CN107103000A - It is a kind of based on correlation rule and the integrated recommended technology of Bayesian network - Google Patents
It is a kind of based on correlation rule and the integrated recommended technology of Bayesian network Download PDFInfo
- Publication number
- CN107103000A CN107103000A CN201610096873.6A CN201610096873A CN107103000A CN 107103000 A CN107103000 A CN 107103000A CN 201610096873 A CN201610096873 A CN 201610096873A CN 107103000 A CN107103000 A CN 107103000A
- Authority
- CN
- China
- Prior art keywords
- bayesian network
- node
- correlation rule
- network
- bayesian
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
For the intelligent recommendation technology of business web site, learn bayesian network structure on the basis of correlation rule, the conditional probability of each network address is accessed based on optimal Bayesian network forecasting active user, find out the maximum N number of network address of conditional probability and recommend user.First, analysis is associated to network address, and correlation rule is sorted by lifting degree.Then, the relation according to correlation rule front and back pieces, initial Bayesian network is changed into by correlation rule.Structure learning subsequently is carried out to initial Bayesian network, optimal bayesian network structure is found, and learn the parameter of optimal bayesian network structure.Finally predict that active user accesses the probability of unknown network address using bayes method, N number of network address of maximum probability is recommended into user.The present invention is creatively by correlation rule and two kinds of data digging methods of Bayesian network are integrated is used on the intelligent recommendation of website, and both approaches combine complementary, improve the accuracy and operation efficiency of model.
Description
Technical field
The present invention relates to the intelligent recommendation technical field of commercial class website, and in particular to by correlation rule and Bayesian network
Integrated proposed algorithm.
Background technology
Internet and information technology fast development make it that business web site transaction is more and more frequent, and substantial amounts of information aggregation is got up
Form magnanimity information.User is helped rapidly and accurately to search out oneself information interested from magnanimity information, setting up one has
The commending system of effect, can make business web site set up the loyal customer base of stable enterprise, improve user satisfaction.
Correlation rule is very simple due to form, using convenient, is rapidly progressed.Correlation rule can be used for finding commercial affairs
Contact in the transaction data base of website between disparity items, these rules reflect the website browsing pattern of user.It was found that these
Rule, which can be applied, recommends network address interested to user.Since correlation rule is proposed, it has also become the recommended technology of main flow.
But correlation rule can not express the contact between Different Rule, this greatly limits correlation rule answering under complicated case
With.And Bayesian network is indicated how a series of conditional probability function related to nodes being combined into patterned form
For the joint probability distribution function of an entirety.One Bayesian network includes a structural model and associated one group
Conditional probability distribution function.Structural model is a directed acyclic graph, and node therein illustrates stochastic variable, and it is for mistake
The description of certain characteristic of the entities such as journey, event, state, side then represents the probability dependency between variable.Each node in figure
There is the conditional probability distribution function of a node in the case of its father node is given.
Because Bayesian network is the expression of the joint probability of variable, so during being made inferences to node state,
Each factor can be considered（Father node）Influence, because correlation rule and Bayesian network are all using probability theory as theory
Basis, it may be considered that use Bayesian network amendment correlation rule, and predict by way of probabilistic inference active user's visit
The probability to not browsing network address is asked, by obtained sort result, recommends probability top-N network address.
The present invention browses information according to user's history, finds the connection between disparity items in business web site transaction data base
System, the rule of these website browsing patterns for reflecting user is expressed with directed acyclic graph, pre- based on Bayesian network
The conditional probability that active user accesses each network address is surveyed, the maximum N number of network address of conditional probability is found out and recommends user.Provide the user
The substantial connection set up a web site while the service of personalization between user, allows user to produce dependence to commending system, so that
The loyal customer base of stable enterprise is set up, realizes that client's chain reaction is rised in value, improves customer satisfaction.By improving service effect
Rate helps consumer to save transaction cost etc., formulates targetedly marketing strategy policy, promotes enterprise to send out at a high speed steadily in the long term
Exhibition.
The content of the invention
The present invention is directed to the intelligent recommendation technology of business web site, learns Bayesian network knot on the basis of correlation rule
Structure, is proposed correlation rule and the integrated proposed algorithm of Bayesian network.
Method comprises the following steps：First, analysis is associated to network address, and correlation rule is sorted by lifting degree.So
Afterwards, the relation according to correlation rule front and back pieces, initial Bayesian network is changed into by correlation rule.Subsequently to initial Bayes
Network carries out Structure learning, finds optimal bayesian network structure, and learns the parameter of optimal bayesian network structure, now
Optimal Bayesian network is equivalent to the amendment to correlation rule.Finally non-Hownet is accessed using bayes method prediction active user
The probability of location, user is recommended by N number of network address of maximum probability.
