CN106095978B - Score in predicting and recommended method based on space proximity recommender system - Google Patents
Score in predicting and recommended method based on space proximity recommender system Download PDFInfo
- Publication number
- CN106095978B CN106095978B CN201610443878.1A CN201610443878A CN106095978B CN 106095978 B CN106095978 B CN 106095978B CN 201610443878 A CN201610443878 A CN 201610443878A CN 106095978 B CN106095978 B CN 106095978B
- Authority
- CN
- China
- Prior art keywords
- article
- user
- score
- scoring
- spatially
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims description 21
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 3
- 241001269238 Data Species 0.000 claims description 2
- 230000000717 retained effect Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of recommender system score in predicting and proposed algorithm based on space proximity, the foundation characteristic that similarity in recommender system between article is spatially laid out as it, and the similar alternatively foundation of article is carried out with its space proximity, by investigating the connection between recommender system commodity, film and other items comprehensively, by the topological layout of article network spatially, more accurate scoring calculating and recommendation service are realized.It improves recommender system and recommends the diversity of article, and the accuracy recommended is made to get a promotion.
Description
Technical field
It is the present invention relates to a kind of information processing technology, in particular to a kind of based on the pre- of space proximity recommender system scoring
Survey and recommended method.
Background technique
Personalized service is paid close attention to and is studied by many research fields as hot issue.One of personalized service is important to grind
Studying carefully content is personalized recommendation, it can find article set therewith with similar interests according to the interest of user, then base
The interested article of similitude recommended user between article.Recommend to be suitable for its own to user using personalized recommendation technology
Film, commodity etc., allow user to obtain meeting the Item Information liked and select tendency rapidly, and this is for recommending
For system, both the trust of available user, obtained more favors and used, existed but also system recommendation saves user
Consumed system resource in a large amount of navigation processes to save the operation costs such as bandwidth, and is preferably user service.
As most successful recommended technology is applied in Technologies of Recommendation System in E-Commerce, traditional Collaborative Filtering Recommendation Algorithm exists
Two aspect problems.It relies solely on the similarity between user to choose the neighbours of target user, causes to recommend precision lower, and
And when recommending, have the shortcomings that the diversity of Recommendations is poor.
It is convenient and efficient to realize therefore, it is necessary to a kind of more efficiently and there is appropriate multifarious recommender system
Service.
Summary of the invention
The problem that the present invention be directed to traditional proposed algorithm accuracy is low, system is vulnerable, proposes one kind and is based on
The score in predicting and recommended method of space proximity recommender system, by investigate comprehensively recommender system commodity, film and other items it
Between connection realize more accurate scoring calculating and recommendation service by the topological layout of article network spatially.
The technical solution of the present invention is as follows: a kind of score in predicting and recommended method based on space proximity recommender system, tool
Body includes the following steps:
1) the user's scoring library established in current recommender system: traversal active user obtains each user to all films
Scoring record, below article represent film, list part of user's score data；
2) all user's score datas are normalized with following formula:
Wherein riαScoring for user Ui to article α, ri maxAnd ri minRepresent user Ui scoring record in best result and
Minimum point, if best result and minimum split-phase etc., normalized value can be assigned to 0, ei α ∈ [- 1,1], ei α is user Ui after normalization
To the score value of article α；
3): the correlation between article is calculated, for any two article α and β, correlation calculations formula are as follows:
Wherein m, which is represented, participates in the number of users that prediction calculates, and k α indicates the degree of article α, scores for user the article
Number, kiIndicate the degree of user Ui, the degree of correlation S α β for the number for the article that the user scored, between article α and article β
It is considered as the similarity of article α Yu article β, aiαRepresent whether user Ui comments excessively article α, value is 1 or 0, aiα=
1 expression user Ui comments excessively article α, aiα=0 expression user Ui does not comment excessively article α；
4): using correlation S α β between article as side associated between article, using article as joint structure network, net
Article is only existed in network, user is not present, and the weight on side is S α β value；
