CN106528643B - Multi-dimensional comprehensive recommendation method based on social network - Google Patents
Multi-dimensional comprehensive recommendation method based on social network Download PDFInfo
- Publication number
- CN106528643B CN106528643B CN201610894662.7A CN201610894662A CN106528643B CN 106528643 B CN106528643 B CN 106528643B CN 201610894662 A CN201610894662 A CN 201610894662A CN 106528643 B CN106528643 B CN 106528643B
- Authority
- CN
- China
- Prior art keywords
- user
- entity
- item
- representing
- interest
- 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
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
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0255—Targeted advertisements based on user history
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a multi-dimensional comprehensive recommendation method based on a social network, which comprises the following steps: s1: establishing a social network model and a community model; s2: defining user closeness, user entity interest degree and entity similarity, wherein the user closeness reflects the closeness of the relationship between two users, the user entity interest degree reflects the interest degree of the user to the entity, and the entity similarity reflects the consistency of the attributes of the two entities; s3: and performing multi-dimensional entity comprehensive recommendation to the users in the social network according to the user closeness, the user entity interest degree and the entity similarity. Compared with the prior art, the method and the system provided by the invention have the advantages that the influence of the social network business website is evaluated from the user relationship dimension, the entity similarity dimension and the user interest dimension, so that the recommendation of related entities is better carried out, and the method and the system have wide practical application prospects.
Description
Technical Field
The invention relates to the technical field of social networks, in particular to a multi-dimensional comprehensive recommendation method based on a social network.
Background
In the evolution process that the original physical activities of the human society gradually move to the virtual space of the internet, the interaction relationship between people gradually forms a complex network which is staggered with each other, and the social network also becomes one of the important applications of the internet. In recent years, as people have more and more demands on obtaining and transmitting information, social networks are good media for meeting the demands, so that the social networks are rapidly developed. The various social network sites (such as QQ, Facebook, MySpace, etc.) are increasingly developed, and now become the main platform for life, work and entertainment at all ages. Under such circumstances, social network-based merchandise recommendation is widely applied to websites such as e-commerce and online knowledge sharing, for example: the business websites such as the American society and the Taobao are gradually attracting more and more attention of related researchers.
In the traditional commodity recommendation, commodity recommendation is often performed according to some hot sales orders, other related attributes of commodities are ignored, and a huge social network is not considered to be formed among consumers for performing multi-dimensional recommendation. This results in a low recommendation rate and an insignificant recommendation effect. Firstly, when a user searches a certain commodity, ordering operation is performed only if the relevant attributes of the expected commodity are basically matched with the expected objects in the center; secondly, the social network formed by the users can also be utilized, the social network is mainly formed due to the reasons of work, study, consanguinity, entertainment and the like, and users in a common community in the social network are likely to have similar favorite funs. Therefore, a recommendation system is needed to take these factors into consideration and to be sophisticated to achieve a higher recommendation rate.
In order to improve the recommendation effect, commodity recommendation based on the social network is generated. People form a complex social network which is staggered with each other due to the relations of work, entertainment, learning, blooding margin and the like, obviously, a plurality of similar preference characteristics exist among friends possibly, when a certain person in the social network purchases a certain commodity, a commodity candidate set is generated according to the related attributes (such as type, price, quality and sales) of the commodity based on the social network through commodity recommendation of the social network, then a consumer candidate set is generated according to the social network formed by the consumer, and certain commodities meeting certain conditions in the commodity candidate set are recommended to a certain consumer in the consumer candidate set through a related algorithm. Obviously, compared with the traditional commodity recommendation system, the commodity recommendation system based on the social network fuses the social network, is considered more comprehensively, can achieve better recommendation effect and improves the recommendation rate to a certain extent. The invention mainly provides a multi-dimensional comprehensive recommendation method based on a social network aiming at the problems.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a multi-dimensional comprehensive recommendation method based on a social network, which is used for evaluating the influence of a social network business website from a user relationship dimension, an entity similarity dimension and a user interest dimension, so that the recommendation of related entities is better performed, and the method has a wide practical application prospect.
The purpose of the invention can be realized by the following technical scheme:
a multi-dimensional comprehensive recommendation method based on a social network comprises the following steps:
s1: establishing a social network model G (V, E) and a community model C (C)<CV,CE>Where V represents a set of users in social network G, E represents a set of relationships between users, CV represents a set of users in community C who have the same hobbies,CE represents a collection of relationships between users having the same hobbies,namely, users with the same interests and hobbies form a community C;
s2: defining user closeness, user entity interest degree and entity similarity, wherein the user closeness reflects the closeness of the relationship between two users and is related to a comment stability index, an inter-user mutual confidence index, an inter-user interaction frequency comprehensive index, a common neighbor and a similar community index, the user entity interest degree reflects the interest degree of the users to the entity and is related to user interest evidence behaviors, and the entity similarity reflects the consistency of the attributes owned by the two entities and is related to an entity type similarity index, an entity price similarity index, an entity quality similarity index and an entity sales similarity index;
s3: and performing multi-dimensional entity comprehensive recommendation to the users in the social network according to the user closeness, the user entity interest degree and the entity similarity.
The step S3 specifically includes:
302: obtaining entities according to entity similarityEntity candidate recommendation set of Representing entity item' to be recommended and entityThe entity similarity between the two, gamma represents a similarity threshold;
303: acquiring a friend set friend (O) of an initial user according to the friend relationship of the social network of the initial user;
then, a user candidate recommendation set of the initial user is obtained according to the user closeness closeness(vR_userO) represents the user v to be recommendedR_userAnd starting user O, α being the closeness threshold;
304: judging a user v to be recommendedR_userUser entity interest EI (v) of entity item' to be recommendedR_userIf the | item' ∈ R _ item) is larger than the interestingness threshold β, if so, the user v to be recommended is presented with the informationR_userRecommending entity item' to be recommended, if not, not recommending;
305: and updating all the users to be recommended of the recommended entities in the step 304 as initial users, and jumping to the step 303 until all the users in the social network are traversed and recommended, and ending.
