CN108628999A - It is a kind of based on explicit and implicit information video recommendation method - Google Patents
It is a kind of based on explicit and implicit information video recommendation method Download PDFInfo
- Publication number
- CN108628999A CN108628999A CN201810411065.3A CN201810411065A CN108628999A CN 108628999 A CN108628999 A CN 108628999A CN 201810411065 A CN201810411065 A CN 201810411065A CN 108628999 A CN108628999 A CN 108628999A
- Authority
- CN
- China
- Prior art keywords
- video
- user
- information
- explicit
- implicit information
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000009412 basement excavation Methods 0.000 claims abstract description 11
- 238000001914 filtration Methods 0.000 claims abstract description 11
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 5
- 238000009795 derivation Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 235000010627 Phaseolus vulgaris Nutrition 0.000 description 2
- 244000046052 Phaseolus vulgaris Species 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
Abstract
The invention discloses a kind of based on explicit and implicit information video recommendation method, includes the following steps：1) score information and user social relationship information of the extraction user to video；2) scoring according to user to video and user social contact relation excavation user's explicit information；3) video implicit information is excavated to the scoring of video according to user；4) user's explicit information and video implicit information are merged, is added it in collaborative filtering.The present invention overcomes recommendation information in traditional collaborative filtering recommending method is single and Sparse Problem, more accurately video can be recommended.
Description
Technical field
The present invention relates to personalized recommendation fields, more particularly to based on user social contact cyberrelationship and video incidence relation
Collaborative filtering recommending method.
Background technology
With the fast development of TV media, see that video becomes one kind that people instantly like and amuses and diverts oneself mode.One
Aspect video resource brings spiritual enjoyment to user, and but then, the video resource of magnanimity causes problem of information overload,
User finds oneself interested video in the desired short time becomes extremely difficult.
As a kind of, to meet the system and tool of users ' individualized requirement, internet can be effectively relieved in commending system
Problem of information overload.Collaborative filtering based on matrix decomposition is then to solve the problems, such as that commending system score in predicting is most widely used
Technology.Currently, a lot of research work is all advanced optimized around what matrix decomposition be directed to.In order to alleviate matrix point
The problems such as faced Sparse of solution and cold start-up, additional information is added in commending system by many researchers, such as with
The explicit informations such as family social information, item attribute, user comment and geographical location.But existing research work is mostly directly by this
A little explicit informations incorporate in optimization aim.Rarely have work that goal in research is placed in the further excavation to scoring explicit information.
A kind of explicit expression of the scoring as user and project interaction, the implicit information being richly stored with, these implicit informations can be with
The useful supplement that user preference process is understood as commending system, so as to further increase the performance of commending system.
Invention content
For the deficiency of existing recommended technology, the invention discloses a kind of based on explicit and implicit information video recommendations side
Method.This method is dissolved into traditional collaborative filtering method, the party by excavating user's explicit information and video implicit information
Method can realize better recommendation effect.
In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is：It is a kind of based on explicit and implicit information
Video recommendation method includes the following steps：
1) score information and user social relationship information of the extraction user to video；
2) scoring according to user to video and user social contact relation excavation user's explicit information；
3) video implicit information is excavated to the scoring of video according to user；
4) user's explicit information and video implicit information are merged, is added it in collaborative filtering.
In above-mentioned steps 1) in, score information and user social relationship information of the user to video are extracted based on this method.
It comprises the steps of：
11) user is captured to the scoring of video as user video rating matrix R from video resource website；
12) the social networks matrix between user and user is extracted from the website, the use of each row and column value is 0 or 1
User social contact relational matrix T, T are describeduvOn the contrary=1 indicates that user u trusts user v, then be 0.
Above-mentioned steps 2) in, according to user to the scoring of video and user social contact relation excavation user's explicit information include with
Lower step：
21) for user ui, useTo indicate the neighbor user of this user.Use user
The cosine similarities of scoring portray the relationship each other between user, and excavation obtains the explicit information of user.