5): all items being used as node initializing be placed in space (0,0) point, and in subsequent method and step respectively
Adjust the position of each node spatially；
6): calculate separately the gravitation and repulsion of article between any two:
Calculate repulsion formula are as follows:
Wherein c is adjustable empirical parameter, can influence the speed of topological layout；D is the sky between article a and article b
Between distance, Da、DbIt is the angle value of article a and article b in the article network constructed in step 4)；The gravitation of article a and article b
Gravitation calculation formula are as follows:
Wherein DaIt is the angle value of article α in a network, S α β is correlation between article a and b, and e is Euler's numbers；
7): being based on gravitation repulsion calculated result, more new article a and the position of article b spatially, new coordinate is according to such as
Lower formula updates:
x'a=xa+(xa-xb+θ)·(fa+fr)
y'a=ya+(ya-ya+θ)·(fa+fr)
x'b=xb+(xb-xb+θ)·(fa+fr)
y'b=yb+(yb-yb+θ)·(fa+fr)
Wherein x'aIt is the new x-axis coordinate of article a, xaIt is article a old coordinate in x-axis, θ is a constant, works as xa-xb≠
When 0, θ=0, xa-xbWhen=0, θ=0.5, other coordinate meanings are similar；
8): as (xa-xb+ θ) and (ya-yb+ θ) calculated result simultaneously be less than given ε, then judge article α in the position in space
It sets and is just fixed up, spatially layout method terminates after all items are fixed, otherwise, step 7) and 8) is repeated,
Complete layout spatially；
9): after completing layout spatially, the article β for doing scoring for target user Ui search does not score for scoring
The article crossed selectes α that any one the did not scored coordinate (x on article spacea,ya), it is radius model in r using r as radius
The article β that middle search user Ui did scoring is enclosed, is retained as neighbours, the n neighbours of α are selected；
10): according to following formula to user UiThe each article β searched predicts article α:
WhereinWithIt is the history average score of article α Yu article β respectively,It is prediction of the user Ui to article a
Score value, riβScoring for user Ui to article β；
11): the article α that optionally a target user Ui did not scored repeats step 9) and uses with all targets 10), are traversed
The article that family Ui did not scored recommends user Ui for highest p of score in predicting.
The beneficial effects of the present invention are: the present invention is based on the score in predicting of space proximity recommender system and recommendation side
Method, the foundation characteristic that the similarity in recommender system between article is spatially laid out as it, and with its space proximity
The similar alternatively foundation of article is carried out, recommender system is improved and recommends the diversity of article, and obtain the accuracy recommended
To promotion.
Detailed description of the invention
Fig. 1 is that the present invention is based on the score in predicting of space proximity recommender system and recommended method flow chart；
Fig. 2 is article of the present invention spatially layout method flow chart；
Fig. 3 is the schematic diagram that all films are spatially laid out in film recommender system of the present invention.
Specific embodiment
Assuming that Ui(being indicated in formula with subscript i) is the user of film recommender system, once in site databases
Several films are scored, and score value is distributed between 1-5.
Recommender system score in predicting based on space proximity and recommended method flow chart as shown in Figure 1, specific steps are such as
Under:
S1: user's scoring library in current recommender system is established:
Active user is traversed, each user is obtained and the scoring of all films is recorded, lists part of user's scoring number
According to as shown in table 1.
Table 1
S2: user's score data is normalized with following formula:
Wherein riαScoring for user Ui to film α, ri maxAnd ri minRepresent user Ui scoring record in best result and
Minimum point, if best result and minimum split-phase etc., normalized value can be assigned to 0, ei α ∈ [- 1,1], ei α is user Ui after normalization
To the score value of film α, with initial value the difference is that: its codomain is distributed between positive and negative 1, and considers user to different electricity
The TOP SCORES quantity of shadow has modified user to film prejudice that may be present；Obtain data as shown in table 2；
Table 2
S3: target user UiIt searches for article α (setting α as film 3)；
S4: the similitude between article is calculated, for any two article α and β, correlation calculations formula are as follows:
Wherein k α indicates the degree (number to score for user the article) of article α, kiIndicate the degree of user Ui (for the use
The number for the article that family was scored).S α β indicates influence of the article α to article β, is considered as the similarity of article α Yu article β,
But article similarity be it is oriented, i.e. influence between two articles is different.In addition, aiαUser Ui is represented whether to article α
It comments excessively, value is 1 or 0, aiα=1 expression user Ui comments excessive (how much no matter scoring), a to article αiα=0 indicates user
Ui does not comment excessively article α, and m, which is represented, participates in the number of users that prediction calculates.
