CN108960897A - A kind of various dimensions user collaborative filtered recommendation method of combination correlation rule - Google Patents
A kind of various dimensions user collaborative filtered recommendation method of combination correlation rule Download PDFInfo
- Publication number
- CN108960897A CN108960897A CN201810584551.5A CN201810584551A CN108960897A CN 108960897 A CN108960897 A CN 108960897A CN 201810584551 A CN201810584551 A CN 201810584551A CN 108960897 A CN108960897 A CN 108960897A
- Authority
- CN
- China
- Prior art keywords
- user
- commodity
- context
- target user
- scoring
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Economics (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of various dimensions user collaborative filtered recommendation methods of combination correlation rule.Context dimension has been fused in user collaborative filter algorithm by the present invention first, calculates user's similarity and commodity projection scoring under the influence of context factors.The frequent item set for not including in target user's commodity purchasing record is excavated with FP-growth association rule algorithm simultaneously, Result is fused in recommendation list.The present invention considers the influence of context various dimensions factor, and the accuracy of recommender system can be improved.
Description
Technical field
The present invention relates to proposed algorithm technical field, specifically a kind of various dimensions user collaborative of combination correlation rule
Filtered recommendation method.
Background technique
With coming for web2.0 epoch, internet enters the epoch of data explosion.The data of magnanimity to user with
Also to have flooded valid data while enriching.And recommender system can analyze the history buying behavior and other overall situations of user
Information recommends him may interested commodity for user.Focus of the user between quickly positioning in magnanimity commodity is helped, is mentioned
High information matches efficiency, while but also businessman is more targeted when advertisement is launched, help to realize user and
Win-win between businessman.
The data traffic that the development of the communication technology carries mobile terminal is increasing, and user can be on mobile terminal
Complete many purchases and article housing choice behavior.Each purchase of user has its specific context, such as time, place, week
Enclose same administrative staff etc..And these contextual informations also can largely influence the decision of user.And mobile device can obtain
These contextual informations, such as current time are got, it is same whether the current location of the available user of GPS positioning and its surrounding have
Thing, friend, household exist.
Summary of the invention
Aiming at the defects existing in the prior art, the technical problem to be solved in the present invention is to provide a kind of combinations
The various dimensions user collaborative filtered recommendation method of correlation rule.
Present invention technical solution used for the above purpose is: a kind of various dimensions user association of combination correlation rule
Same filtered recommendation method, comprising the following steps:
The purchaser record of user is collected, the contextual information when purchaser record user buys, generating includes context
User-rating matrix of information；
Calculate user's similarity；
Determine the set of neighbor user；
Determine context similarity；
Predict commodity scoring；
Calculate the relevant frequent item set of target user's history purchase commodity；
Generate recommendation list.
Whether the contextual information includes the time, place, weather, same to administrative staff, temperature, humidity, season, is that section is false
Day, whether be weekend, daytime or night, whether be any several combination in the meal time.
Calculating user's similarity, specifically:
Wherein, sa,u,cIndicate similarity value of the target user a and user u under the conditions of contextual information c, IaIndicate target
The item destination aggregation (mda) that user a scored, IuIndicate the commodity set that user u was contacted, i indicates that target user a and user u is total
With the commodity to score, ra,iIndicate scoring of the target user a to commodity i,Indicate target user a in current contextual situation
Under average score, ru,iIndicate scoring of the user u to commodity i,Indicate that user u being averaged under current contextual situation is commented
Point.
The set of the determining neighbor user, specifically:
Choose the set of neighbor user of several the maximum users of user's similarity as target user.
The determining context similarity, specifically:
Wherein, simt(x, y, i) be context x and y under t dimension to commodity i similarity, u is user,It is user u
In xtTo the scoring of commodity i under context dimension,It is the average of commodity i,It is user u in ytUnder context dimension
Scoring to commodity i,It is xtThe standard deviation of context dimension,It is ytThe standard deviation of context dimension.