Described is integrated with Bayesian network by correlation rule, is that original transaction collection is closed using Apriori algorithm
Connection rule, after being sorted by lifting degree, converts it into bayesian network structure.The structure of Bayesian network is a directed acyclic
Each node in figure, figure uniquely corresponds to a stochastic variable, and the state of node corresponds to the value of stochastic variable.In figure
Directed edge represents variable（Node）Between condition（Cause and effect）Dependence.Also contain one between the former piece and consequent of correlation rule
Dependence is planted, the thinking of conversion is exactly to come out this dependence in correlation rule with the representation of Bayesian network.
Described Bayesian forecasting, one group of random vector variable is regarded as by whether all network address are accessed, active user's
History access record is exactly a sample, with reference to this sample data and parameter priori, predicts the accessed probability of some network address.
Variable includes Bayesian network node and non-Bayesian network node two parts, it is assumed that Bayesian network node condition is independent, non-
Bayesian network node is separate, and Bayesian network node and non-Bayesian network node are separate.
The described correlation rule proposed algorithm integrated with Bayesian network is comprised the following steps that：
1) data prediction.On the basis of the Exploring Analysis to initial data, it is found that or model unrelated with analysis target needs
The data of processing, are handled for such data.Converted by data cleansing, data integration and data, at initial data
Manage into the input data of model needs.WhereinCollect for user,For address set.
2) correlation rule.Data set D is changed into transaction set DT first, it is contemplated that, this hair corresponding with Bayesian network
Bright analysis has the correlation rule former piece replacement problem of single consequent attribute status, therefore Apriori algorithm need to only be retrieved
All 2 frequent item sets in transaction database, the rule for meeting user's minimum lift degree, and root are constructed using frequent item set
Rule is arranged by order from big to small according to lifting degree.
3) correlation rule is changed into Bayesian network.The thinking of conversion is exactly by this dependence in correlation rule
Come out with the representation of Bayesian network.The former piece network address of correlation ruleWithAccumulate between consequent network address
Contain a kind of dependence.Bayesian network is such as
Fruit is present from nodePoint to nodeDirected edge, then existPoint toDirection on,Status condition depend onState, claimIt isA father node,Father node collection can be expressed as.The Xiang Yubei of correlation rule
The node of this network of leaf be it is corresponding, in Bayesian network point represent be a variable, refer to user whether browse network address this
Individual two-valued variable, and it is a state of this variable that the item in correlation rule, which is represented, i.e. user access network address this event.
So just each node and its father node are found out from top to bottom according to lifting degree.
4) optimal Bayesian network is found, and estimates parameter.The present invention learns the knot of Bayesian network using MCMC methodology
Structure, by Gibbs sampling algorithms, using local side increase, is deleted and the reverse suggestion point being uniformly distributed as sampling process
Cloth, and produce after being restrained using sampling process estimate Bayesian network from the network structure sample of target Stationary Distribution
Architectural feature, build optimal Bayesian network.Carried out on the basis of the tangible optimal bayesian network structure of study of parameter,
The parameter that the present invention is estimated using bayes method, the conditional probability distribution of node is all multinomial distribution, therefore parameter is total to
Yoke priori and its Posterior distrbutionp are all the distributions of Di Li Crays.From the perspective of correlation rule, the learning process of Bayesian network
The actually optimization process of correlation rule.
5) Bayesian forecasting.Target network addressWhether be accessed is stochastic variable, variable is divided into Bayesian network node
With non-Bayesian network node, non-Bayesian network node is separate, non-Bayesian network node and Bayesian network node
It is separate, if soIt is non-Bayes's node, predicted value is its marginal probability；Bayesian network node condition is independent,
Bayesian network node and non-Bayesian network node are separate, ifIt is Bayes's node, predicted value is conditional probability
Product.
6) recommendation is made.The predicted value of individual network address is sorted, the network address for therefrom selecting predicted value top-N recommends current use
Family.Then model is evaluated in terms of precision and timeliness two.Statistical accuracy method uses index mean absolute error
（MAE）.Decision support precision index uses accurate rate（precision）And recall rate（recall）.The Shi Xiaoyong response times
Weigh.
The present invention is creatively by correlation rule and two kinds of data digging methods of Bayesian network are integrated is used in website intelligence
On recommending.On the one hand, correlation rule due to form it is very simple, using convenient, but the contact between Different Rule can not be expressed,
Application under complicated case is greatly limited.And Bayesian network represents the joint probability of variable, each can be considered
Factor（Father node）Influence.On the other hand, it is computationally intensive when Bayesian network number of nodes is big, complicated, learning time
Long, correlation rule can obtain the dependence between node faster, can be quick by the Bayesian network of correlation rule
Study is optimal structure.Both approaches combine complementary, empirical tests（See below literary algorithm example）Improve the accuracy of model
And operation efficiency.The present invention accesses the probability of unknown network address with Bayesian forecasting active user, and bayes method can make priori
Knowledge and data are organically combined, and when sample data is sparse, priori can be made full use of to obtain reliable results, some mistakes
In unexpected winner or popular network address, as the abnormity point of analysis, it can also be made full use of in Bayesian network analysis, and steadily and surely tied
Really.