According to correlation calculations formula, film correlation calculations are carried out, it is as shown in table 3 that correlation between film can be obtained:
Table 3
S5: using correlation S α β between article as side associated between article.Using article as one net of joint structure
Network only exists article (there is no user) in network, when β > 0 similitude S α between article α and article β, then exist a line from
Article α is connected to article β, and the weight on side is S α β value.
S6: all items being used as node initializing be placed in space (0,0) point, and in subsequent method and step respectively
Adjust the position of each node spatially, this part method (S6-S9) article spatially layout method process as shown in Figure 2
Figure.
S7: the gravitation and repulsion of article between any two are calculated separately
Calculate repulsion formula are as follows:
Wherein c is adjustable empirical parameter, can influence the speed of topological layout.D is the sky between article α and article b
Between distance, Da、DbIt is the angle value of article α and article b in the article network constructed in steps of 5, is meant that: is similar to article α
The high article of property is more, DaIt is bigger.It should be noted that the value is different from k α angle value defined in S4 step, DaIt is in S5 step
The network of generation, only article, k α indicate the degree that article α is scored by user in the network, and value represents user to the article
The number of scoring, user are more to α scoring number, and k α is bigger.
The gravitation gravitation calculation formula of article α and article b are as follows:
Wherein DaIt is the angle value of article α in a network, S α β is correlation between article a and b, and e is Euler's numbers.
S8: it is based on gravitation repulsion calculated result, more new article a and the position of article b spatially, for example, for article a
With article b, the coordinate of article a is (xa,ya), the coordinate of article b is (xb,yb).Their new coordinates according to following formula more
It is new:
x'a=xa+(xa-xb+θ)·(fa+fr)
y'a=ya+(ya-ya+θ)·(fa+fr)
x'b=xb+(xb-xb+θ)·(fa+fr)
y'b=yb+(yb-yb+θ)·(fa+fr)
Wherein x'aIt is the new x-axis coordinate of article a, xaIt is article a old coordinate in x-axis, θ is a constant, works as xa-xb≠
When 0, θ=0, xa-xbWhen=0, θ=0.5, other coordinate meanings are similar.
S9: by taking article a as an example, as (xa-xb+ θ) and (ya-yb+ θ) calculated result is less than given ε (this simultaneously
Previously given value such as 0.5) illustrates that article is spatially only moved the displacement of very little, then position of the article a in space
Just it is fixed up.Spatially layout method terminates after all items are fixed, otherwise, repeats S7 and S8 step.It is complete
It is as shown in Figure 3 at the cinema network layout spatially of layout.
S10: after completing layout spatially, search for target user Ui or score article a, selectes a in article sky
Between upper coordinate (xa,ya), using r as radius (r initial value is 0), neighbours' (not including a) of a are selected, because of r initial value very little,
So its neighbours' initial number is 0, since it is desired that constantly increasing r, until containing n neighbours in r radius, and meet Ui
Excessive requirement is all commented each of these.By taking film 3 in Fig. 3 as an example, according to the distant relationships of spatially distance, film 3
Neighbours from nearby to being far respectively as follows: film 5, film 4, film 6, film 2, film 1.It is 1 (practical when 3 neighbours of film choose n
Situation is generally 50-100) when, because user Ui does not see flash back past events 5 and film 4, film 6 is selected to predict user
Scoring of the Ui to film 3.