The user closeness satisfies the following formula:
wherein, close (v)su,vtu)、cf_L(vsu,vtu)、mf_L(vsu,vtu)、if_L(vsu,vtu) And ncf _ L (v)su,vtu) Respectively representing source users vsuAnd target user vtuUser closeness, comment stability index, inter-user mutual credibility index, inter-user interaction frequency comprehensive index and common neighbor and similar community index.
The comment stability indicator satisfies the following formula:
wherein cf _ L (v)su,vtu) Representing source user vsuAnd target user vtuComment on stability index of (v) COM (v)su,vtu) Representing source user vsuFor target user vtuSet of published reviews, | COM (v)su,vtu) I represents a set of comments COM (v)su,vtu) Total number of comments, com (v)su,vtu)iRepresenting source user vsuFor target user vtuThe value of the ith comment of (c),expressed in the comment set COM (v)su,vtu) Average of all reviews in (1);
the inter-user mutual credibility index satisfies the following formula:
wherein mf _ L (v)su,vtu) Representing source user vsuAnd target user vtuThe inter-user mutual trust degree index between the users,is a ternary array of the elements and is,representing source user vsuFor target user vtuThe average of all the evaluation values is calculated,representing source user vsuAt target user vtuThe proportion of all the issued messages that are forwarded,representing source user vsuAt target user vtuThe proportion of information that is praised among all the published information,is a ternary array of the elements and is,representing a target user vtuFor source user vsuThe average of all the evaluation values is calculated,representing a target user vtuAt source user vsuThe proportion of all the issued messages that are forwarded,representing a target user vtuAt source user vsuThe information ratio for praise is carried out in all the issued information;
the comprehensive index of the interaction frequency among the users meets the following formula:
wherein if _ L (v)su,vtu) Representing source user vsuAnd target user vtuThe comprehensive index of the interaction frequency between users, | Unit | represents the total number of Unit time slices, lkRepresents the k unit time slice ukThe index value of the interaction frequency between the two users,respectively represent the k unit time slice ukEndogenous user vsuFor target user vtuThe number of interactions, the average interaction time length, the average interaction time interval,respectively represent the k unit time slice ukEndogenous user vsuThe maximum interaction times, the longest interaction time length and the shortest interaction time interval of all the neighbor users of the user are determined;
the common neighbor and similar community indexes satisfy the following formula:
wherein, ncf _ L (v)su,vtu) Representing source user vsuAnd target user vtuCommon neighbors between and similar community indices,representing source user vsuThe community in which the user is located is,representing a target user vtuIn the ith communityIn, source user vsuAnd target user vtuThe respective neighbor user sets are respectively CCi(vsu) And CCi(vtu),|CCi| represents CCiNumber of users, | CCi(vsu)∩CCi(vtu) | represents CCi(vsu) And CCi(vtu) Number of users, | CC, within the intersectioni(vsu)∪CCi(vtu) | represents CCi(vsu) And CCi(vtu) The number of users in the union.
The user entity interestingness satisfies the following formula:
in the formula, EI (v | item) represents the user entity interest degree of the user v to the entity item, n represents the total times of interest evidence behaviors appearing when the user faces the entity item,irepresenting the ith interest evidence behavior of the user appearing when facing the entity item, classifying all the interest evidence behaviors appearing in the user history, DjRepresenting the j-th interest evidence category, right (D), which is a direct interest evidencej) Representing a j-th interest evidence category DjWeight of (1), Pv(Dj| item) represents for user v, when entity item appears, the interest evidence category DjThe probability of occurrence.
The following relationships exist between the weights of different interest evidence categories:
in the formula (I), the compound is shown in the specification,right(Dq) Representing a q-th type interest evidence category DqID represents the set of all interest evidence categories that the user has historically appeared, Dq,Dj∈ ID, | ID | represents the total number of interest evidence categories, p (D)q) Representing an interest evidence category DqProbability of occurrence in user past interest evidence behavior, L (D)q) Show and q type interest evidence category DqNumber of linked interest evidence categories, Link (D)j) Show the j-th type interest evidence category DjSet of linked interest evidence categories, Dq→DjRepresents DjIs formed by DqCaused by, i.e. DqAnd DjAnd linking.
The entity similarity satisfies the following formula:
in the formula, itemaAnd itembTo represent two different entities, sim (item)a,itemb)、sim_type(itema,itemb)、sim_price(itema,itemb)、sim_sale(itema,itemb) Respectively representing entity itemaWith entity itembEntity similarity, entity type similarity index, entity price similarity index, entity sales volume similarity index, sim _ quality (item)a) Representing entity itemaThe entity quality similarity index of (1).