Above-mentioned steps 3) in, the implicit information for excavating video to the scoring of video according to user comprises the steps of：
31) according to the index of video similarity is portrayed, similarity two-by-two is calculated between all videos, to target video j,
The similarity of all videos and video j are subjected to descending sort, the highest video composition video j's of K similarity is close before selecting
Neighbour's collection Aj, excavate and obtain the local implicit information of video；
32) the global implicit information of video is obtained by the weighting excavation between video popularity and user's scoring.
Above-mentioned steps 4) in, user's explicit information and video implicit information are merged, traditional collaborative filtering skill is added it to
In art.It comprises the steps of：
41) consider two kinds of information and propose comprehensive explicit and implicit information model；
42) stochastic gradient descent method optimization object function is used.
Compared with prior art, the present invention its remarkable advantage is：The method (SILGR) of the present invention is in recommendation results accuracy
On be significantly improved, the introducing of user's explicit information and video implicit information overcomes traditional collaborative filtering recommending method and recommends
The single problem for leading to recommendation results inaccuracy with Sparse of information, can more accurately recommend video.
Description of the drawings
Fig. 1 is the embodiment of the present invention based on explicit and implicit information recommendation method frame.
Fig. 2 is the flow chart of the stochastic gradient descent method of the embodiment of the present invention.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, is clearly retouched to technical solution in the embodiment of the present invention
It states.
As shown in Figure 1, being included the following steps based on explicit and implicit information recommendation method：
1) score information and user social relationship information of the extraction user to video；
2) scoring according to user to video and user social contact relation excavation user's explicit information；
3) video implicit information is excavated to the scoring of video according to user；
4) user's explicit information and video implicit information are merged, is added it in traditional collaborative filtering.
Above-mentioned steps 1) in, score information and user social relationship information of the user to video are extracted based on this method.Packet
Containing following steps：
11) data are captured from film scoring website bean cotyledon, bean cotyledon has various types of film videos, user can be to video
It scores, social networks can also be established between user and user.The video scoring of extraction is used as user video rating matrix
R, user are usually 0 to 5 to video scoring.
12) user social contact relationship is extracted, user social contact relational matrix T, T are described for 0 or 1 using each row and column valueuv
On the contrary=1 indicates that user u trusts user v, then be 0.
Above-mentioned steps 2) in, according to user to the scoring of video and user social contact relation excavation user's explicit information include with
Lower step：
21) explicit information of user discloses the trusting relationship between their friends between user, for user ui, I
Use Ni={ uk| T (i, k)=1 } indicate this user's neighbor user.Indicate social similarity matrix, SikTable
Show user uiAnd ukTrusting relationship intensity.SikIt can be defined by following formula, be come using the cosine similarities that user scores
Portray the relationship each other between user.
The user semantic matrix in matrix decomposition model is modified by user's explicit information, is proposed aobvious based on user
The recommendation method of formula information：
For capturing the bias relation between user, SikIt is bigger, user uiAnd ukContact is closer.
Above-mentioned steps 3) in, video implicit information is excavated to the scoring of video according to user and is comprised the steps of：
31) contact of the local implicit information illustration of video between video and its similar video can pass through and be associated with rule
Then this contact of technology mining.One correlation rule can indicate as follows：li,j:I → j considers the support of correlation rule
The similarity based method IAS of asymmetric relation between video is portrayed in degree and confidence level, proposition：
Support (l in formulai,j) it is correlation rule li,jSupport, Confidence (li,j) it is correlation rule li,j's
Confidence level.RmaxIt is the maximum value to score in rating matrix, | Ru| it is the number of user in rating matrix, | Vi∩Vj| be simultaneously to
User's set of video i and video j scorings.Wherein support and confidence calculations formula difference is as follows：
Support(li,j)=| Vi∩Vj|/|Ru|
Confidence(li,j)=| Vi∩Vj|/|Vi|
According to the index for portraying video similarity, calculate between all videos that similarity will to target video j two-by-two
The similarity of all videos and video j carries out descending sort, the neighbour of the highest video composition video j of K similarity before selecting
Collect Aj, propose the recommendation method such as formula based on video local implicit information：
In formulaGjqIndicate video vjAnd vqThe confidence level of the correlation rule of composition, G are unsymmetrical matrix.