The prediction commodity scoring, specifically:
Wherein, Pa,i,cIt scores under c context condition the prediction of commodity i for target user a,Indicate target user a
The average score that commodity provide is bought under c context condition, u ' is the neighbor user of target user a, NcIt is target user a
The set of neighbor user u', sima,u′,cIt is target user a and similarity of a certain neighbor user u' under c context condition,
Ru′,i,cIndicate comprehensive score of the neighbor user u' on the scoring of commodity i after considering that c context influences,It is neighbor user u'
To the average score of all purchase commodity under c contextual situation, i is that target user a is not yet bought, and its neighbor user sense is emerging
Interest and higher commodity of giving a mark；
Wherein, c is context locating for target user, and x is the locating context of certain scoring record, and t is the dimension of context
Degree, simt(c, x, i) indicates context c and similarity of the context x in t dimension, ru′,i,cIt is neighbor user u' above or below c
To the scoring of commodity i under literary dimension, k indicates the weight of context impact factor.
Calculating target user's history buys the relevant frequent item set of commodity, specifically:
The frequent item set occurred in user's purchaser record is calculated with FP-growth algorithm, and is selected recent with target user
Buy the relevant support of commodity it is highest several.
The generation recommendation list, specifically:
By several commodity best in commodity scoring and frequent item set fusion, obtains recommendation list and show
To target user.
The present invention has the following advantages and beneficial effects:
1, the present invention considers the influence of context various dimensions factor, and the accuracy of recommender system can be improved.
2, present invention incorporates association rule minings, and more scientific and reasonable decision-making foundation can be provided for recommender system.
3, the present invention has stronger application, inherently considers Various Complex situation and carries out modeling analysis, can be very
Good correspondence practical application scene is extended.
Detailed description of the invention
Fig. 1 is the method for the present invention flow chart；
Fig. 2 is that user's similarity clusters schematic diagram；
Fig. 3 is the exemplary diagram recommended.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and embodiments.
Definition:
(1) it user: is indicated in text with u, refers to all users using the website or website.
(2) target user: being indicated in text with a, indicates that user for needing to carry out commercial product recommending for it.
(3) neighbor user: being indicated in text with u', indicates other users similar with target user's interest.
Step 1: collecting the purchaser record of user, and contextual information when user's purchase is contained in this purchaser record, raw
At user-rating matrix comprising contextual information.
Step 2: user's similarity is calculated
Wherein, sa,u,cIndicate the similarity value of target user a and user u between under c context, IaIndicate that target is used
The item destination aggregation (mda) that family a scored, IuIndicate the commodity set that user u was contacted.I indicates that target user a and user u is common
The commodity to score.ra,iIndicate scoring of the target user a to commodity i,Indicate target user a under current contextual situation
Average score.Similarly ru,iIndicate scoring of the user u to commodity i,Indicate user u being averaged under current contextual situation
Scoring.Wherein, target user indicates that user that recommender system is used, and system will carry out commodity for the target user and push away
It recommends.
Step 3: the set of neighbor user is determined
Usually phase can be chosen according to calculating formula of similarity (formula of step 2) when determining neighbor user group
Like spending the cluster neighbours (N can take 10,20,30 etc., depending on the circumstances) of maximum N number of user as target user.And
Contextual information can help user filtering to fall user's scoring record that part differs greatly with context under current recommendation environment.
Because some commodity decisions and current a certain context factors are in close relations, which is referred to as rigid context, is pushing away
It must be taken into consideration and meet in recommending.And some scoring records for being unsatisfactory for current context can be filtered out preferentially, it is adjacent calculating
It occupies when clustering similarity without a moment's thought.
As shown in Figure 3, wherein I1、I2、I3、I4Indicate four commodity, U1、U2、U3Illustrate four users.User U1Once
To I1、I2、I3Generated interest.User U2Once to I3、I4Generated interest.User U3Once to I2、I3Generated interest.Point
After the interest for having analysed three users, U is found1User and U3The Interest Similarity of user is larger, then by U1Interested and U3Not
The commodity I of contact1User U is recommended3.Herein, U3It is the target user of recommender system, U1It is neighbours' use of target user
Family, I1It is the commercial product recommending made to target user.