The entity type similarity index is calculated based on the entity directory classification tree, and meets the following formula:
in the formula, sim _ type (item)a,itemb) Representing entity itemaWith entity itembAn entity type similarity index of caRepresenting entity itemaHierarchy within the entity directory classification tree, cbRepresenting entity itembHierarchy within the entity directory classification tree, distance (item)a,itemb) Representing entity item within entity directory classification treeaWith entity itembThe path length between;
the entity price similarity index meets the following formula:
in the formula, sim _ price (item)a,itemb) Representing entity itemaWith entity itembPrice similarity index of entityaRepresenting entity itemaPrice of (price)bRepresenting entity itembThe price of (c);
the entity quality similarity index satisfies the following formula:
in the formula, sim _ quality (item)a) Representing entity itemaThe index of similarity of mass of entities, scoreaRepresenting entity itemaScore of entity quality of (1), scoremaxRepresents the highest score of all entity quality scores;
the entity sales volume similarity index satisfies the following formula:
in the formula, sim _ sale (item)a,itemb) Representing entity itemaWith entity itembThe entity sales similarity index of (1), saleaRepresenting entity itemaTotal sales of (1), salebRepresenting entity itembTotal sales of (a).
Compared with the prior art, the invention has the following advantages:
1. different from the traditional technology of recommending based on a single standard, the method develops entity recommendation in the social network based on multiple dimensions: selecting users with close relation for recommendation according to the user closeness as a standard from the user relation dimension; starting from the entity similarity dimension, selecting entities with high similarity for recommendation according to the entity similarity serving as a standard; starting from the user interest dimension, selecting an entity with high user interest for recommendation according to the user entity interest as a standard, and closely fusing the entity recommendation with the internal information of the social network, so that the entity recommendation is more comprehensive, a better recommendation effect can be achieved, and the method has a wide practical application prospect;
2. the recommendation process of the invention takes the initial user who purchases the entity as the initial, establishes the recommendation judgment condition, and gradually recommends the entity to the adjacent users of the initial user, thereby achieving the purpose of recommending the entity of all users, having wide recommendation range, having no omission, achieving the purpose of real-time recommendation, and having accurate selection of the recommended object and the recommended entity.
3. In the user closeness index calculation process, the closeness of the user relationship is more comprehensively evaluated by calculating the four aspects of stable user mutual evaluation, user mutual trust degree, user interaction frequency, common user neighbors and communities;
4. in the user entity interestingness calculation process provided by the invention, the calculation is carried out by adopting the related data of the direct behavior evidence of the user, and a weight calculation method based on association is established among the user interest evidences, so that the calculation accuracy of the user interestingness is ensured;
5. in the calculation of the entity similarity provided by the invention, four aspects are considered: and the entity type similarity, the entity price similarity, the entity quality similarity and the entity sales volume similarity are calculated more comprehensively.
6. The method is based on the principle that the higher the weight of the interest evidence category is, the relationship existing between the weights of different interest evidence categories is given, after the initial value of the weight value is randomly given, the final weight value can be calculated according to the iterative operation of a relationship formula, compared with the method of directly setting the initial value, the weight of the interest evidence category used by the method can reflect the occupied component of the interest evidence behavior more accurately, and the value of the user entity interest degree is more accurate.
Drawings
FIG. 1 is a schematic diagram of the method of the present invention;
FIG. 2 is a simplified flow chart of the recommendation method of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
The present invention gives the following definitions for social networks:
define 1 social network model: the social network G may be formatted as a graph, with G ═ V, (E), V representing a set of users in the social network, and ViRepresenting individual users, viThe middle subscript i represents the number of the user, and E is used for describing the set of relationships among the users;
defining 2 a community model: the community C is composed of users with the same interest and hobbies and is a sub-graph of the social network, and C is equal to<CV,CE>When a user belongs to a community, it is expressed as:CV represents a set of users in community C who have the same interest and hobbies, and the relationship among community users is represented as:CE represents a set of relationships among users having the same interests;
define 3 user affinity: reflecting how close the relationship between the two users is. Closeness among computing users (denoted as: closeness (v)su,vtu) Wherein v issuRepresents the source user, vtuRepresenting the target user), the present invention primarily considers four aspects: stability of comments, inter-user mutual reliability, inter-user interaction frequency and inter-userCommon neighbors and similar communities;
defining 4 user entity interest degrees: entity interest EI (v) held by user to entityi|itemk) Reflecting user's item to a certain entitykDegree of interest, itemkThe middle subscript k represents the number of the entity, and in the present invention, the interest of the entity is expressed by the direct behavior of the user or other direct witness evidence, such as: evaluation, browsing time, praise, forwarding and the like.
Definition 5 entity similarity: entity similarity represents the degree of agreement between the attributes possessed by two entities. The similarity between two entities is denoted sim (item)a,itemb) Wherein itema、itembRepresenting two different entities. The entity similarity is developed mainly based on four attributes represented in the attribute set, and the entity attributes are represented as { type, price, quality, sale }, that is, the entity attributes include the type, price, quality, sale, etc. to which the entity belongs.
A multi-dimensional comprehensive recommendation method based on a social network comprises the following steps:
s1: establishing a social network model G (V, E) and a community model C (CV, CE), and defining concepts of user closeness, user entity interest degree and entity similarity;
s2: evaluating according to the definitions of user closeness, user entity interest and entity similarity;
the user closeness reflects the closeness of the relationship between two users and is related to a comment stability index, an inter-user mutual credibility index, an inter-user interaction frequency comprehensive index, a common neighbor and similar community index, the user entity interest reflects the interest degree of the user to the entity and is related to user interest evidence behaviors, and the entity similarity reflects the consistency of the attributes owned by the two entities and is related to an entity type similarity index, an entity price similarity index, an entity quality similarity index and an entity sales similarity index;
s3: and performing multi-dimensional entity comprehensive recommendation to the users in the social network according to the user closeness, the user entity interest degree and the entity similarity.
In step S2, (1) calculating user closeness
1) Review stability index
And the comment stability index reflects the fluctuation state of the comment value given by the source user to the target user in past comments. In the invention, the higher the comment stability value is, the more consistent the comment of the source user to the target user is reflected; conversely, the lower the comment stability value is, the more unstable the comment value given by the source user to the target user is, that is, the difference is larger. In the present invention, the set COM (v)su,vtu) To represent a source user vsuFor target user vtuA set of published reviews. Assume in the set COM (v)su,vtu) The average of all comments in (1) isThen, the stability index cf _ L (v) of the comment of the source user to the target usersu,vtu) The value of (d) can be calculated by the following equation:
wherein, | COM (v)su,vtu) I represents a set of comments COM (v)su,vtu) Total number of comments, com (v)su,vtu)iRepresenting source user vsuFor target user vtuThe value of the ith comment of (1).
For example, if source user vsuFor target user vtuThe evaluation was collected as COM (v)su,vtu)={com(vsu,vtu)1=0.9,com(vsu,vtu)2=1.0,com(vsu,vtu)30.8 }; then
2) Inter-user mutual confidence index
The inter-user mutual trust degree represents the matching degree of the two users to the mutual trust degree, namely the more consistent the mutual trust degree between the users is, the higher the mutual relationship compactness is. In the invention, the mutual credibility index between users is embodied by three aspects: review mutual confidence (reliability of comment), mutual forwarding (forwarding), and mutual approval (approving). In the present invention, for source user vsuAnd target user vtuIn other words, the values between two directly related users using the above three aspects are expressed as a ternary arrayAnd
Then the source user vsuAnd target user vtuInter-user mutual confidence index mf _ L (v) betweensu,vtu) The calculation formula of (a) is as follows:
in the formula (I), the compound is shown in the specification,to representThe (c) th element of (a),to representThe ith element in (1).
For example, if source user vsuAnd target user vtuThere are three mutual confidence values:
then:
3) inter-user interaction frequency indicator
The interaction frequency index in the invention is calculated by using three aspects: the number of interactions per unit time, the average length of time of the interactions and the average interaction time interval.
For source user vsuLet us assume the kth unit time slice ukThe maximum value of the number of interactions with all the neighbor users isThe longest duration of interaction among all the past interaction records isThe shortest interaction time interval isAt the k unit time slice ukInner, source user vsuFor target user vtuThe number of interactions isSource user vsuFor target user v in its pasttuThe average time length in all interactions of (1) isAverage interaction time interval of
Then the source user vsuAnd target user vtuInter-user interaction frequency comprehensive index if _ L (v) therebetweensu,vtu) The following formula is satisfied:
where | Unit | represents the total number of past time slices per Unit, lkRepresents the k unit time slice ukAnd (4) interacting the frequency index values between the two users.
4) Common neighbors and similar community metrics.
In the invention, the user intimacy is considered to be reduced along with the increase of the community size. The smaller the community scale is, the more frequent the user interaction is, and the higher the intimacy is; conversely, the larger the community size, the less user interaction and the lower the intimacy. Likewise, the number of common neighbors also reflects the affinity between two users, i.e., the more common neighbors there are, the higher the affinity between users.
Suppose thatAndrespectively representing source users vsuAnd target user vtuThe community in which it is located. In each common communityMiddle (CC)iMiddle subscript i denotes the number of the common community), source user vsuAnd target user vtuWith their neighbor sets, respectively CCi(vsu) And CCi(vtu). In the present invention, | CCiI denotes CC in CommunityiNumber of members, | CCi(vsu)∩CCi(vtu) I is represented in the set CCi(vsu)∩CCi(vtu) Number of neighbors in, | CCi(vsu)∪CCi(vtu) I is represented in the set CCi(vsu)∪CCi(vtu) The number of neighbors in (1). Source user vsuAnd target user vtuCommon neighbors between and similar community indices ncf _ L (v)su,vtu) The calculation expression of (a) is as follows:
finally, the calculation source user v can be obtained according to the four indexessuAnd target user vtuCloseness of users between closenss (v)su,vtu) The expression is as follows:
(2) calculating user entity interest degree of user for entity
User entity interestingness is evaluated by user direct behavior or other direct witness evidence. The invention utilizes direct interest evidence (behavior) to calculate entity interest degree. The user direct interest evidence comprises: forwarding (forwarding), praise (serving), attention (following), comments (comments), and the like. The user entity interest degree calculation principle in the invention is as follows:
for oneThe number of individual entity items is,irepresenting the ith interest evidence behavior of the user appearing when facing the entity item, and assuming that the interest evidence behavior of the user repeatedly appears n times in the past for one entity item, the interest evidence behavior set is represented as phi ═ f1,2,...,i,...}: meanwhile, classifying previous various interest evidence behaviors of the user v, and recording as: d1,D2,...,Dj,...,DmM is the total number of interest evidence categories, DjRepresents the j-th interest evidence category, which is the type of direct interest evidence, and P (D)j) Representing an interest evidence category DjFrequency of occurrence in all evidence of interest behaviors. And P isv(Dj| item) represents for user v, when entity item appears, the interest evidence category DjThe probability of occurrence. Each type of interest evidence category DjIs right (D)j),right(Dj)∈[0,1]. Then, the calculation expression of the user entity interest level EI (v | item) of the user v for the entity item is as follows:
in the formula, Pv(item|Dj) For user v, when the interest evidence category DjWhen the probability is present, the probability corresponds to the entity item, that is, the user v is in the interest evidence category D in all past interest evidencesjExpressed in the ratio of one entity item, P (D)j)、Pv(item|Dj) The observation value can be obtained directly through numerical statistics.
For weight right (D)j) The calculation method of (2) is as follows: assuming that the interest evidence y is caused by other interest evidence x, it can be written as: x → y. The invention uses ID to represent the set of all interest evidence categories appearing in user history, wherein ID is { D }1,D2,...,Dj,...,DmAnd f and the firstj interest evidence categories DjConnected interest evidence set Link (D)j) Can be expressed as:
in the formula, DqRepresenting the q-th interest evidence category, Dq,Dj∈ID,DqIs also AND DjThe linked (connected) interest evidence category (which can be direct evidence or indirect evidence), i.e. DjIs formed by DqCaused by (with D)q→DjRepresentation).
Then, the weight right (D) is calculatedj) The equation of (1) is as follows:
wherein right (D)q) Representing a q-th type interest evidence category DqRepresents the total number of interest evidence categories, | ID | ═ m, p (D)q) Representing an interest evidence category DqProbability of occurrence in user past interest evidence behavior, L (D)q) Show and q type interest evidence category DqThe number of linked interest evidence categories.
The right (D)j) The calculation formula (2) represents the weight relationship among the related interest evidence, and the operation principle is that the higher the weight of the interest evidence category, the higher the weight of the associated interest evidence category. The initial value of the weight value can be randomly given, but the final weight value can be calculated after iterative operation according to the formula, so that the error caused by direct setting of the weight value is reduced.
(3) Calculating entity similarity between entities
1) Entity type similarity index
The method carries out type similarity index calculation based on the entity directory classification tree. Classifying different entities according to the directory to form an entity directory classification tree with a tree structure, namely representing the entity directory as a tree model; at the same time, the invention introducesDirectory tree hierarchy, order ciRepresenting the i-th level of directory in the entity directory classification tree, e.g. the first level of directory of the entity directory classification tree, i.e. denoted c1。
Suppose there are two different entities, denoted item respectivelyaAnd itemb. Distance (item) for the present inventiona,itemb) To represent entity itemaThe directory hierarchy and entity itembPath distance between directory levels where it is located. The invention calculates the type similarity index as follows:
1. if the directory hierarchy of two entities belongs to the same layer number, the entity itemaWith entity itembEntity type similarity index sim _ type (item) of (1)a,itemb) The calculation formula of (a) is as follows:
in the formula, caRepresenting entity itemaHierarchy within the entity directory classification tree, cbRepresenting entity itembA hierarchy within the entity directory classification tree;
2. if the directories of the two entities are not in the same level, the similarity of the types of the two entities is as follows:
2) entity price similarity index
In the invention, the entity price similarity index is calculated as:
in the formula, sim _ price (item)a,itemb) Representing entity itemaWith entity itembPrice similarity index of entityaRepresenting entity itemaPrice of (price)bRepresenting entity itembThe price of (c);
3) entity quality similarity index
In the invention, the entity quality similarity index is used for calculating the quality similarity between a certain entity and the current candidate recommendation set so as to judge whether the entity meets the recommendation requirement. The highest score of all evaluations, assuming entity quality, is scored as scoremaxThe average of all entity quality scores in the current candidate recommendation set is designated scoreaveThen entity itemaOverall quality similarity index sim _ quality (item) with all currently recommendable entitiesa) The calculation formula of (a) is as follows:
in the formula, scoreaRepresenting entity itemaAn entity quality score of (a); if it satisfiesThen the entity itemaSimilar in quality index to the current candidate recommendation set.
4) Similarity index of entity sales
Entity itemaWith entity itembThe entity sales volume similarity index sim _ sample (item)a,itemb) The calculation formula of (a) is as follows:
in the formula, saleaRepresenting entity itemaTotal sales of (1), salebRepresenting entity itembTotal sales of (a).
Finally, the entity similarity value can be calculated by the four indexes, and the calculation formula is as follows:
if sim (item)a,itemb) At least α (α is a tightness threshold),then the entity itemaIncorporation of entity itembα is the threshold value of similarity of entities.
As shown in fig. 2, the step S3 specifically recommends the following steps:
1) when a user purchases an entityWhen the user is initially marked as an initial user O, a friend set of the user is marked as friend (O) according to the friend relationship of the user in the social network, and friends in the set meet the condition' vi∈friend(O)∩closeness(vi,O)≥α∩viUnpurchased entityWhen the method is adopted, the following steps are carried out:
① for each entity item present in all social networksk(itemkSubscript k denotes the kth entity in the entity set ITEM), if satisfiedThen item will bekPutting the R _ item into a set, wherein elements in the R _ item are represented by tiem'; the set of all entities fulfilling the above conditions is called the entity candidate recommendation set R _ item (purchased by the originating user O))。
2) For all vi∈ friend (O), the following work is done:
① calculation of closeness (v)iO) and whether the condition close (v) is satisfied is determinediO) is greater than threshold α if satisfied, viPutting the user candidate recommendation set R _ user into a set of user candidate recommendations, wherein the elements in the set R _ user are represented by vR_userRepresents; if not, directly ignoring vi。
② randomly selecting a to-be-recommended entity tiem' from the R _ items, and calculating EI (v)R_user| tiem' ∈ R _ item), and judgesBreaking EI (v)R_userIf | tiem' ∈ R _ item) is satisfied greater than a threshold value βR_userRecommending entity tiem 'and deleting entity tiem' from R _ item; otherwise, no recommendation is made and the tiem is deleted directly.
Thirdly, repeating the first step and the second step in the step 2) until the entity in the R _ item is empty;
and fourthly, repeating the step one and the step two until all the users of the R _ user are recommended.
3) Taking all users with recommended entities in the R _ user as initial users, respectively aiming at all the initial users in the R _ user, continuously recommending the entity candidate recommendation set R _ item generated in the step 1) to friends of a new initial user, and repeating the step 2); until all users (nodes) in the social network are traversed through the recommendations.
FIG. 1 is a framework of a recommendation algorithm of the present invention, wherein an arrow ① in FIG. 1 indicates that a social network is generated based on a user candidate recommendation set, an arrow ② generates a new user candidate recommendation set based on the generated social network, an arrow ③ indicates that a plurality of user candidate recommendation sets form a large new social network, an arrow ④ indicates that the recommendation algorithm and the social network are established on the basis of each other, an arrow ⑤ indicates that a final user candidate recommendation set is generated based on the generated large social network, an arrow ⑥ indicates that entities in the entity candidate set generated by the recommendation algorithm are recommended to users in the user candidate recommendation set, and an arrow ⑦ indicates that an entity candidate set is obtained based on the recommendation algorithm, FIG. 1 shows that the recommendation algorithm is core and performs comprehensive recommendation of three dimensions of the entities based on the social network, the three dimensions are respectively an entity similarity, a user closeness, and a user interest degree of the entitieskAnd entitiesIf the similarity value is greater than the threshold value gamma, the entity item is determinedkPutting the entity candidate set R _ item into the entity candidate set R _ item, otherwise, not putting the entity candidate set R _ item into the entity candidate set R _ item; second, compute the userviCloseness to user O, if the value of closeness is greater than threshold α, then user v is assignediPutting the user candidate set R _ user; then, calculating the user v in the user candidate set R _ userR_userFor the interest degree of the entity item' in the entity candidate set R _ item, if the value of the interest degree is larger than the threshold value β, the entity item is processedkRecommending to user vR_userOtherwise, it is not recommended.
Therefore, the recommendation method of the invention can initiate comprehensive recommendation timely along the social network around the purchased entity and the user of the purchased entity after the user purchases the entity, and the comprehensive recommendation relates to three dimensions: the user closeness index enables the closeness of the user relationship to be evaluated more comprehensively, the user entity interest degree ensures the calculation accuracy of the user interest degree, the entity similarity enables the entity similarity degree to be calculated more comprehensively, evaluation on influence of social network business websites such as Taobao and Mei is achieved, recommendation of related entities is better carried out, and the method has practical application prospect.
Claims (7)
1. A multi-dimensional comprehensive recommendation method based on a social network is characterized by comprising the following steps:
s1: establishing a social network model G (V, E) and a community model C (C)<CV,CE>Where V represents a set of users in social network G, E represents a set of relationships between users, CV represents a set of users in community C who have the same hobbies,CE represents a collection of relationships between users having the same hobbies,
s2: defining user closeness, user entity interest degree and entity similarity, wherein the user closeness reflects the closeness of the relationship between two users and is related to a comment stability index, an inter-user mutual confidence index, an inter-user interaction frequency comprehensive index, a common neighbor and a similar community index, the comment stability index reflects the fluctuation state of a comment value given by a source user to a target user in past comments, the user entity interest degree reflects the interest degree held by the user to the entity and is related to user interest evidence behaviors, and the entity similarity reflects the consistency of attributes owned by the two entities and is related to an entity type similarity index, an entity price similarity index, an entity quality similarity index and an entity sales similarity index;
calculating the entity interestingness by using direct interest evidence;
s3: according to the user closeness, the user entity interest degree and the entity similarity, carrying out multi-dimensional entity comprehensive recommendation on the users in the social network;
the step S3 specifically includes:
301: when a user purchases an entityWhen, initially mark the user as an originating user O,∈ entity set ITEM;
302: obtaining entities according to entity similarityEntity candidate recommendation set of Representing entity item' to be recommended and entityThe entity similarity between the two, gamma represents a similarity threshold;
303: acquiring a friend set friend (O) of an initial user according to the friend relationship of the social network of the initial user;
then, a user candidate recommendation set of the initial user is obtained according to the user closeness closeness(vR_userO) represents the user v to be recommendedR_userAnd starting user O, α being the closeness threshold;
304: judging a user v to be recommendedR_userUser entity interest EI (v) of entity item' to be recommendedR_userIf the | item' ∈ R _ item) is larger than the interestingness threshold β, if so, the user v to be recommended is presented with the informationR_userRecommending entity item' to be recommended, if not, not recommending;
305: and updating all the users to be recommended of the recommended entities in the step 304 as initial users, and jumping to the step 303 until all the users in the social network are traversed and recommended, and ending.
2. The social network-based multi-dimensional comprehensive recommendation method according to claim 1, wherein the user closeness satisfies the following formula:
wherein, close (v)su,vtu)、cf_L(vsu,vtu)、mf_L(vsu,vtu)、if_L(vsu,vtu) And ncf _ L (v)su,vtu) Respectively representing source users vsuAnd target user vtuUser closeness, comment stability index, inter-user mutual credibility index and inter-user interaction frequencyRate integration indicators and co-neighborhood and similar community indicators.