32) the global implicit information of video discloses influence power power of the single video in entirely scoring video collection,
We portray the global implicit information of video using the weighting of video popularity and user's score information.Specific calculation is such as
Under：
zj=| rj|/n
R in formulajIt is user's set of video j scorings, n is the total quantity of user；
For each user, we define a kind of inherent preference of user, are defined as follows：
I (u) refers to user u and comments excessive video collection in formula
Video j and the interior relationship in preference of user can be described as follows：
In formulaRefer to the average score of video j, QujRefer to the inherent preference of all users to video j scorings,Refer to inherent
The average value of preference vector.Using following formula by correlation coefficient value domain mapping to [0,1]：
zj=(Corrj+1)/2 (3)
Iterative formula (1) (2) (3) restrains when meeting following formula, just obtains the global implicit information power of each video
Weight；
The recommendation method such as formula based on video overall situation implicit information of proposition：
Above-mentioned steps 4) in, user's explicit information and video implicit information are merged, traditional collaborative filtering skill is added it to
In art, the learning process for adjusting user and the potential feature vector of video comprises the steps of：
41) step 2) and step 3) promote recommendation quality from user's explicit information and video implicit information respectively, comprehensive
Consider that two kinds of information propose comprehensive explicit and implicit information method, such as formula：
λ in formulau,λv, λ is regular terms parameter, and the contribution from user's explicit information is by λuControl, implicitly believes from video
The contribution of breath is by λvControl, last λ Controlling model complexity prevent over-fitting.
42) the above minimum object function is set as L, using gradient descent method to L derivations, obtains following result：
According to above-mentioned derivation, the video recommendation method process (as shown in Figure 2) based on explicit and implicit information is provided.
Above embodiment is merely illustrative of the invention's technical idea, and protection scope of the present invention cannot be limited with this, all
It is any change done on the basis of technical solution according to technological thought proposed by the present invention, each falls within present invention protection model
Within enclosing.
Claims (8)
1. a kind of based on explicit and implicit information video recommendation method, which is characterized in that include the following steps：
Step 1, score information and user social relationship information of the extraction user to video；
Step 2, the scoring according to user to video and user social contact relation excavation user's explicit information；
Step 3, video implicit information is excavated to the scoring of video according to user；
Step 4, user's explicit information and video implicit information are merged, is added in collaborative filtering.
2. according to claim 1 based on explicit and implicit information video recommendation method, which is characterized in that step 1 is wrapped
Include following steps：
Step 11, user is captured to the scoring of video as user video rating matrix R from video resource website；
Step 12, the social networks matrix between user and user is extracted from the website, the use of each row and column value is 0 or 1
User social contact relational matrix T, T are describeduvOn the contrary=1 indicates that user u trusts user v, then be 0.
3. according to claim 1 based on explicit and implicit information video recommendation method, which is characterized in that step 2 is wrapped
Include following steps：
Step 21, for user ui, use Ni={ uk| T (i, k)=1 } indicate the neighbor user of this user；Use user
The cosine similarities of scoring portray the relationship each other between user, and excavation obtains the explicit information of user.
4. according to claim 3 based on explicit and implicit information video recommendation method, which is characterized in that step 2 into
One step includes the following steps：
In step 21,Indicate social similarity matrix, SikIndicate user uiAnd ukTrusting relationship intensity, SikPass through
Following formula definition, the relationship each other between user is portrayed using the cosine similarities of user's scoring,
The recommendation method based on user's explicit information of proposition：
For capturing the bias relation between user, SikIt is bigger, user uiAnd ukContact is closer.