Step 4: the determination of context similarity
Assuming that context selected by a system has z different dimensions, i.e.,
C=(c1,c2,...,cz)
Wherein ct(t=1 ..., z) it is a certain context dimension (such as time, place, weather).Two scorings record x,
Similarity of the context on dimension t between y can be denoted as simt(x,y).We are with the context dimension to the shadow of scoring
The degree of sound measures the similarity between two context variables:
Wherein, u is user,It is user u in xtTo the scoring of commodity i under context dimension.It is being averaged for commodity i
Point.It is user u in ytTo the scoring of commodity i under context dimension.It is xtThe standard deviation of context dimension.It is ytOn
The hereafter standard deviation of dimension.We measure the influence degree of the scoring of identical commodity i up and down according to different context environmentals
Literary x and y is under t dimension to commodity i similarity.
Step 5: commodity score in predicting
Wherein, Pa,i,cIt scores under c context condition the prediction of commodity i for target user a.Indicate target user a
The average score that commodity provide is bought under c contextual situation.U' is the neighbor user of target user a.N is target user a
The set of neighbor user.Sim (a, u', c) is target user a and similarity of a certain neighbor user u' under c context condition.
Ru′,i,cIndicate comprehensive score of the neighbor user u' on the scoring of commodity i after considering that c context influences.It is neighbor user u'
To the average score of all purchase commodity under c contextual situation.The formula is by neighbours' cluster user of target user a to quotient
The comprehensive prediction scoring for obtaining target user a to commodity i of similarity between the scoring and user of product.Wherein i is target user a
It not yet buys, and its neighbour is interested and higher commodity of giving a mark.
Wherein c is context locating for target user, and x is the locating context of certain scoring record, Ru,i,cIt is neighbor user
U' is recorded in the synthesis size under the influence of c context to the scoring of commodity i.T is the dimension of context.It is understood that context can
There are many specific dimensions, different performances, such as time, place, related personnel are had according to the difference that data acquire.And when
Between dimension can be specifically divided into season, week, moment, festivals or holidays etc. again.Sim (c, x, t) indicates context c and context x in t
Similarity in dimension.K indicates the weight of context impact factor, can be determined by experiment.
Step 6: the relevant frequent item set of target user's history purchase commodity is calculated
The frequent item set occurred in user's purchaser record is calculated with FP-growth algorithm, is selected and is purchased in the recent period with target user
It is highest several to buy the relevant support of commodity.
Step 7: recommendation list is generated
The highest frequent episode fusion of prediction scoring (scoring in step 5) best N number of commodity and support (is put one
It rises and is presented in recommendation list.Such as assume that the length of recommendation list is 8, the item of 3 correlation rules recommendation is just put, 5 associations are put
With the recommendation items being obtained by filtration.According to the difference of recommendation list presentation mode, amalgamation mode is adjustable), recommendation list is obtained,
By the merchandise display in list to target user.
In Fig. 1, pass through the purchaser record and its context and mesh of the log information acquisition user of server first
Mark the current contextual information of user.Data are cleaned and are pre-processed.Remove the nonstandard dirty data of data structure.Then
Data prediction is carried out, the effective information in log is extracted, user-rating matrix comprising context is arranged in.With
Method in invention gradually carries out various dimensions user collaborative filtering, obtains several highest Recommendations to be selected of prediction scoring.
Meanwhile rule digging is associated to the purchasing history to user in database, it finds out and buys commodity phase in the recent period with target user
The frequent item set of pass, choose support it is highest several, be fused in recommendation list.Recommendation results are finally stored in log
In.Algorithm is adjusted by the feedback of user.
In Fig. 2, central node is equivalent to target user, and the homochromy node of surrounding, which is equivalent to, is chosen for neighborhood
User, i.e., with the higher user of target user's similarity.
Claims (8)
1. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule, which comprises the following steps:
The purchaser record of user is collected, the contextual information when purchaser record user buys, generating includes contextual information
User-rating matrix；
Calculate user's similarity；
Determine the set of neighbor user；
Determine context similarity；
Predict commodity scoring；
Calculate the relevant frequent item set of target user's history purchase commodity；
Generate recommendation list.
2. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule according to claim 1, feature
It is, whether the contextual information includes the time, place, weather, same to administrative staff, temperature, humidity, season, is festivals or holidays, is
It is no be weekend, daytime or night, whether be any several combination in the meal time.
3. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule according to claim 1, feature
It is, calculating user's similarity, specifically:
Wherein, sa,u,cIndicate similarity value of the target user a and user u under the conditions of contextual information c, IaIndicate target user
The item destination aggregation (mda) that a scored, IuIndicate the commodity set that user u was contacted, i indicates that target user a and user u is commented jointly
The commodity divided, ra,iIndicate scoring of the target user a to commodity i,Indicate target user a under current contextual situation
Average score, ru,iIndicate scoring of the user u to commodity i,Indicate average score of the user u under current contextual situation.
4. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule according to claim 1, feature
It is, the set of the determining neighbor user, specifically:
Choose the set of neighbor user of several the maximum users of user's similarity as target user.
5. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule according to claim 1, feature
It is, the determining context similarity, specifically:
Wherein, simt(x, y, i) be context x and y under t dimension to commodity i similarity, u is user,It is user u in xt
To the scoring of commodity i under context dimension,It is the average of commodity i,It is user u in ytTo quotient under context dimension
The scoring of product i,It is xtThe standard deviation of context dimension,It is ytThe standard deviation of context dimension.
6. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule according to claim 1, feature
It is, the prediction commodity scoring, specifically:
Wherein, Pa,i,cIt scores under c context condition the prediction of commodity i for target user a,Indicate target user a in c
The average score that commodity provide is bought under context condition, u' is the neighbor user of target user a, NcIt is the neighbour of target user a
Occupy the set of user u', sima,u',cIt is target user a and similarity of a certain neighbor user u' under c context condition,
Ru',i,cIndicate comprehensive score of the neighbor user u' on the scoring of commodity i after considering that c context influences,It is neighbor user u'
To the average score of all purchase commodity under c contextual situation, i is that target user a is not yet bought, and its neighbor user sense is emerging
Interest and higher commodity of giving a mark；
Wherein, c is context locating for target user, and x is the locating context of certain scoring record, and t is the dimension of context,
simt(c, x, i) indicates context c and similarity of the context x in t dimension, ru',i,cIt is that neighbor user u' is tieed up in c context
To the scoring of commodity i under degree, k indicates the weight of context impact factor.
7. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule according to claim 1, feature
It is, calculating target user's history buys the relevant frequent item set of commodity, specifically:
The frequent item set occurred in user's purchaser record is calculated with FP-growth algorithm, and is selected and bought in the recent period with target user
The relevant support of commodity it is highest several.
8. a kind of various dimensions user collaborative filtered recommendation method of combination correlation rule according to claim 1, feature
It is, the generation recommendation list, specifically:
By several commodity best in commodity scoring and frequent item set fusion, obtains recommendation list and show mesh
Mark user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810584551.5A CN108960897A (en) | 2018-06-08 | 2018-06-08 | A kind of various dimensions user collaborative filtered recommendation method of combination correlation rule |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810584551.5A CN108960897A (en) | 2018-06-08 | 2018-06-08 | A kind of various dimensions user collaborative filtered recommendation method of combination correlation rule |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108960897A true CN108960897A (en) | 2018-12-07 |
Family
ID=64493305
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810584551.5A Pending CN108960897A (en) | 2018-06-08 | 2018-06-08 | A kind of various dimensions user collaborative filtered recommendation method of combination correlation rule |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108960897A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112241894A (en) * | 2019-07-16 | 2021-01-19 | 百度时代网络技术（北京）有限公司 | Content delivery method and device and terminal |
| CN112669083A (en) * | 2020-12-30 | 2021-04-16 | 杭州趣链科技有限公司 | Commodity recommendation method and device and electronic equipment |