S11: according to following formula to user UiSearch or scoring article α are predicted:
WhereinWithIt is the history average score of article α Yu article β respectively, n is neighbours' number of article α,It is
Prediction score value of the user Ui to article α, riβScoring for user Ui to article β, β belong to the neighborhood of article α, it includes
The article (being determined by S10) of n.By calculating, obtaining user Ui is 3.75 to the scoring of film 3.Similarly, it can calculate and pre-
Surveying user is 4.166667 to the scoring of film 4；Scoring to film 5 is 2.8.
S12: by the n neighbours of α, highest p of score in predicting is recommended user Ui.As p=1, pushed away to user Ui
Recommend film 4.
Claims (1)
1. a kind of score in predicting and recommended method based on space proximity recommender system, which is characterized in that specifically include as follows
Step:
1) the user's scoring library established in current recommender system: traversal active user obtains each user and comments all films
Member record, below article represent film, list part of user's score data；
2) all user's score datas are normalized with following formula:
Wherein riαScoring for user Ui to article α, ri maxAnd ri minRepresent the best result and minimum in user Ui scoring record
Point, if best result and minimum split-phase etc., normalized value can be assigned to 0, eiα∈ [- 1,1], eiαIt is user Ui after normalization to object
The score value of product α；
3): the correlation between article is calculated, for any two article α and β, correlation calculations formula are as follows:
Wherein m representative participates in the number of users that prediction calculates, and k α indicates the degree of article α, is the number that user scores to the article,
kiIndicate the degree of user Ui, the correlation s for the number for the article that the user scored, between article α and article βαβIt can regard as
It is the similarity of article α Yu article β, aiαRepresent whether user Ui comments excessively article α, value is 1 or 0, aiα=1 indicates
User Ui comments excessively article α, aiα=0 expression user Ui does not comment excessively article α；
4): with correlation s between articleαβAs side associated between article, using article as joint structure network, in network only
There are articles, and user is not present, and the weight on side is sαβValue；
5): all items being used as node initializing be placed in space (0,0) point, and adjusted separately in subsequent method and step
The position of each node spatially；
6): calculate separately the gravitation and repulsion of article between any two:
Calculate repulsion formula are as follows:
Wherein c is adjustable empirical parameter, can influence the speed of topological layout；D be space between article a and article b away from
From Da、DbIt is the angle value of article a and article b in the article network constructed in step 4)；
The gravitation gravitation calculation formula of article a and article b are as follows:
Wherein DaIt is the angle value of article a in a network, sαβIt is correlation between article a and b, e is Euler's numbers；
7): being based on gravitation repulsion calculated result, more new article a and the position of article b spatially, new coordinate is according to following public affairs
Formula updates:
x'a=xa+(xa-xb+θ)·(fa+fr)
y'a=ya+(ya-ya+θ)·(fa+fr)
x'b=xb+(xb-xb+θ)·(fa+fr)
y'b=yb+(yb-yb+θ)·(fa+fr)
Wherein x'aIt is the new x-axis coordinate of article a, xaIt is article a old coordinate in x-axis, θ is a constant, works as xa-xbWhen ≠ 0, θ
=0, xa-xbWhen=0, θ=0.5, other coordinate meanings are similar；
8): as (xa-xb+ θ) and (ya-yb+ θ) calculated result simultaneously be less than given ε, then judge article a space position just
It is fixed up, spatially layout method terminates after all items are fixed, otherwise, repeats step 7) with 8), completes
Layout spatially；
9): after completing layout spatially, the article β that scoring was done in target user Ui search being used to score not score
Article selectes article α that any one the did not scored coordinate (x on article spacea,ya), it is radius model in r using r as radius
The article β that middle search user Ui did scoring is enclosed, is retained as neighbours, the n neighbours of α are selected；
10): according to following formula to user UiThe each article β searched predicts article α:
WhereinWithIt is the history average score of article α Yu article β respectively,It is that user Ui scores to the prediction of article α
Value, riβScoring for user Ui to article β；