5. according to claim 1 based on explicit and implicit information video recommendation method, which is characterized in that step 3 is wrapped
Include following steps：
Step 31, according to the index for portraying video similarity, similarity two-by-two is calculated between all videos, to target video j,
The similarity of all videos and video j are subjected to descending sort, the highest video composition video j's of K similarity is close before selecting
Neighbour's collection Aj, excavate and obtain the local implicit information of video；
Step 32, the global implicit information of video is obtained by the weighting excavation between video popularity and user's scoring.
6. according to claim 5 based on explicit and implicit information video recommendation method, which is characterized in that step 3 into
One step includes the following steps：
In step 31, a correlation rule of video i and video j are represented by：li,j:I → j is considering support and is setting
On the basis of reliability, the similarity based method IAS of asymmetric relation between video is portrayed in proposition, and calculation formula is as follows：
Support (l in formulai,j) it is correlation rule li,jSupport, Confidence (li,j) it is correlation rule li,jConfidence
Degree；RmaxIt is the maximum value to score in rating matrix, | Ru| it is the number of user in rating matrix, | Vi∩Vj| it is simultaneously to video
User's set of i and video j scorings；Wherein support and confidence calculations formula difference is as follows：
Support(li,j)=| Vi∩Vj|/|Ru|
Confidence(li,j)=| Vi∩Vj|/|Vi|
The recommendation method such as formula based on video local implicit information of proposition：
In formulaGjqIndicate video vjAnd vqThe confidence level of the correlation rule of composition, G are unsymmetrical matrix；
In step 32, the weighting between video popularity and user's scoring, calculation is as follows：
zj=| rj|/n
R in formulajIt is user's set of video j scorings, n is the total quantity of user；
For each user, the inherent preference of user is defined as follows：
I (u) refers to user u and comments excessive video collection in formula；
Video j and the interior relationship in preference of user can be described as follows：
In formulaRefer to the average score of video j, QujRefer to the inherent preference of all users to video j scorings,Refer to inherent preference
The average value of vector；Using following formula by correlation coefficient value domain mapping to [0,1]；
zj=(Corrj+1)/2 (3)
Iterative formula (1) (2) (3) restrains when meeting following formula, just obtains the global implicit information weight of each video；
The recommendation method such as formula based on video overall situation implicit information of proposition：
7. according to claim 1 based on explicit and implicit information video recommendation method, which is characterized in that step 4 is wrapped
Include following steps：
Step 41：Consider two kinds of information and proposes the video recommendation method based on explicit and implicit information；
Step 42：Use stochastic gradient descent method optimization object function.
8. according to claim 7 based on explicit and implicit information video recommendation method, which is characterized in that step 4 into
One step includes the following steps：
In step 41, considers two kinds of information and propose the video recommendation method based on explicit and implicit information, such as formula：
λ in formulau,λv, λ is regular terms parameter, and the contribution from user's explicit information is by λuControl, from video implicit information
Contribution is by λvControl, last λ Controlling model complexity prevent over-fitting；
In step 42, if it is that L obtains following result using gradient descent method to L derivations to minimize object function above：
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810411065.3A CN108628999B (en) | 2018-05-02 | 2018-05-02 | Video recommendation method based on explicit and implicit information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810411065.3A CN108628999B (en) | 2018-05-02 | 2018-05-02 | Video recommendation method based on explicit and implicit information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108628999A true CN108628999A (en) | 2018-10-09 |
| CN108628999B CN108628999B (en) | 2022-11-11 |
Family
ID=63695397