Brief description of the drawings
The flow chart of Fig. 1 correlation rules and the integrated proposed algorithm of Bayesian network；
The flow chart of Fig. 2 step 1 data predictions；
Fig. 3 step 3 correlation rules change into the general expression example of Bayesian network, to a correlation rule, the Bayes's knot built by the algorithm class of step 3 of the present invention
Structure；
The mean absolute error of Fig. 4 models of the present invention and the comparison figure of traditional association rule-based algorithm；
The accurate rate and recall rate of Fig. 5 models of the present invention and the comparison figure of traditional association rule-based algorithm；
The comparison figure of the time performance of Fig. 6 models of the present invention and traditional Bayes net algorithm.
Embodiment
Technical scheme is described in detail with reference to the accompanying drawings and examples.
Fig. 1 gives business web site personalized recommendation method process, comprises the following steps that：
Step 1：Data prediction.On the basis of the Exploring Analysis to initial data, or model unrelated with analysis target is found
Data to be processed are needed, are handled for such data.Converted by data cleansing, data integration and data, by original number
According to the input data for being processed into model needs.WhereinCollect for user,For address set.
Step 2：Correlation rule.Data set D is changed into transaction set DT first, it is contemplated that corresponding with Bayesian network, this
Invention only correlation rule former piece replacement problem of the analysis with single consequent attribute status, therefore Apriori algorithm only needs retrieval
All 2 frequent item sets gone out in transaction database, the rule for meeting user's minimum lift degree is constructed using frequent item set, and
Rule is arranged by order from big to small according to lifting degree.
Step 3：Correlation rule is changed into Bayesian network.The thinking of conversion is exactly will association
This dependence in rule is come out with the representation of Bayesian network.The former piece network address of correlation ruleWithContain a kind of dependence between consequent network address
.Bayesian network is if there is from nodePoint to nodeDirected edge, then existPoint toDirection on,State
Condition is depended onState, claimIt isA father node,Father node collection can be expressed as.Association rule
Item then is corresponding with the node of Bayesian network, and what the point in Bayesian network was represented is a variable, whether refers to user
Browse network address this two-valued variable, and it is a state of this variable that the item in correlation rule, which is represented, i.e. user accesses net
This event of location.So just each node and its father node are found out from top to bottom according to lifting degree.
Step 4：Optimal Bayesian network is found, and estimates parameter.The present invention learns Bayesian network using MCMC methodology
Structure, by Gibbs sampling algorithms, using local side increase, delete and reverse be uniformly distributed building as sampling process
View distribution, and produce after being restrained using sampling process estimate Bayes from the network structure sample of target Stationary Distribution
The architectural feature of network, builds optimal Bayesian network.Enter on the basis of the tangible optimal bayesian network structure of the study of parameter
It is capable, the parameter that the present invention is estimated using bayes method, the conditional probability distribution of node is all multinomial distribution, therefore parameter
Conjugate prior and its Posterior distrbutionp be all Di Li Crays distribution.From the perspective of correlation rule, the study of Bayesian network
Process is actually the optimization process of correlation rule.
Step 5：Bayesian forecasting.Target network addressWhether be accessed is stochastic variable, variable is divided into Bayesian network
Node and non-Bayesian network node, non-Bayesian network node are separate, non-Bayesian network node and Bayesian network
Node is separate, if soIt is non-Bayes's node, predicted value is its marginal probability；Bayesian network node condition is only
Vertical, Bayesian network node and non-Bayesian network node are separate, ifIt is Bayes's node, predicted value is that condition is general
The product of rate.
Step 6：Make recommendation.The predicted value of individual network address is sorted, the network address for therefrom selecting predicted value top-N is recommended and worked as
Preceding user.Then model is evaluated in terms of precision and timeliness two.Statistical accuracy method uses index mean absolute error
（MAE）.Decision support precision index uses accurate rate（precision）And recall rate（recall）.The Shi Xiaoyong response times
Weigh.
Described step 1 is described as follows：
Extract in certain Legal website three months（2015-02-01~2015-04-29）The access data of In Guangzhou Area user are used as original
Beginning data set.A total of 837450 records of its data volume, including user number, access time and accession page.From original
Data in data to repeated data, unrelated with analysis target（The page of lawyer's login assistant）And webpages（After html
The webpage sewed）Cleaned.The network address of page turning belongs to same type of webpage, and these webpages need to reduce its original classification.In advance
245515 records are extracted after processing, 108204 users, 48573 network address, are used as experimental data set altogether.Note user collects, address set is, time interval collection is, by interpretation shaping such as data source。
Described step 2 is described as follows：
Pretreated data are converted into transaction database DT first, then call Apriori algorithms to be obtained using iteration
Frequent Set set, generates binomial Frequent Set, according to support, lifting degree and confidence level generation Strong association rule, and by lifting
Degree sequence.The generating algorithm of correlation rule is specific as follows：
Input：Transaction Information, minimum support, minimum lift degree；Minimum confidence
Degree；.
Output：Correlation rule。
Step：
1）The support of 1 item collection is calculated, 1 Frequent Set is found；
2）According to Apriori algorithm byGenerate the Candidate Set of 2 Frequent Sets；
3）Permutation and combination goes out Candidate SetIn 2 item collections, calculate the support of each 2 item collection；
2 Frequent Sets are filtered out,
4）Correlation rule is initialized；
5）To 2 of any 2 Frequent Sets, if, and meetOr
, then haveOr。
6）By correlation ruleAccording to the order arrangement of lifting degree from big to small.
Described step 3 is described as follows：
Remember that the Bayesian network for having merged correlation rule is, whereinIt is Bayesian network knot
Structure,The random vector variable whether accessed for correlation rule former piece and consequent network address, directed edge
CollectionThe condition dependence between front and back pieces network address is represented, if there is from nodePoint to nodeDirected edge, then existRefer to
ToDirection on,Status condition depend onState, claimIt isA father node,Father node collection can
To be expressed as；For the set of conditional probability distribution, by the bar of each node
Part probability tabular value is constituted.Represent nodeIn his father's set of nodeCondition under the influence of value is general
Rate.
The establishment of the Bayesian network also needs to meet following three hypothesis：
Assuming that one：Assuming that influence of all former pieces to consequent is all independent.
Assuming that two：All former pieces have been use up assuming that having arranged.
Assuming that three：Assuming that the former piece without appearance does not influence on consequent.
Bayesian network structureThe algorithm of generation is as follows：
Input：Transaction data set (TDS), correlation rule to be modified；
Output：Bayesian network structure。
Step：
1）Initialisation structures formula；
2）For ruleIn each Frequent SetIf,, then；
3）Construct pointerPoint to ruleOriginal position.
4）Take outThe rule of sensing,If,Between be not present directed walk, then；
5）IfDo not arrive alsoIf the end,Next rule is pointed to, step 4 is returned to；
6）DeleteIn there is no the connected node in side.
Described step 4 includes following sub-step：
S4.1：Bayesian network structure learning：
According to correlation ruleThe address set that includes of item collectionBrowsed record, find corresponding user collection, constitute and use
Family network address matrix, it is used as training data matrix：
The present invention learns the structure of Bayesian network using MCMC methodology, by Gibbs sampling algorithms, local arc is increased,
Delete and the reverse suggestion distribution being uniformly distributed as sampling process, and produce after being restrained using sampling process come from mesh
The network structure sample of Stationary Distribution is marked to estimate the architectural feature of Bayesian network.
Add auxiliary variable,
The effect of auxiliary variable is description bayesian network structure, note vector：
,
The sampling process of Gibbs sampling algorithms is as follows：
1）The Bayesian network that will be constructed according to correlation ruleAs original state, now
。
2）It is rightCirculating sampling
3）
4） …
5）
6） …
7）
8） …
9）
Obtained after algorithm above convergenceStable distritation, the Bayesian network thus constructed be required optimum network structure。
S4.2：Bayesian network parameters learn：
The present invention learns the parameter of Bayesian network using MCMC methodology, it is assumed that the estimation of each distributed constant is separate.
According to parameter independence assumption, the conditional probability of node
WhereinFor network addressAccessed number of times,For total instance number.Order,Then there is stochastic variableThe parameter of condition distribution
,Combination condition be distributed as
Gibbs sampling algorithms are applied from the Posterior distrbutionp of parameter, specific sampling process is as follows：
1）The Bayesian network that will be constructed according to correlation ruleAs original state, now
2）It is rightCirculating sampling
3）
4）…
5）
6）…
7）
8）…
9）
Wherein。
The derivation of Posterior distrbutionp is as follows：
Due toLikelihood function be multinomial distribution, thereforeConjugate gradient descent method be Di Li Crays distribution：
Wherein super Study first。
Posterior distrbutionp be
Wherein,,, when
When,For network addressAnd network addressThe number of times being accessed simultaneously, whenWhen,For network addressAccessed time
Number.
Described step 5 is described as follows：
Active userHistory access network address record, regard stochastic variable asA sample,
The purpose of Bayesian forecasting is exactly in known sample dataUnder conditions of, predict each network addressAccessed probability, network address is interviewed
The stochastic variable note whether asked,There are two kinds of situations of Bayesian network node and non-Bayesian network node.
1. work as network addressNetwork address outside for training sample, i.e.,。
2. work as network addressFor the network address in training sample, i.e.,。
This is a Higher Dimensional Integration, calculates complicated, we seek another thinking, it is contemplated that,,, andIndependently of, thereforeBi-distribution.
Wherein, note
Then haveBe desired for：
Therefore
Network addressAccessed predicted value is：
Described step 6 is described as follows：
Mean absolute errorThe error of predicted value and actual value is represented, being averaged for recommendation results is exhausted
Recommendation quality smaller to error is better.Accurate rate, recall rate.Wherein
It is the number for recommending correct article in recommendation results,It is the number of the article of mistake recommendation in recommendation results,It is
The number for the article that should be recommended but not appear in recommendation results.The accurate rate and recall rate of recommendation results are higher, then
The recommendation results finally given are more accurate.Two tables are respectively with regard to traditional association rule algorithm, and correlation rule and Bayes below
System integrating algorithm comparison mean absolute error value and accurate rate and recall rate.Time performance depends on data scale, divides below
Just traditional Bayesian network and correlation rule are not compared the response time with Bayesian network Integrated Algorithm.
Mean absolute error under the different pieces of information scale of table 1
The accurate rate and recall rate of the different recommendation numbers of table 2
Timeliness under the different pieces of information scale of table 3
In summary analyze, the present invention is by correlation rule and two kinds of data digging methods of Bayesian network are integrated is used in website intelligence
On recommending, combine two methods complementary, improve the accuracy and operation efficiency of model.Correlation rule faster can be obtained
To the dependence between node, structure can be optimal with Fast Learning by the Bayesian network of correlation rule.Bayes
Predict that active user accesses the probability of unknown network address, bayes method can be such that priori and data organically combine, in sample
When notebook data is sparse, priori can be made full use of to obtain reliable results, some excessively unexpected winners or popular network address are used as analysis
Abnormity point, can also be made full use of in Bayesian network analysis, and obtain sane result.
Claims (4)
1. for the intelligent recommendation technology of commercial class website, it is characterised in that correlation rule is integrated with Bayesian network, it is based on
Bayesian network forecasting active user accesses the conditional probability of each network address, finds out the maximum N number of network address of conditional probability and recommends use
Family, method comprises the following steps：First, analysis is associated to network address, correlation rule is obtained, and sorted by lifting degree, then,
According to the relation of correlation rule front and back pieces, correlation rule is changed into initial Bayesian network, subsequently to initial Bayesian network
Network carries out Structure learning, finds optimal bayesian network structure, learns the parameter of optimal bayesian network structure, finally using shellfish
Leaf this method prediction active user accesses the probability of unknown network address, and N number of network address of maximum probability is recommended into user.
2. the intelligent recommendation technology according to claim 1 for commercial class website, it is characterised in that described will association
Rule is integrated with Bayesian network, is that original transaction collection is obtained with single consequent attribute status using Apriori algorithm
Correlation rule, after being sorted by lifting degree, converts it into bayesian network structure, the structure of Bayesian network is a directed acyclic
Each node in figure, figure uniquely corresponds to a stochastic variable, and the state of node corresponds in the value of stochastic variable, figure
Directed edge represents variable（Node）Between condition（Cause and effect）Dependence, also contains one between the former piece and consequent of correlation rule
Dependence is planted, the thinking of conversion is exactly to come out this dependence in correlation rule with the representation of Bayesian network.
3. the intelligent recommendation technology according to claim 1 for commercial class website, it is characterised in that described Bayes
Prediction, regards one group of random vector variable as, the history access record of active user is exactly one by whether all network address are accessed
Sample, with reference to this sample data and parameter priori, predicts the accessed probability of some network address, variable includes Bayesian network section
Point and non-Bayesian network node two parts, it is assumed that Bayesian network node condition is independent, non-Bayesian network node is mutually only
Vertical, Bayesian network node and non-Bayesian network node are separate.
4. the intelligent recommendation technology of commercial class website according to claim 1, described by correlation rule and Bayesian network
The algorithm of the integrated recommendation of network is comprised the following steps that：
1）Data prediction, on the basis of the Exploring Analysis to initial data, it is found that or model unrelated with analysis target needs
The data of processing, are handled for such data, are converted by data cleansing, data integration and data, at initial data
Manage into the input data of model needs, whereinCollect for user,For address set；
2）Correlation rule, changes into transaction set DT by data set D first, it is contemplated that corresponding with Bayesian network, and the present invention is only
Correlation rule former piece replacement problem of the analysis with single consequent attribute status, therefore Apriori algorithm need to only retrieve affairs
All 2 frequent item sets in database, construct the rule for meeting user's minimum lift degree using frequent item set, and according to carrying
Liter degree is arranged rule by order from big to small；
3）Correlation rule is changed into Bayesian network, the thinking of conversion is exactly to use this dependence in correlation rule
The representation of Bayesian network comes out, the former piece network address of correlation ruleWithContain between consequent network address
A kind of dependenceIf Bayesian network is deposited
From nodePoint to nodeDirected edge, then existPoint toDirection on,Status condition depend onState,
ClaimIt isA father node,Father node collection can be expressed as, the item of correlation rule and Bayesian network
Node is corresponding, and what the point in Bayesian network was represented is a variable, refers to whether user browses network address this two-valued variable,
And it is a state of this variable that the item in correlation rule, which is represented, i.e., user accesses network address this event, so according to carrying
Liter degree from top to bottom just finds out each node and its father node；
4）Optimal Bayesian network is found, and estimates parameter, the present invention learns the structure of Bayesian network using MCMC methodology, leads to
Gibbs sampling algorithms are crossed, using local side increase, are deleted and the reverse suggestion distribution being uniformly distributed as sampling process, and
What is produced after being restrained using sampling process estimates the knot of Bayesian network from the network structure sample of target Stationary Distribution
Carried out on the basis of structure feature, the optimal Bayesian network of structure, the tangible optimal bayesian network structure of study of parameter, this hair
The parameter of bright use bayes method estimation, the conditional probability distribution of node is all multinomial distribution, therefore the conjugation of parameter is first
Test and its Posterior distrbutionp is all the distribution of Di Li Crays, from the perspective of correlation rule, the learning process of Bayesian network is actual
On be correlation rule optimization process；
5）Bayesian forecasting, it is stochastic variable that whether target network address is accessed, variable is divided into Bayesian network node and Fei Bei
This network node of leaf, non-Bayesian network node is separate, and non-Bayesian network node and Bayesian network node are mutually only
It is vertical, if soIt is non-Bayes's node, predicted value is its marginal probability；Bayesian network node condition is independent, Bayesian network
Network node and non-Bayesian network node are separate, ifIt is Bayes's node, predicted value is the product of conditional probability；
6）Recommendation is made, the predicted value of individual network address is sorted, the network address for therefrom selecting predicted value top-N recommends active user,
Then model is evaluated in terms of precision and timeliness two, statistical accuracy method uses index mean absolute error（MAE）, certainly
Plan supports precision index to use accurate rate（precision）And recall rate（recall）, when effectiveness response time weigh.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610096873.6A CN107103000A (en) | 2016-02-23 | 2016-02-23 | It is a kind of based on correlation rule and the integrated recommended technology of Bayesian network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610096873.6A CN107103000A (en) | 2016-02-23 | 2016-02-23 | It is a kind of based on correlation rule and the integrated recommended technology of Bayesian network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107103000A true CN107103000A (en) | 2017-08-29 |
Family
ID=59659010
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610096873.6A Pending CN107103000A (en) | 2016-02-23 | 2016-02-23 | It is a kind of based on correlation rule and the integrated recommended technology of Bayesian network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107103000A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107678858A (en) * | 2017-09-30 | 2018-02-09 | 广东欧珀移动通信有限公司 | application processing method, device, storage medium and electronic equipment |
| CN108491931A (en) * | 2018-03-29 | 2018-09-04 | 四川升拓检测技术股份有限公司 | A method of non-destructive testing precision is improved based on machine learning |