11): the article α that optionally a target user Ui did not scored repeats step 9) with 10), traverses all target user Ui
The article not scored recommends user Ui for highest p of score in predicting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610443878.1A CN106095978B (en) | 2016-06-20 | 2016-06-20 | Score in predicting and recommended method based on space proximity recommender system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610443878.1A CN106095978B (en) | 2016-06-20 | 2016-06-20 | Score in predicting and recommended method based on space proximity recommender system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106095978A CN106095978A (en) | 2016-11-09 |
| CN106095978B true CN106095978B (en) | 2019-05-24 |
Family
ID=57238013
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610443878.1A Expired - Fee Related CN106095978B (en) | 2016-06-20 | 2016-06-20 | Score in predicting and recommended method based on space proximity recommender system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106095978B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4972966B2 (en) * | 2006-03-16 | 2012-07-11 | 日本電気株式会社 | Relationship display method and apparatus, and program thereof |
| CN102893275A (en) * | 2010-05-14 | 2013-01-23 | 微软公司 | Automated social networking graph mining and visualization |
| CN104182543A (en) * | 2014-09-05 | 2014-12-03 | 上海理工大学 | Similarity propagation and popularity dimensionality reduction based mixed recommendation method |
-
2016
- 2016-06-20 CN CN201610443878.1A patent/CN106095978B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4972966B2 (en) * | 2006-03-16 | 2012-07-11 | 日本電気株式会社 | Relationship display method and apparatus, and program thereof |
| CN102893275A (en) * | 2010-05-14 | 2013-01-23 | 微软公司 | Automated social networking graph mining and visualization |
| CN104182543A (en) * | 2014-09-05 | 2014-12-03 | 上海理工大学 | Similarity propagation and popularity dimensionality reduction based mixed recommendation method |
Non-Patent Citations (1)
| Title |
|---|
| 《Predicting online ratings based on the opinion spreading process》;He XingSheng等;《Physica A: Statistical Mechanics and its Applications》;20150518;第2015卷(第436期);第658-664页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106095978A (en) | 2016-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11216861B2 (en) | Color based social networking recommendations | |
| US10402917B2 (en) | Color-related social networking recommendations using affiliated colors | |
| US20190179838A1 (en) | Method and apparatus for providing book recommendation service | |
| CN104199896B (en) | The video similarity of feature based classification is determined and video recommendation method | |
| CN103955464B (en) | It is a kind of that the recommendation method perceived is merged based on situation | |
| CN103886048B (en) | Cluster-based increment digital book recommendation method | |
| CN103744966A (en) | Item recommendation method and device | |
| CN103064903A (en) | Method and device for searching images | |
| US10565262B2 (en) | Methods for refining search results in an application | |
| US20150248720A1 (en) | Recommendation engine | |
| CN104915860A (en) | Commodity recommendation method and device | |
| CN104778284A (en) | Spatial image inquiring method and system | |
| CN101957825A (en) | Method for searching image based on image and video content in webpage | |
| CN106095974A (en) | Commending system score in predicting based on network structure similarity and proposed algorithm | |
| CN103412880B (en) | Method and device for determining implicit associated information between multimedia resources | |
| CN105677810A (en) | Online shopping product searching system based on keyword analysis | |
| US20140280098A1 (en) | Performing application search based on application gaminess | |
| Ying et al. | PGRank: Personalized geographical ranking for point-of-interest recommendation | |
| Tan et al. | Preference-oriented mining techniques for location-based store search | |
| US10318843B2 (en) | Method and apparatus for image processing and comparison based on spatial relationships between image features | |
| US20150227583A1 (en) | Managing search results | |
| CN106095978B (en) | Score in predicting and recommended method based on space proximity recommender system | |
| US8745074B1 (en) | Method and system for evaluating content via a computer network | |
| Phan et al. | Collaborative recommendation of photo-taking geolocations | |
| Yu et al. | Geographical proximity boosted recommendation algorithms for real estate |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190524 |