| CN115544242A (en) * | 2022-12-01 | 2022-12-30 | 深圳市智加云栖科技有限公司 | Big data based similar commodity model selection recommendation method |
| CN116757794A (en) * | 2023-08-17 | 2023-09-15 | 酒仙网络科技股份有限公司 | Big data-based product recommendation method in wine selling applet |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103514255A (en) * | 2013-07-11 | 2014-01-15 | 江苏谐云智能科技有限公司 | Method for collaborative filtering recommendation based on item level types |
| CN107220365A (en) * | 2017-06-07 | 2017-09-29 | 四川大学 | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing |
-
2018
- 2018-06-08 CN CN201810584551.5A patent/CN108960897A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103514255A (en) * | 2013-07-11 | 2014-01-15 | 江苏谐云智能科技有限公司 | Method for collaborative filtering recommendation based on item level types |
| CN107220365A (en) * | 2017-06-07 | 2017-09-29 | 四川大学 | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing |
Non-Patent Citations (4)
| Title |
|---|
| ANNIE CHEN 等: "Context-aware collaborative filtering system: predicting the user’s preference in the ubiquitous computing environment", 《CHI 2005 CONFERENCE ON HUMAN FACTORS IN COMPUTING SYSTEMS》 * |
| SURYA KANT 等: "Merging user and item based collaborative filtering to alleviate data sparsity", 《 INTERNATIONAL JOURNAL OF SYSTEM ASSURANCE ENGINEERING AND MANAGEMENT》 * |
| 章诗杰: "移动环境下上下文感知的协同过滤推荐模型研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 陈平华 等: "一种结合关联规则的协同过滤推荐算法", 《小型微型计算机系统》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112241894A (en) * | 2019-07-16 | 2021-01-19 | 百度时代网络技术（北京）有限公司 | Content delivery method and device and terminal |
| CN112669083A (en) * | 2020-12-30 | 2021-04-16 | 杭州趣链科技有限公司 | Commodity recommendation method and device and electronic equipment |
| CN115544242A (en) * | 2022-12-01 | 2022-12-30 | 深圳市智加云栖科技有限公司 | Big data based similar commodity model selection recommendation method |
| CN116757794A (en) * | 2023-08-17 | 2023-09-15 | 酒仙网络科技股份有限公司 | Big data-based product recommendation method in wine selling applet |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | A time-aware personalized point-of-interest recommendation via high-order tensor factorization | |
| EP3779841B1 (en) | Method, apparatus and system for sending information, and computer-readable storage medium | |
| CN108960897A (en) | A kind of various dimensions user collaborative filtered recommendation method of combination correlation rule | |
| Xie et al. | Learning graph-based poi embedding for location-based recommendation | |
| Zheng et al. | Mining correlation between locations using human location history | |
| Li et al. | Mining user similarity based on location history | |
| US9495456B2 (en) | Selecting, ranking, and/or presenting microsite content | |
| US20150112963A1 (en) | Time and location based information search and discovery | |
| KR20140071417A (en) | Mobile advertising using data networks based on intelligence data associated with internet-connectable devices derived using graph models | |
| WO2019056661A1 (en) | Search term pushing method and device, and terminal | |
| CN105488233A (en) | Reading information recommendation method and system | |
| US10228258B2 (en) | Determining top venues from aggregated user activity location data | |
| CN106960044B (en) | Time perception personalized POI recommendation method based on tensor decomposition and weighted HITS | |
| CN103077220A (en) | User group correlation degree-based personalized recommendation method and system | |
| Yang et al. | Unsupervised learning of parsimonious general-purpose embeddings for user and location modeling | |
| CN105809475A (en) | Commodity recommendation method compatible with O2O applications in internet plus tourism environment | |
| KR101519879B1 (en) | Apparatus for recommanding contents using hierachical context model and method thereof | |
| Liu et al. | Where your photo is taken: Geolocation prediction for social images | |
| CN112241489B (en) | Information pushing method, device, readable storage medium and computer equipment | |
| CN104867015B (en) | A kind of article delivery person recommendation method based on user's moving projection | |
| KR101945726B1 (en) | Contents-converging dynamic advertising system | |
| CN105718573B (en) | A kind of concern relation for user interest is extracted and mask method | |
| KR102403681B1 (en) | System and Method for Inducing Intelligent Fuzzy Based on User Loacation and Keyword | |
| Chang et al. | Hotel recommendation based on surrounding environments | |
| CN106951459A (en) | Improvement Collaborative Filtering Recommendation Algorithm based on Information Entropy |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181207 |