| CN109947752A (en) * | 2019-01-28 | 2019-06-28 | 西南民族大学 | A kind of automaticdata cleaning method based on DeepDive |
| CN110162694A (en) * | 2019-04-02 | 2019-08-23 | 莫毓昌 | A kind of recommender system and method based on pairs of correlation rule |
| CN110489602A (en) * | 2019-07-25 | 2019-11-22 | 上海乂学教育科技有限公司 | Knowledge point ability value predictor method, system, equipment and medium |
| CN111325254A (en) * | 2020-02-12 | 2020-06-23 | 支付宝(杭州)信息技术有限公司 | Method and device for constructing conditional relation network and processing conditional service |
| CN111723973A (en) * | 2020-05-15 | 2020-09-29 | 西安交通大学 | Learning effect optimization method based on user behavior causal relationship in MOOC log data |
| CN112149607A (en) * | 2020-10-08 | 2020-12-29 | 吉林大学 | Remote intelligent operation and maintenance method based on Bayesian algorithm |
| CN112528316A (en) * | 2020-09-18 | 2021-03-19 | 江苏方天电力技术有限公司 | Privacy protection lineage workflow publishing method based on Bayesian network |
| CN112784346A (en) * | 2021-02-07 | 2021-05-11 | 殿汇空间(上海)信息科技有限公司 | Building structure autonomous design method, system, terminal and storage medium |
| CN112800669A (en) * | 2021-01-25 | 2021-05-14 | 上海海事大学 | Multiple approximation technology error evaluation method based on probability map model in special accelerator |
| CN112822052A (en) * | 2021-01-08 | 2021-05-18 | 河海大学 | Network fault root cause positioning method based on network topology and alarm |
| CN113139079A (en) * | 2021-04-15 | 2021-07-20 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Music recommendation method and system |
| CN113256275A (en) * | 2021-07-14 | 2021-08-13 | 支付宝(杭州)信息技术有限公司 | Expert system updating method, service processing method and device |
| CN114491183A (en) * | 2022-02-15 | 2022-05-13 | 国家电网有限公司 | Intelligent data entry method based on Bayesian algorithm |
| CN114548709A (en) * | 2022-02-07 | 2022-05-27 | 昆明尽蓝信息技术有限公司 | Intelligent decision-making method and system based on Bayesian network data enabling enterprise |
| CN114971400A (en) * | 2022-06-24 | 2022-08-30 | 东南大学溧阳研究院 | User side energy storage polymerization method based on Dirichlet distribution-multinomial distribution conjugate prior |
| CN116779055A (en) * | 2023-06-26 | 2023-09-19 | 中国矿业大学(北京) | Coal composition data analysis method based on graph model |
| CN117371450A (en) * | 2023-11-08 | 2024-01-09 | 重庆市信息通信咨询设计院有限公司 | NLP-based event text feature association analysis method, system and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080027890A1 (en) * | 2006-07-31 | 2008-01-31 | Microsoft Corporation | Bayesian probability accuracy improvements for web traffic predictions |
| CN105183841A (en) * | 2015-09-06 | 2015-12-23 | 南京游族信息技术有限公司 | Recommendation method in combination with frequent item set and deep learning under big data environment |
-
2016
- 2016-02-23 CN CN201610096873.6A patent/CN107103000A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080027890A1 (en) * | 2006-07-31 | 2008-01-31 | Microsoft Corporation | Bayesian probability accuracy improvements for web traffic predictions |
| CN105183841A (en) * | 2015-09-06 | 2015-12-23 | 南京游族信息技术有限公司 | Recommendation method in combination with frequent item set and deep learning under big data environment |
Non-Patent Citations (5)
| Title |
|---|
| SUNG-SHUN WENG: "APPLYING BAYESIAN NETWORK AND ASSOCIATION RULE ANALYSIS FOR PRODUCT RECOMMENDATION", 《INTERNATIONAL JOURNAL OF ELECTRONIC BUSINESS MANAGEMENT》 * |
| 何岩: "《统计稀疏学习中的贝叶斯非参数建模方法及其应用研究》", 30 April 2014, 浙江工商大学出版社 * |
| 史会峰: "基于MCMC算法贝叶斯网络的学习", 《华北电力大学学报》 * |
| 赵守香: "《大数据分析与应用》", 31 December 2015, 航空工业出版社 * |
| 赵海丰: "关联规则挖掘及贝叶斯网表示研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107678858B (en) * | 2017-09-30 | 2019-08-23 | Oppo广东移动通信有限公司 | Application processing method, device, storage medium and electronic equipment |
| CN107678858A (en) * | 2017-09-30 | 2018-02-09 | 广东欧珀移动通信有限公司 | application processing method, device, storage medium and electronic equipment |
| CN108491931B (en) * | 2018-03-29 | 2023-12-19 | 四川升拓检测技术股份有限公司 | Method for improving nondestructive testing precision based on machine learning |
| CN108491931A (en) * | 2018-03-29 | 2018-09-04 | 四川升拓检测技术股份有限公司 | A method of non-destructive testing precision is improved based on machine learning |
| CN109947752A (en) * | 2019-01-28 | 2019-06-28 | 西南民族大学 | A kind of automaticdata cleaning method based on DeepDive |
| CN110162694A (en) * | 2019-04-02 | 2019-08-23 | 莫毓昌 | A kind of recommender system and method based on pairs of correlation rule |
| CN110489602A (en) * | 2019-07-25 | 2019-11-22 | 上海乂学教育科技有限公司 | Knowledge point ability value predictor method, system, equipment and medium |
| CN111325254A (en) * | 2020-02-12 | 2020-06-23 | 支付宝(杭州)信息技术有限公司 | Method and device for constructing conditional relation network and processing conditional service |
| CN111325254B (en) * | 2020-02-12 | 2022-06-28 | 支付宝(杭州)信息技术有限公司 | Method and device for constructing conditional relational network and processing conditional service |
| CN111723973A (en) * | 2020-05-15 | 2020-09-29 | 西安交通大学 | Learning effect optimization method based on user behavior causal relationship in MOOC log data |
| CN112528316A (en) * | 2020-09-18 | 2021-03-19 | 江苏方天电力技术有限公司 | Privacy protection lineage workflow publishing method based on Bayesian network |
| CN112528316B (en) * | 2020-09-18 | 2022-07-15 | 江苏方天电力技术有限公司 | Privacy protection lineage workflow publishing method based on Bayesian network |
| CN112149607A (en) * | 2020-10-08 | 2020-12-29 | 吉林大学 | Remote intelligent operation and maintenance method based on Bayesian algorithm |
| CN112822052A (en) * | 2021-01-08 | 2021-05-18 | 河海大学 | Network fault root cause positioning method based on network topology and alarm |
| CN112822052B (en) * | 2021-01-08 | 2022-03-29 | 河海大学 | Network fault root cause positioning method based on network topology and alarm |
| CN112800669B (en) * | 2021-01-25 | 2024-03-15 | 上海海事大学 | Method for evaluating various approximate technical errors based on probability map model in special accelerator |
| CN112800669A (en) * | 2021-01-25 | 2021-05-14 | 上海海事大学 | Multiple approximation technology error evaluation method based on probability map model in special accelerator |
| CN112784346B (en) * | 2021-02-07 | 2022-12-06 | 殿汇空间(上海)信息科技有限公司 | Building structure autonomous design method, system, terminal and storage medium |
| CN112784346A (en) * | 2021-02-07 | 2021-05-11 | 殿汇空间(上海)信息科技有限公司 | Building structure autonomous design method, system, terminal and storage medium |
| CN113139079B (en) * | 2021-04-15 | 2022-10-04 | 合肥中聚源智能科技有限公司 | Music recommendation method and system |
| CN113139079A (en) * | 2021-04-15 | 2021-07-20 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Music recommendation method and system |
| CN113256275A (en) * | 2021-07-14 | 2021-08-13 | 支付宝(杭州)信息技术有限公司 | Expert system updating method, service processing method and device |
| CN113256275B (en) * | 2021-07-14 | 2021-11-02 | 支付宝(杭州)信息技术有限公司 | Expert system updating method, service processing method and device |
| CN114548709A (en) * | 2022-02-07 | 2022-05-27 | 昆明尽蓝信息技术有限公司 | Intelligent decision-making method and system based on Bayesian network data enabling enterprise |
| CN114491183A (en) * | 2022-02-15 | 2022-05-13 | 国家电网有限公司 | Intelligent data entry method based on Bayesian algorithm |
| CN114971400A (en) * | 2022-06-24 | 2022-08-30 | 东南大学溧阳研究院 | User side energy storage polymerization method based on Dirichlet distribution-multinomial distribution conjugate prior |
| CN114971400B (en) * | 2022-06-24 | 2023-01-31 | 东南大学溧阳研究院 | User side energy storage polymerization method based on Dirichlet distribution-multinomial distribution conjugate prior |
| CN116779055A (en) * | 2023-06-26 | 2023-09-19 | 中国矿业大学(北京) | Coal composition data analysis method based on graph model |
| CN116779055B (en) * | 2023-06-26 | 2024-03-15 | 中国矿业大学(北京) | Coal composition data analysis method based on graph model |
| CN117371450A (en) * | 2023-11-08 | 2024-01-09 | 重庆市信息通信咨询设计院有限公司 | NLP-based event text feature association analysis method, system and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107103000A (en) | It is a kind of based on correlation rule and the integrated recommended technology of Bayesian network | |
| CN103064945B (en) | Based on the Situational searching method of body | |
| CN107590243B (en) | The personalized service recommendation method to be sorted based on random walk and diversity figure | |
| Ma et al. | Combining tag correlation and user social relation for microblog recommendation | |
| Zhang et al. | Enabling kernel-based attribute-aware matrix factorization for rating prediction | |
| Neto et al. | Bivariate symbolic regression models for interval-valued variables | |
| Wang et al. | Towards real-time demand-aware sequential POI recommendation | |
| Zhang et al. | Modeling the heterogeneous duration of user interest in time-dependent recommendation: A hidden semi-Markov approach | |
| CN112765480B (en) | Information pushing method and device and computer readable storage medium | |
| Wang et al. | Diversified and scalable service recommendation with accuracy guarantee | |
| Park et al. | Uniwalk: Explainable and accurate recommendation for rating and network data | |
| CN110674407A (en) | Hybrid recommendation method based on graph convolution neural network | |
| Hsu et al. | Unsupervised ranking using graph structures and node attributes | |
| CN106960044B (en) | Time perception personalized POI recommendation method based on tensor decomposition and weighted HITS | |
| CN102135905B (en) | User customization-based body matching system and method | |
| CN104484431A (en) | Multi-source individualized news webpage recommending method based on field body | |
| Zhang et al. | Dynamic scholarly collaborator recommendation via competitive multi-agent reinforcement learning | |
| Komkhao et al. | Incremental collaborative filtering based on Mahalanobis distance and fuzzy membership for recommender systems | |
| CN110119478A (en) | A kind of item recommendation method based on similarity of a variety of user feedback datas of combination | |
| Yigit et al. | Extended topology based recommendation system for unidirectional social networks | |
| Zou et al. | TrustRank: a Cold-Start tolerant recommender system | |
| Liphoto et al. | A survey on recommender systems | |
| Zhang et al. | Improving augmented reality using recommender systems | |
| Hazrati et al. | Entity representation for pairwise collaborative ranking using restricted Boltzmann machine | |
| Li et al. | Session Recommendation Model Based on Context‐Aware and Gated Graph Neural Networks |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170829 |
|
| WD01 | Invention patent application deemed withdrawn after publication |