with high similarity of subgroup members;
the subgroup weight refers to the weight of the subgroup in the current group obtained by calculation according to the tolerance of the members in the group and the Litta behavior index;
performing similarity matching between the group preference and the initial group recommendation list to obtain a final group recommendation list, wherein the step of performing similarity matching between the group preference and the initial group recommendation list comprises the following steps:
similarity matching is carried out by utilizing a cross-item attribute similarity calculation formula, and the calculation mode is as follows:
Figure FDA0003614548140000041
wherein the content of the first and second substances,
Figure FDA0003614548140000042
representing the target group G pair GiThe value of the preference of (c) is,
Figure FDA0003614548140000043
representative movie msWhether or not the subject feature g is contained iniMovie msIs the set of movies contained in the initial group recommendation list top-N, k being the number of movie theme features.
2. The method of claim 1, wherein the tolerance index of the members in the group is a measure of the probability of the members in the group contracting the group recommendation list based on their subgroups and social behaviors
Figure FDA0003614548140000044
The calculation formula is as follows:
Figure FDA0003614548140000045
wherein, X1And X2Representing the social activity and social influence of the user, alpha, beta and gamma are parameters, the value ranges of alpha and beta are-1-0, the value ranges of gamma are 0-1, and X is1And X2The calculation methods of (A) are respectively as follows:
Figure FDA0003614548140000046
Figure FDA0003614548140000047
wherein the content of the first and second substances,
Figure FDA0003614548140000048
representing user ujIs the set of people of interest of (a),
Figure FDA0003614548140000049
representing user ujIs focused on a person's preferred subject matter feature set,
Figure FDA00036145481400000410
indicates the length of time for which the user is registered,
Figure FDA00036145481400000411
a set of fans that represent the user is represented,
Figure FDA00036145481400000412
representing user ujMutual powder aggregation of (1);
the groupRival behavior index of members of the group
Figure FDA00036145481400000413
The calculation formula is as follows:
Figure FDA00036145481400000414
wherein the content of the first and second substances,
Figure FDA00036145481400000415
representing user ujThe mutual powder aggregation of (A) and (B),
Figure FDA00036145481400000416
representing user u1Mutual powder aggregation of (1);
weights occupied by subgroups in the current group
Figure FDA00036145481400000417
The calculation formula is as follows:
Figure FDA00036145481400000418
wherein, | SGx| represents subgroup SG in current groupxThe number of people in (A) is,
Figure FDA0003614548140000051
representative user ujThe tolerance of the pressure sensor is indicated by the tolerance index,
Figure FDA0003614548140000052
representative user ujFace friend u1Users u in timejRival behavior index of (SG)x-{ujRepresents subgroup SGxUser u is dividedjOther users than the others.
CN201810951568.XA 2018-08-20 2018-08-20 Method for reducing group recommendation list based on subgroup and social behavior Active CN109101642B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810951568.XA CN109101642B (en) 2018-08-20 2018-08-20 Method for reducing group recommendation list based on subgroup and social behavior

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810951568.XA CN109101642B (en) 2018-08-20 2018-08-20 Method for reducing group recommendation list based on subgroup and social behavior

Publications (2)

Publication Number Publication Date
CN109101642A CN109101642A (en) 2018-12-28
CN109101642B true CN109101642B (en) 2022-06-24

Family

ID=64850411

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810951568.XA Active CN109101642B (en) 2018-08-20 2018-08-20 Method for reducing group recommendation list based on subgroup and social behavior

Country Status (1)

Country Link
CN (1) CN109101642B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110087679A1 (en) * 2009-10-13 2011-04-14 Albert Rosato System and method for cohort based content filtering and display
CN104050258A (en) * 2014-06-15 2014-09-17 中国传媒大学 Group recommendation method based on interest groups
CN106909607A (en) * 2017-01-06 2017-06-30 南京邮电大学 A kind of collaborative filtering group recommending method based on random perturbation technology

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110087679A1 (en) * 2009-10-13 2011-04-14 Albert Rosato System and method for cohort based content filtering and display
CN104050258A (en) * 2014-06-15 2014-09-17 中国传媒大学 Group recommendation method based on interest groups
CN106909607A (en) * 2017-01-06 2017-06-30 南京邮电大学 A kind of collaborative filtering group recommending method based on random perturbation technology

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
点击欺诈群体检测与发现;董亚楠 等;《计算机应用研究》;20150329;第33卷(第6期);第1771-1774、1805页 *

Also Published As

Publication number Publication date
CN109101642A (en) 2018-12-28

Similar Documents

Publication Publication Date Title
CN108733798B (en) Knowledge graph-based personalized recommendation method
KR101009830B1 (en) Compatibility scoring of users in a social network
US20170235788A1 (en) Machine learned query generation on inverted indices
WO2017181612A1 (en) Personalized video recommendation method and device
JP5432243B2 (en) Send and react to media object queries
CN107133277B (en) A kind of tourist attractions recommended method based on Dynamic Theme model and matrix decomposition
US8204878B2 (en) System and method for finding unexpected, but relevant content in an information retrieval system
CN107833117B (en) Bayesian personalized sorting recommendation method considering tag information
CN109947987B (en) Cross collaborative filtering recommendation method
US20150278350A1 (en) Recommendation System With Dual Collaborative Filter Usage Matrix
CN107436914B (en) Recommendation method and device
CN103324636A (en) System and method for recommending friend in social network
CN108446350B (en) Recommendation method based on topic model analysis and long and short interests of user
CN108334592A (en) A kind of personalized recommendation method being combined with collaborative filtering based on content
CN115391576A (en) Cluster-based federal graph learning method and system
KR101693429B1 (en) System for identifying human relationships around users and coaching based on identified human relationships
CN109726747A (en) Recommend the data fusion sort method of platform based on social networks
CN108629010A (en) It is a kind of to recommend method based on the web services of theme and service combination information
CN112102029A (en) Knowledge graph-based long-tail recommendation calculation method
CN109977299A (en) A kind of proposed algorithm of convergence project temperature and expert's coefficient
CN115712780A (en) Information pushing method and device based on cloud computing and big data
CN113873330A (en) Video recommendation method and device, computer equipment and storage medium
Neve et al. ImRec: Learning reciprocal preferences using images
García-Sánchez et al. The Simpsons did it: Exploring the film trope space and its large scale structure
CN109101642B (en) Method for reducing group recommendation list based on subgroup and social behavior

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