3. The social network-based multidimensional comprehensive recommendation method according to claim 1, wherein the comment stability index satisfies the following formula:
wherein cf _ L (v)su,vtu) Representing source user vsuAnd target user vtuComment on stability index of (v) COM (v)su,vtu) Representing source user vsuFor target user vtuSet of published reviews, | COM (v)su,vtu) I represents a set of comments COM (v)su,vtu) Total number of comments, com (v)su,vtu)iRepresenting source user vsuFor target user vtuThe value of the ith comment of (c),expressed in the comment set COM (v)su,vtu) Average of all reviews in (1);
the inter-user mutual credibility index satisfies the following formula:
wherein mf _ L (v)su,vtu) Representing source user vsuAnd target user vtuThe inter-user mutual trust degree index between the users,is a ternary array of the elements and is,representing source user vsuFor target user vtuAll ofThe average value of the evaluation values is,representing source user vsuAt target user vtuThe proportion of all the issued messages that are forwarded,representing source user vsuAt target user vtuThe proportion of information that is praised among all the published information,is a ternary array of the elements and is,representing a target user vtuFor source user vsuThe average of all the evaluation values is calculated,representing a target user vtuAt source user vsuThe proportion of all the issued messages that are forwarded,representing a target user vtuAt source user vsuThe information ratio for praise is carried out in all the issued information;
the comprehensive index of the interaction frequency among the users meets the following formula:
wherein if _ L (v)su,vtu) Representing source user vsuAnd target user vtuBetween usersInteraction frequency synthesis index, | Unit | represents the total number of Unit time slices, lkRepresents the k unit time slice ukThe index value of the interaction frequency between the two users,respectively represent the k unit time slice ukEndogenous user vsuFor target user vtuThe number of interactions, the average interaction time length, the average interaction time interval,respectively represent the k unit time slice ukEndogenous user vsuThe maximum interaction times, the longest interaction time length and the shortest interaction time interval of all the neighbor users of the user are determined;
the common neighbor and similar community indexes satisfy the following formula:
wherein, ncf _ L (v)su,vtu) Representing source user vsuAnd target user vtuCommon neighbors between and similar community indices,representing source user vsuThe community in which the user is located is,representing a target user vtuIn the ith communityIn, source user vsuAnd target user vtuThe respective neighbor user sets are respectively CCi(vsu) And CCi(vtu),|CCi| represents CCiNumber of users, | CCi(vsu)∩CCi(vtu) | represents CCi(vsu) And CCi(vtu) Number of users within the intersection, | CCi (v)su)∪CCi(vtu) | represents CCi(vsu) And CCi(vtu) The number of users in the union.
4. The social network-based multi-dimensional comprehensive recommendation method according to claim 1, wherein the user entity interestingness satisfies the following formula:
in the formula, EI (v | item) represents the user entity interest degree of the user v to the entity item, n represents the total times of interest evidence behaviors appearing when the user faces the entity item,irepresenting the ith interest evidence behavior of the user appearing when facing the entity item, classifying all the interest evidence behaviors appearing in the user history, DjRepresenting the j-th interest evidence category, right (D), which is a direct interest evidencej) Representing a j-th interest evidence category DjWeight of (1), Pv(Dj| item) represents for user v, when entity item appears, the interest evidence category DjThe probability of occurrence.
5. The social network-based multi-dimensional comprehensive recommendation method according to claim 4, wherein the following relationships exist among the weights of different interest evidence categories:
in the formula, right (D)q) Representing a q-th type interest evidence category DqWeight of (1), ID representsSet of all interest evidence categories, D, of user historical appearanceq,Dj∈ ID, | ID | represents the total number of interest evidence categories, p (D)q) Representing an interest evidence category DqProbability of occurrence in user past interest evidence behavior, L (D)q) Show and q type interest evidence category DqNumber of linked interest evidence categories, Link (D)j) Show the j-th type interest evidence category DjSet of linked interest evidence categories, Dq→DjRepresents DjIs formed by DqCaused by, i.e. DqAnd DjAnd linking.
6. The social network-based multidimensional comprehensive recommendation method according to claim 1, wherein the entity similarity satisfies the following formula:
in the formula, itemaAnd itembTo represent two different entities, sim (item)a,itemb)、sim_type(itema,itemb)、sim_price(itema,itemb)、sim_sale(itema,itemb) Respectively representing entity itemaWith entity itembEntity similarity, entity type similarity index, entity price similarity index, entity sales volume similarity index, sim _ quality (item)a) Representing entity itemaThe entity quality similarity index of (1).
7. The social network-based multidimensional comprehensive recommendation method according to claim 1, wherein the entity type similarity index is calculated based on an entity directory classification tree, and satisfies the following formula:
in the formula (I), the compound is shown in the specification,sim_type(itema,itemb) Representing entity itemaWith entity itembAn entity type similarity index of caRepresenting entity itemaHierarchy within the entity directory classification tree, cbRepresenting entity itembHierarchy within the entity directory classification tree, distance (item)a,itemb) Representing entity item within entity directory classification treeaWith entity itembThe path length between;
the entity price similarity index meets the following formula:
in the formula, sim _ price (item)a,itemb) Representing entity itemaWith entity itembPrice similarity index of entityaRepresenting entity itemaPrice of (price)bRepresenting entity itembThe price of (c);
the entity quality similarity index satisfies the following formula:
in the formula, sim _ quality (item)a) Representing entity itemaThe index of similarity of mass of entities, scoreaRepresenting entity itemaScore of entity quality of (1), scoremaxRepresents the highest score of all entity quality scores;
the entity sales volume similarity index satisfies the following formula:
in the formula, sim _ sale (item)a,itemb) Representing entity itemaWith entity itembThe entity sales similarity index of (1), saleaRepresenting entity itemaTotal sales of (1), salebTo representEntity itembTotal sales of (a).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610894662.7A CN106528643B (en) | 2016-10-13 | 2016-10-13 | Multi-dimensional comprehensive recommendation method based on social network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610894662.7A CN106528643B (en) | 2016-10-13 | 2016-10-13 | Multi-dimensional comprehensive recommendation method based on social network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106528643A CN106528643A (en) | 2017-03-22 |
CN106528643B true CN106528643B (en) | 2020-10-16 |
Family
ID=58331863
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610894662.7A Active CN106528643B (en) | 2016-10-13 | 2016-10-13 | Multi-dimensional comprehensive recommendation method based on social network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106528643B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107436914B (en) * | 2017-06-06 | 2020-06-23 | 北京星选科技有限公司 | Recommendation method and device |
CN109062914B (en) * | 2017-06-12 | 2020-10-23 | 东软集团股份有限公司 | User recommendation method and device, storage medium and server |
CN107818165A (en) * | 2017-10-31 | 2018-03-20 | 平安科技(深圳)有限公司 | Marketing client screening technique, electronic installation and storage medium based on tag library |
CN108184207B (en) * | 2017-12-20 | 2020-06-30 | 中国移动通信集团江苏有限公司 | Method, device, equipment and medium for determining resident community of communication user |
CN108376371A (en) * | 2018-02-02 | 2018-08-07 | 众安信息技术服务有限公司 | A kind of internet insurance marketing method and system based on social networks |
CN109727059A (en) * | 2018-11-20 | 2019-05-07 | 北京云和互动信息技术有限公司 | A kind of evaluation method and system based on big data |
CN110096614B (en) * | 2019-04-12 | 2022-09-20 | 腾讯科技(深圳)有限公司 | Information recommendation method and device and electronic equipment |
CN112446777B (en) * | 2019-09-03 | 2023-11-17 | 腾讯科技(深圳)有限公司 | Credit evaluation method, device, equipment and storage medium |
CN110807052B (en) * | 2019-11-05 | 2022-08-02 | 佳都科技集团股份有限公司 | User group classification method, device, equipment and storage medium |
CN111241420B (en) * | 2020-01-10 | 2020-11-10 | 云境商务智能研究院南京有限公司 | Recommendation method based on social network information diffusion perception |
CN112163169A (en) * | 2020-09-29 | 2021-01-01 | 海南大学 | Multi-mode user emotion analysis method based on knowledge graph |
CN114386764B (en) * | 2021-12-11 | 2022-12-16 | 上海师范大学 | GRU and R-GCN based OJ platform topic sequence recommendation method |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8095432B1 (en) * | 2009-01-30 | 2012-01-10 | Intuit Inc. | Recommendation engine for social networks |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130325587A1 (en) * | 2009-01-21 | 2013-12-05 | Truaxis, Inc. | System and method for managing campaign effectiveness by a merchant |
US20110047013A1 (en) * | 2009-05-21 | 2011-02-24 | Mckenzie Iii James O | Merchandising amplification via social networking system and method |
CN102426686A (en) * | 2011-09-29 | 2012-04-25 | 南京大学 | Internet information product recommending method based on matrix decomposition |
CN103514304B (en) * | 2013-10-29 | 2017-01-18 | 海南大学 | Project recommendation method and device |
CN103995823A (en) * | 2014-03-25 | 2014-08-20 | 南京邮电大学 | Information recommending method based on social network |
CN104268171B (en) * | 2014-09-11 | 2017-09-19 | 东北大学 | The social network friend recommendation system and method trusted based on the social activity of activity phase Sihe |
-
2016
- 2016-10-13 CN CN201610894662.7A patent/CN106528643B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8095432B1 (en) * | 2009-01-30 | 2012-01-10 | Intuit Inc. | Recommendation engine for social networks |
Non-Patent Citations (1)
Title |
---|
"一种社交网络群组间信息推荐的有效方法";黄震华 等;《电子学报》;20150630;第43卷(第6期);1090-1093 * |
Also Published As
Publication number | Publication date |
---|---|
CN106528643A (en) | 2017-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106528643B (en) | Multi-dimensional comprehensive recommendation method based on social network | |
US11659050B2 (en) | Discovering signature of electronic social networks | |
Yang et al. | Friend or frenemy? Predicting signed ties in social networks | |
CN107833117B (en) | Bayesian personalized sorting recommendation method considering tag information | |
KR100961782B1 (en) | Apparatus and method for presenting personalized goods information based on artificial intelligence, and recording medium thereof | |
KR100961783B1 (en) | Apparatus and method for presenting personalized goods and vendors based on artificial intelligence, and recording medium thereof | |
CN111125453B (en) | Opinion leader role identification method in social network based on subgraph isomorphism and storage medium | |
US20150220972A1 (en) | Management Of The Display Of Online Ad Content Consistent With One Or More Performance Objectives For A Webpage And/Or Website | |
CN109918562B (en) | Recommendation method based on user community and scoring combined community | |
TW201501059A (en) | Method and system for recommending information | |
US11107093B2 (en) | Distributed node cluster for establishing a digital touchpoint across multiple devices on a digital communications network | |
US20140188994A1 (en) | Social Neighborhood Determination | |
CN105761154A (en) | Socialized recommendation method and device | |
CN111666413B (en) | Commodity comment recommendation method based on reviewer reliability regression prediction | |
KR101459537B1 (en) | Method and system for Social Recommendation with Link Prediction | |
CN114723535A (en) | Supply chain and knowledge graph-based item recommendation method, equipment and medium | |
Zhang et al. | Multi-view dynamic heterogeneous information network embedding | |
Ma | Recommendation of sustainable economic learning course based on text vector model and support vector machine | |
Zhou et al. | Model and implementation of e-commerce recommendation system based on user clustering | |
Zhang | Research on collaborative filtering recommendation algorithm based on social network | |
CN114861079A (en) | Collaborative filtering recommendation method and system fusing commodity features | |
Goldstein et al. | Are we there yet? Analyzing progress in the conversion funnel using the diversity of searched products | |
Zhang et al. | Multi-dimensional recommendation scheme for social networks considering a user relationship strength perspective | |
CN110751180B (en) | Spurious comment group division method based on spectral clustering | |
CN110347923B (en) | Traceable fast fission type user portrait construction method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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 |