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810411065.3A Active CN108628999B (en) | 2018-05-02 | 2018-05-02 | Video recommendation method based on explicit and implicit information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108628999B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866181A (en) * | 2019-10-12 | 2020-03-06 | 平安国际智慧城市科技股份有限公司 | Resource recommendation method, device and storage medium |
| CN110955775A (en) * | 2019-11-11 | 2020-04-03 | 南通大学 | Drawing book recommendation method based on implicit inquiry |
| CN114282101A (en) * | 2021-12-20 | 2022-04-05 | 北京百度网讯科技有限公司 | Training method and device of product recommendation model, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104935963A (en) * | 2015-05-29 | 2015-09-23 | 中国科学院信息工程研究所 | Video recommendation method based on timing sequence data mining |
| CN106682114A (en) * | 2016-12-07 | 2017-05-17 | 广东工业大学 | Personalized recommending method fused with user trust relationships and comment information |
| CN106777051A (en) * | 2016-12-09 | 2017-05-31 | 重庆邮电大学 | A kind of many feedback collaborative filtering recommending methods based on user's group |
-
2018
- 2018-05-02 CN CN201810411065.3A patent/CN108628999B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104935963A (en) * | 2015-05-29 | 2015-09-23 | 中国科学院信息工程研究所 | Video recommendation method based on timing sequence data mining |
| CN106682114A (en) * | 2016-12-07 | 2017-05-17 | 广东工业大学 | Personalized recommending method fused with user trust relationships and comment information |
| CN106777051A (en) * | 2016-12-09 | 2017-05-31 | 重庆邮电大学 | A kind of many feedback collaborative filtering recommending methods based on user's group |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866181A (en) * | 2019-10-12 | 2020-03-06 | 平安国际智慧城市科技股份有限公司 | Resource recommendation method, device and storage medium |
| CN110955775A (en) * | 2019-11-11 | 2020-04-03 | 南通大学 | Drawing book recommendation method based on implicit inquiry |
| CN114282101A (en) * | 2021-12-20 | 2022-04-05 | 北京百度网讯科技有限公司 | Training method and device of product recommendation model, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108628999B (en) | 2022-11-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109241405B (en) | Learning resource collaborative filtering recommendation method and system based on knowledge association | |
| US9704026B1 (en) | Systems and methods for building and using social networks in image analysis | |
| CN105612514B (en) | System and method for image classification by associating contextual cues with images | |
| US11341207B2 (en) | Generating app or web pages via extracting interest from images | |
| CN104834686B (en) | A kind of video recommendation method based on mixing semantic matrix | |
| US9269016B2 (en) | Content extracting device, content extracting method and program | |
| CN101872346B (en) | Method for generating video navigation system automatically | |
| CN105139211B (en) | Product brief introduction generation method and system | |
| CN108628999A (en) | It is a kind of based on explicit and implicit information video recommendation method | |
| Rudinac et al. | Learning crowdsourced user preferences for visual summarization of image collections | |
| CN112507248A (en) | Tourist attraction recommendation method based on user comment data and trust relationship | |
| CN104636371A (en) | Information recommendation method and device | |
| EP4379579A2 (en) | Facilitating use of images as search queries | |
| US20120030711A1 (en) | Method or system to predict media content preferences | |
| CN103064903A (en) | Method and device for searching images | |
| CN110532480B (en) | Knowledge graph construction method for recommending human-read threat information and threat information recommendation method | |
| CN112069326B (en) | Knowledge graph construction method and device, electronic equipment and storage medium | |
| CN104199838B (en) | A kind of user model constructing method based on label disambiguation | |
| CN108241619A (en) | A kind of recommendation method based on the more interest of user | |
| US10552473B2 (en) | Systems and methods for processing media content that depict objects | |
| CN102999489B (en) | The picture retrieval method of a kind of community website page and system | |
| Vasudevan et al. | Image-based recommendation engine using VGG model | |
| EP2835748A1 (en) | Systems and methods for image classification by correlating contextual cues with images | |
| US10838995B2 (en) | Generating distinct entity names to facilitate entity disambiguation | |
| TWI556123B (en) | News tracking and recommendation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |