CN102023993B - Cluster page ranking equipment and method based on clustering/classification and time - Google Patents

Abstract

The invention provides cluster page ranking equipment and method based on clustering/classification and time. The cluster page ranking equipment comprises a searcher, a cluster builder, a cluster page ranking calculator, a cluster trend generator and a cluster trend ranking device, wherein the searcher is configured to search relevant documents from data sets according to given query statements and calculate document related values of the searched documents, thus obtaining related document sets of the sequencing; the cluster builder is configured to cluster or classify the related document sets, thus obtaining a cluster; the cluster page ranking calculator based on time is configured to calculate cluster page ranking values (TCP values) based on cluster calculation on the basis of the cluster, and is a combination of document link values based on time of all the documents in the cluster and is used as a combination of the page ranking values based on time, author ranking values based on time and document library ranking values based on time of all the documents in the cluster; the cluster trend generator is configured to calculate the future TCP value of the cluster according to the TCP value; and the cluster trend ranking device is configured to sequence future TCP values, thus obtaining trend.

Description

Bunch page ranking equipment and method based on clustering/classification and time

Technical field

The present invention relates to trend analysis and document ranking optimization, more particularly, the present invention relates to bunch page ranking equipment and the method based on clustering/classification and time of the trend that can find the sub-field of a specific area and analyze and find this a little field.

Background technology

In the field of trend analysis and document ranking optimization, reference paper 1 (US20050234877A1, " System and method for searching using a temporal dimension ") disclose Query Result has been carried out upper sequence of time, consider author authority and the authority of publishing house of each result document during sequence.Wherein, time-based rank has been used based on document publication time and poor aging function and a meeting make the time of delivering document value more of a specified duration reduce faster ratio now.But this scheme is just calculated the time-based page rank value of single document and is predicted its trend, and is used for prediction single document trend.

Reference paper 2 (US20080071763A1, " Dynamic updating of display andranking for search results ", EMC CORP) disclose again to searching order, it has adopted the page rank value that the first of Search Results is sorted, with the second portion of clustering method insertion Search Results in the first result.

Reference paper 3 (US20070143300A1, " System and method for monitoringevolution over time of temporal content ", ASK JEEVES INC) query statement of inputting according to the user is disclosed, receive and store time-based content, analysis entities occurs determining time-based content trend.

Reference paper 4 (US20060089924A1, " Document categorisation system ") discloses a document classification system, comprises a cluster device and a filtering module.System comprises that one is used for determining document classification trend and the trend analysis device of determining new bunch over time, and this system can be used for the merge module that spreadsheet is used.

in addition, at reference paper 5 (Hassan Sayyadi, Lise Getoory, " FutureRank:RankingScientific Articles by Predicting their Future PageRank ", Society for Industrialand Applied Mathematics (SIAM) Data Mining Conference (SDM 2009)) in, its method will be quoted, the future trend that author and publication time combine and effectively the science article sorted and predict article, but the method does not relate to the Classification and clustering method, to the value of increase on the page rank value of Search Results with time correlation, and only be used for prediction single document trend.

In addition, also have some following systems in prior art, wherein CiteSpace be one with the trend visualization system.But this system has only considered the cooperative relationship between the author, future trend is not predicted.

ThemeRiver is a trend, pattern discrimination system.But it does not consider the context of network, Future Data is not predicted yet, only has the statistics to historical data.

And Google trend analysis Google Webpage search is used for calculating the number of times that the user makes word.Because do not consider user's authority, thus make word do not represent often it authority.

In sum, in the conventional method, the user is at first according to given field search relevant documentation, then the number of documents in the known sub-field of simple computation or manually sum up document content, analysis time and anticipation trend just.Calculate the content relation that number of documents can not disclose document, and the manual analysis method is consuming time and not objective.

Summary of the invention

Therefore, the objective of the invention is to solve above-mentioned shortcoming of the prior art, thereby for a given field, find the sub-field that is just occurring and the trend of predicting this a little field.

In order to solve the problems of the prior art, this patent has proposed bunch page ranking equipment and the method based on the time of clustering/classification.It can automatically be found sub-field and adopt time-based linking relationship to calculate objectively the relative importance in sub-field, i.e. trend.

According to an aspect of the present invention, a kind of bunch page ranking equipment based on clustering/classification and time is provided, comprise: searcher, be configured to receive the given query statement of user, according to the document correlation of described query statement from data centralization search relevant documentation and the calculating document of searching for, thereby the set of relevant documents that obtains sorting, and with described set of relevant documents output; Cluster and grow up to be a useful person, be configured to receive from the set of relevant documents of described searcher output, obtain bunch thereby described set of relevant documents is carried out cluster or classification, and with described bunch of output; Time-based bunch of page rank counter, be configured to from described cluster grow up to be a useful person receive export bunch, based on a bunch time-based bunch of page rank value of calculating, and export described time-based bunch of page rank value, described time-based bunch of page rank value is the combination of the time-based document links value of all documents in described bunch, and as the combination of time-based page rank value, time-based author's rank value and the time-based document library rank value of all documents in described bunch; Bunch trend maker, be configured to receive described time-based bunch of page rank value from described time-based bunch of page rank counter, and according to the time-based bunch of page rank value in future of described time-based bunch of page rank value compute cluster, and export the time-based bunch of page rank value in described future; With a bunch trend rank device, be configured to receive from bunch trend maker the time-based bunch of page rank value in described future, thereby and the time-based bunch of page rank value in described future sorted obtain trend.

In the present invention, described time-based bunch of page rank counter further comprises: page rank value computing unit is configured to the time-based page rank value of document in compute cluster; Author's rank value computing unit is configured to time-based author's rank value of document in compute cluster; Document library rank value computing unit is configured to the time-based document library rank value of document in compute cluster; And weighted units, by calculate time-based page rank value from described page rank value computing unit, from time-based author's rank value of described author's rank value computing unit with from the weighted sum of the time-based document library rank value of described document library rank value computing unit, and in accumulation bunch, the weighted sum of all documents produces described time-based bunch of page rank value.

In the present invention, described page rank value computing unit passes through in all time T _iThe weighted sum of time-based page rank changing value of document calculate the document in time T _nTime-based page rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iThe weight of time-based page rank changing value of document be time T _iTo T _nThe aging function of mistiming, and in time T _iThe time-based page rank changing value of document be that all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, and described all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that described searcher obtains.

In the present invention, time-based author's rank value of the document that described author's rank value computing unit calculates is time-based author's rank value sum of all authors of this document, wherein, and by in all time T _iAuthor's the weighted sum of time-based author's rank changing value calculate this author in time T _nTime-based author's rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iAuthor's the weight of time-based author's rank changing value be time T _iTo T _nThe aging function of mistiming, and in time T _iTime-based author's rank value of author write for this author all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, described author write all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that described searcher obtains.

In the present invention, the time-based document library rank value of document that described document library rank value computing unit calculates is the time-based document library rank value of the document place document library, and document library is in time T _nThe time time-based document library rank value be in all time T _iThe time-based document library rank changing value sum of document library, i=1 wherein ..., n, T _i≤ T _n,, and in time T _iThe time-based document library rank changing value of document library be that in the document storehouse, all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, in described document library, all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that described searcher obtains.

In the present invention, described bunch of trend maker comes the time-based bunch of page rank value in the future of compute cluster according to the changing value of time-based bunch of page rank value or the rate of change of time-based bunch of page rank value.

In the present invention, described bunch of trend rank device according to bunch time-based bunch of page rank value in future and the difference between current time-based bunch of page rank value to bunch sorting, wherein, described difference is larger, rank is higher.

According to a further aspect in the invention, a kind of bunch page rank method based on clustering/classification and time is provided, comprise: search step, it is according to the document correlation of being searched for relevant documentation and calculating the document of searching for from data centralization by the given query statement of user, thus the set of relevant documents that obtains sorting; Cluster into step, thereby it carries out cluster or classification to described set of relevant documents and obtains bunch; Time-based bunch of page rank calculation procedure, it is based on the described bunch time-based bunch of page rank value of calculating that obtains in step of clustering into, described time-based bunch of page rank value is the combination of the time-based document links value of all documents in described bunch, and as the combination of time-based page rank value, time-based author's rank value and the time-based document library rank value of all documents in described bunch; Bunch trend generates step, and it is according to the time-based bunch of page rank value in the future of described time-based bunch of page rank value compute cluster; With a bunch trend rank step, thereby its time-based bunch of page rank value to described future sorts and obtains trend.

In the present invention, described time-based bunch of page rank calculation procedure further comprises: page rank value calculation procedure, the time-based page rank value of document in compute cluster; Author's rank value calculation procedure, time-based author's rank value of document in compute cluster; Document library rank value calculation procedure, the time-based document library rank value of document in compute cluster; With the weighting step, by calculate time-based page rank value from described page rank value computing unit, from time-based author's rank value of described author's rank value computing unit with from the weighted sum of the time-based document library rank value of described document library rank value computing unit, and in accumulation bunch, the weighted sum of all documents produces described time-based bunch of page rank value.

In the present invention, by in all time T _iThe weighted sum of time-based page rank changing value of document calculate the document in time T _nTime-based page rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iThe weight of time-based page rank changing value of document be time T _iTo T _nThe aging function of mistiming, and in time T _iThe time-based page rank changing value of document be that all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, and described all point to the document of the document in time T _iWeight and the described search step of time-based page rank changing value in the document correlation that obtains be directly proportional.

In the present invention, time-based author's rank value of described document is time-based author's rank value sum of all authors of this document, wherein, and by in all time T _iAuthor's the weighted sum of time-based author's rank changing value calculate this author in time T _nTime-based author's rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iAuthor's the weight of time-based author's rank changing value be time T _iTo T _nThe aging function of mistiming, and in time T _iTime-based author's rank value of author write for this author all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, described author write all point to the document of the document in time T _iWeight and the described search step of time-based page rank changing value in the document correlation that obtains be directly proportional.

In the present invention, the time-based document library rank value of document that described document library rank value computing unit calculates is the time-based document library rank value of the document place document library, and document library is in time T _nThe time time-based document library rank value be in all time T _iThe time-based document library rank changing value sum of document library, i=1 wherein ..., n-1, T _i≤ T _n,, and in time T _iThe time-based document library rank changing value of document library be that in the document storehouse, all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, in described document library, all point to the document of the document in time T _iWeight and the described search step of time-based page rank changing value in the document correlation that obtains be directly proportional.

In the present invention, come the time-based bunch of page rank value in the future of compute cluster according to the changing value of time-based bunch of page rank value or the rate of change of time-based bunch of page rank value.

In the present invention, according to bunch time-based bunch of page rank value in future and the difference between current time-based bunch of page rank value to bunch sorting, wherein, described difference is larger, rank is higher.

By the application's bunch page ranking equipment and method based on clustering/classification and time, can automatically find sub-field and can automatic Prediction field trend, find following popular sub-field.And, come analytic trend owing to adopting based on page rank, can calculate objectively and analytic trend, when having avoided adopting quoting method due to the past by a large amount of quoting, older document ranking can be higher, and because the less new document of quoting can the very low problem of rank, thereby improved the accuracy of trend analysis.

Description of drawings

Fig. 1 is the block diagram that illustrates according to the overall arrangement of bunch page ranking equipment based on clustering/classification and time of the present invention;

Fig. 2 is the block diagram that illustrates according to the configuration based on time-based bunch of page rank counter in bunch page ranking equipment of clustering/classification and time of the present invention;

Fig. 3 is the process flow diagram that illustrates according to bunch page rank method based on clustering/classification and time of the present invention;

Fig. 4 illustrates according to of the present invention based on calculating the process flow diagram of time-based bunch of page rank value step in bunch page rank method of clustering/classification and time.

Embodiment

Describe specific embodiments of the invention in detail below in conjunction with accompanying drawing.

Fig. 1 is the block diagram that illustrates according to the overall arrangement of bunch page ranking equipment based on clustering/classification and time of the present invention.As described in Figure 1, bunch page ranking equipment 100 based on clustering/classification and time of the present invention comprise searcher 101, cluster grow up to be a useful person 102, time-based bunch of page rank counter 103, bunch trend maker 104 and bunch trend rank device 105.Below, will each assembly based on bunch page ranking equipment 100 of clustering/classification and time be described below in detail.

In described bunch page ranking equipment 100 based on clustering/classification and time, searcher 101 receives the given query statement of user, according to described query statement search full-text index, also calculate the relevant documentation value of the document of searching for from data centralization search relevant documentation, thereby by the set of relevant documents that this relevant documentation value is sorted and obtains sorting, and set of relevant documents is exported.Here, the method for the employing of searcher 101 can comprise statistical method or based on the method for link analysis or their combination.

For example, searcher 101 can adopt the BM25 algorithm (referring to Ed Greengras, " InformationRetrieval:A Survey ", 30November 2000) calculate the correlativity score of document in given query statement and document library, thus corresponding search rank obtained.Here, a given document sets, user input query statement Q, namely a field is described, for example " robotization is printed by office ", the correlativity score score (d, Q) of document d is calculated by following formula:

$\mathrm{score}(d,Q)=\underset{t\∈Q}{\mathrm{\Σ}}\frac{\mathrm{tf}}{K+\mathrm{tf}}\frac{\mathrm{qtf}}{\mathrm{qtf}+k_3}\log (k_2\frac{N}{N_t}+1.0)$ Formula 1

Wherein, t is the word in inquiry Q, and tf is the number of times that t occurs in document d, and qtf is the number of times that t occurs in inquiry Q, and N is the number of files in document library, N _tThe number of files that comprises word t in document library, k ₂And k ₃Parameter, k for example ₂=0.5, k ₃=1000, K is defined as follows

$K=k_1((1-b)+b\frac{l}{\mathrm{avg}\_l})$ Formula 2

Wherein l is the length of document d, and implication is the sum of word in document, and avg_l is the average document length of document library, and namely all document length sums are divided by document number, k ₁With b be parameter, k for example ₁=1.2, b=0.75.

Wherein, the numerical value of score (d, Q) is higher, and the degree of correlation of expression the document d and query statement is higher.

Like this, searcher 101 obtains the degree of correlation of document d and query statement, that is, and and relevant documentation value, thereby and the set of relevant documents that sorts and obtain sorting for document sets according to the degree of correlation.

Clustering grows up to be a useful person 102 receives from the set of relevant documents of searcher 101 outputs, and it is carried out cluster or classification.Grow up to be a useful person in 102 clustering, the method of cluster can be K-averaging method clustering algorithm, fuzzy c-averaging method clustering algorithm, and Graph-theoretical Approach in any one or a plurality of combinations, and the method for classification is the Document Classification Method based on supervision, non-supervisory Document Classification Method, any one in semi-supervised Document Classification Method or a plurality of combinations.

For example, the grow up to be a useful person subset of document of 102 some N that rank in the result of one query is forward of clustering is carried out cluster, and to form different bunches, the document data in each bunch belongs to same feature or theme.Grow up to be a useful person 102 employing K-averaging method clustering algorithms when clustering (referring to Lloyd, S.P. (1957), " Lastsquare quantization in PCM ", Bell Telephone Laboratories Paper Published injournal much later:Lloyd., S.P. (1982)), when being used for N the most forward search result document cluster of rank generated bunch, this algorithm steps comprises:

(1) select clustering parameter k, wherein k can be defined as k=(N/2) 1/2;

(2) select at random k document as k initial classes;

(3) to each class, and 10 words that its occurrence number is maximum (t1 ..., t10) be defined as its cluster centre;

(4) calculate respectively distance between each document and each class

Distance $(d,c)=\frac{s1*l1+s2*l2+...+s10*l10}{\sqrt{s1*s1+...+s10*s10}*\sqrt{l1*l1+...+l10*l10}}$ Formula 3

S1 wherein, s2 ..., s10 is respectively 10 centre word t1 of class c ..., the number of times that t10 occurs, 11,12 ..., 110 is respectively 10 centre word t1 in document d ..., the number of times that t10 occurs, document d will belong to nearest class;

(5) circulation step (3) and (4) until each cluster no longer change.

Like this, cluster and grow up to be a useful person 102 by the set of relevant documents of searcher 101 sequence is carried out cluster and classification, obtain bunch, be i.e. sub-topics.

Time-based bunch of page rank counter 103 grown up to be a useful person from clustering and 102 obtained bunch, thereby based on bunch calculating a time-based bunch of page rank value.The configuration of time-based bunch of page rank counter 103 of the present invention will be described in detail in conjunction with Fig. 2 hereinafter.

Bunch trend maker 104 comes the time-based bunch of page rank value in the future of compute cluster, thereby calculates the trend of each bunch according to time-based bunch of page rank (TCP) value that time-based bunch of page rank counter 103 calculates.The method that described bunch of trend maker 104 adopts can comprise based on the Forecasting Methodology of TCP changing value with based on the Forecasting Methodology of TCP rate of change.

For example, when calculating future trend with the TCP changing value,

${\mathrm{\ΔTCP}}_{T_n}\left(c\right)=T{\mathrm{CP}}_{T_n}\left(c\right)-{\mathrm{TCP}}_{T_{n-1}}\left(c\right)$ Formula 4

${\mathrm{\ΔTCP}}_{T_{n-1}}\left(c\right)={\mathrm{TCP}}_{T_{n-1}}\left(c\right)-{\mathrm{TCP}}_{T_{n-2}}\left(c\right)$ Formula 5

${\mathrm{\ΔTCP}}_{T_{n+1}}\left(c\right)=2*{\mathrm{\ΔTCP}}_{T_n}\left(c\right)-{\mathrm{\ΔTCP}}_{T_{n-1}}\left(c\right)$ Formula 6

Here,

That bunch c is in time T _iTCP value increment.(i＝n-2，n-1，n，n+1)

At last, bunch trend rank device 105 is according to bunch Trend value of bunch trend maker 104 compute clusters, and namely the TCP value increment that calculates of bunch trend maker 104 is to bunch sorting, and wherein, TCP value increment is larger, and rank is higher.Here, the TCP value in described TCP value increment future of being bunch and bunch current TCP value between difference.Here, rank is high bunch is to be about to the popular sub-field that occurs.

Below, be described in further detail as follows with reference to Fig. 2 for time-based bunch of page rank counter 103 of the present invention.

As shown in Figure 2, time-based bunch of page rank counter 103 of the present invention comprises: page rank value computing unit 201, for the time-based page rank value of compute cluster document; Author's rank value computing unit 202 is for time-based author's rank value of compute cluster document; Document library rank value computing unit 203 is for the time-based document library rank value of compute cluster document; With weighted units 204, by calculating time-based page rank value from described page rank value computing unit 201, producing described time-based bunch of page rank value from time-based author's rank value of described author's rank value computing unit 202 with from the weighted sum of the time-based document library rank value of described document library rank value computing unit 203.Below, will each assembly of time-based bunch of page rank counter 103 be described below in detail.

According to the present invention, the time-based page rank value that time-based bunch of page rank counter 103 calculates be bunch in the combination of time-based document links value of all documents.The time-based document links value of document is the time-based page rank value of document, the combination of time-based author's rank value of document and the time-based document library rank value of document.Therefore, in the application's time-based bunch of page rank counter 103, time-based page rank value by document, the weighted sum of time-based author's rank value of document and the time-based document library rank value of document is calculated time-based bunch of page rank value, i.e. TCP value.

In the present invention, page rank value computing unit 201 passes through in all time T _iThe weighted sum of time-based page rank changing value of document calculate the document in time T _nTime-based page rank value, i=0 wherein, 1 ..., n-1, T _i＜T _n, and described in all time T _iThe weight of time-based page rank changing value of document be time T _iTo T _nThe aging function of mistiming, and in time T _iThe time-based page rank changing value of document be that all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, and described all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that searcher 101 obtains.

For example, order

For document A in time T _nThe time time-based page rank value,

${\mathrm{PR}}_{\mathrm{Tn}}\left(A\right)=(1-d)+d*(w_{T_1}*{\mathrm{\ΔPR}}_{T_1}\left(A\right)+...+w_{T_n}*{\mathrm{\ΔPR}}_{T_n}\left(A\right))$

Formula 7

Wherein, d is a constant and 0＜d＜1, for example, and d=0.5, and Be time T _i(the page rank value variable quantity of document A during 0＜i≤n) (, time-based page rank changing value).

For

Weight, satisfy

$w_{T_i}={\mathrm{\α}}^{(T_n-\mathrm{Ti})/12}$ Formula 8

Wherein α is a constant and 0＜α＜1, for example, and α=0.5.

For time T _i(the page rank value variable quantity of document A during 0＜i≤n),

${\mathrm{\ΔPR}}_{\mathrm{Ti}}\left(A\right)=S_{p_1}*\frac{{\mathrm{\ΔPR}}_{T_i}\left(p_1\right)}{C\left(p_1\right)}+...+S_{p_k}*\frac{{\mathrm{\ΔPR}}_{T_i}\left(p_k\right)}{C\left(p_k\right)}$ Formula 9

Wherein

From T _i-1To T _iPoint to the document P of document A _jThe page rank changing value, 0＜j≤k, and k here is from T _i-1To T _iPoint to the number of all documents of document A,

To carry out the document correlation that obtains after normalized search and C (p by searcher 101 _j) be document p _jOut-degree.

In addition, remove outside said method of the present invention, can also use as disclosed method in US20050234877A1, or as the page rank value of single document in disclosed method compute cluster in FutureRank:Ranking Scientific Articles by Predicting theirFuture PageRank one literary composition.

Time-based author's rank value of the document that author's rank value computing unit 202 calculates is time-based author's rank value sum of all authors of this document, wherein, and by in all time T _iAuthor's the weighted sum of time-based author's rank changing value calculate this author in time T _nTime-based author's rank value, i=1 wherein ..., n-, T _i≤ T _n, and described in all time T _iAuthor's the weight of time-based author's rank changing value be time T _iTo T _nThe aging function of mistiming, and in time T _iTime-based author's rank value of author write for this author all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, described author write all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that searcher 101 obtains.

For example, make a ₁..., a _mThe author of document A.

For document A in time T _nThe time time-based author's rank value, For document A in time T _nThe time author a ₁Rank value (wherein, 0＜1≤m, and m is the author's of document A number), have

${\mathrm{AP}}_{T_n}\left(A\right)={\mathrm{AP}}_{T_n}\left(a_1\right)+...+{\mathrm{AP}}_{T_n}\left(a_m\right),$ Formula 10

And, for document A in time T _nThe time author a ₁Rank value

${\mathrm{AP}}_{\mathrm{Tn}}\left(a_l\right)={w^{\′}}_{T_1}*{\mathrm{\ΔAP}}_{T_1}\left(a_l\right)+...+{w^{\′}}_{T_n}*{\mathrm{\ΔAP}}_{T_n}\left(a_l\right)$ Formula 11

Wherein,

Author a ₁In time T _iThe time rank increment (, time-based author's rank changing value of document), wherein, 0＜i≤n.

For

Weight, satisfy

${w^{\′}}_{T_i}={\mathrm{\β}}^{(T_n-\mathrm{Ti})/12}$ Formula 12

Wherein, β is a constant and 0＜β＜1, for example β=0.5.

And for author a _kIn time T _iThe time the rank increment, satisfy

${\mathrm{\ΔAP}}_{\mathrm{Ti}}\left(a_l\right)=\frac{S_{p1}*{\mathrm{\ΔPR}}_{T_i}\left(p_1\right)+...+S_{p_k}*{\mathrm{\ΔPR}}_{T_i}\left(p_k\right)}{\text{k}}$ Formula 13

Wherein,

From T _i-1To T _iPoint to the document P of document A _jThe page rank changing value, 0＜j≤k, and k here is from T _i-1To T _iPoint to the author a of document A ₁The number of all documents of delivering,

It is the document correlation that obtains after normalized search.Here p ₁..., p _kAuthor a ₁The document of delivering.

In addition, the time-based page rank value of document as 201 calculating of page rank value computing unit, remove outside said method of the present invention, can also use as disclosed method in US20050234877A1, or as author's rank value of single document in disclosed method compute cluster in FutureRank:Ranking Scientific Articles by Predicting their Future PageRank one literary composition.

The time-based document library rank value of the document that document library rank value computing unit 203 calculates is the time-based document library rank value of the document place document library, and document library is in time T _nThe time time-based document library rank value be in all time T _iThe time-based document library rank changing value sum of document library, i=1 wherein ..., n, T _i≤ T _n,, and in time T _iThe time-based document library rank changing value of document library be that in the document storehouse, all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, in described document library, all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that described searcher obtains.

For example, make that p is the document library of document A, The rank value of document A place document library, and

The rank value of document library p when time T n,

${\mathrm{JP}}_{T_n}\left(A\right)={\mathrm{JP}}_{T_n}\left(p\right)$ Formula 14

Rank value for document library p when the time T n

${\mathrm{JP}}_{\mathrm{Tn}}\left(p\right)={\mathrm{\ΔJP}}_{T_1}\left(p\right)+...+{\mathrm{\ΔJP}}_{T_n}\left(p\right)$ Formula 15

Here

The rank value increment of document library p when being time T i (, document library rank changing value), 0＜i≤n wherein,

Be expressed from the next

${\mathrm{\ΔJP}}_{\mathrm{Ti}}\left(p_j\right)=\frac{S_{p_1}*{\mathrm{\ΔPR}}_{T_i}\left(p_1\right)+...+S_{p_k}*{\mathrm{\ΔPR}}_{T_i}\left(p_k\right)}{k}$ Formula 16

Wherein

From T _i-1To T _iPoint to the document P of document A _jThe page rank changing value, 0＜j≤k, and k here is from T _i-1To T _iThe number of all documents in the document library P of sensing document A,

The document correlation that obtains after normalized search, p ₁..., p _kIt is the document in document library P.

And, as the time-based page rank value of document of page rank value computing unit 201 calculating and the time-based author's rank value of document of author's rank value computing unit 202 calculating, remove outside said method of the present invention, can also use as disclosed method in US20050234877A1, or as the document library rank value of single document in disclosed method compute cluster in FutureRank:Ranking Scientific Articles by Predicting their Future PageRank one literary composition.

Weighted units 204 time-based T _nThe time the time-based page rank value of document

Time-based author's rank value

With time-based document library rank value

Calculate time-based bunch of document ranking value.

For example, another

The page rank value of bunch c when time T n,

${\mathrm{TCP}}_{T_n}\left(c\right)=\underset{A\∈c}{\mathrm{\Σ}}(k_1*{\mathrm{PR}}_{T_n}\left(A\right)+k_2*{\mathrm{AP}}_{T_n}\left(A\right)+k_3*{\mathrm{JP}}_{T_n}\left(A\right))$ Formula 17

K wherein ₁, k ₂, k ₃The constant parameter greater than zero, for example, k ₁=0.5, k ₂=0.3 and k ₃=0.2.

Like this, the weighted units 204 of time-based bunch of page rank counter 103 by the invention described above is calculated the weighted sum of the time-based document library rank value that time-based page rank value that page rank value computing units 201 calculate, time-based author's rank value that author's rank value computing unit 202 calculates and document library rank value computing unit 203 calculate, and for bunch in the described weighted sum of all documents accumulation, thereby obtain time-based bunch of page rank value.

in addition, those skilled in the art also are appreciated that, disclosed method in using as US20050234877A1, or as in FutureRank:Ranking Scientific Articles by Predicting their FuturePageRank one literary composition during disclosed method, by accumulation calculate bunch in the page rank value of single document, in bunch author's rank value of single document and bunch in the document library rank value of single document obtain the time-based document links value of document, and to bunch in the described document links value of all documents accumulation, also can obtain described time-based bunch of page rank value.

Fig. 3 is the process flow diagram that illustrates according to bunch page rank method based on clustering/classification and time of the present invention.As shown in Figure 3, bunch page rank method based on clustering/classification and time of the present invention comprises step:

Search step S101, in this step, receive the given query statement of user, according to described query statement search full-text index, also calculate the relevant documentation value of the document of searching for from data centralization search relevant documentation, thereby by the set of relevant documents that this relevant documentation value is sorted and obtains sorting, and set of relevant documents is exported.Here, the method that described search step adopts can comprise statistical method or based on the method for link analysis or their combination, above this is being described in the description about searcher 101, therefore, will repeat no more at this.

Cluster into step S102, in this step, be received in the set of relevant documents that produces in search step S101, and it is carried out cluster or classification.The method of the cluster that adopts in clustering into step S102 can be K-averaging method clustering algorithm, fuzzy c-averaging method clustering algorithm, and Graph-theoretical Approach in any one or a plurality of combinations, and the method for the classification of adopting is the Document Classification Method based on supervision, non-supervisory Document Classification Method, any one in semi-supervised Document Classification Method or a plurality of combinations, this is described in 102 the description of above growing up to be a useful person about clustering, therefore, will repeat no more at this.

Time-based bunch of page rank calculation procedure S103 is in this step, based on the time-based bunch of page rank value of bunch calculating of clustering into that step S102 obtains.The treatment scheme of time-based bunch of page rank calculation procedure of the present invention will be described in detail in conjunction with Fig. 4 hereinafter.

Bunch trend generates step S104, in this step, according to time-based bunch of page rank (TCP) value that time-based bunch of page rank calculation procedure S103 calculates, come the time-based bunch of page rank value in the future of compute cluster, thereby calculate the trend of each bunch.Described bunch of trend generates method that step 104 adopts and can comprise based on the Forecasting Methodology of TCP changing value with based on the Forecasting Methodology of TCP rate of change, this is described in above about bunch description of trend maker 104, therefore, will repeat no more at this.

Bunch trend rank step S105, in this step, according to generate in bunch trend that step S104 calculates bunch bunch Trend value (i.e. bunch TCP value increment that trend generation step S104 calculates) to bunch sorting, wherein, TCP value increment is larger, rank is higher.Here, the TCP value in described TCP value increment future of being bunch and bunch current TCP value between difference.And rank is high bunch is to be about to the popular sub-field that occurs.

Below, describe the treatment scheme of time-based bunch of page rank calculation procedure S103 in detail with reference to Fig. 4.Wherein, Fig. 4 illustrates according to of the present invention based on calculating the process flow diagram of time-based bunch of page rank value step in bunch page rank method of clustering/classification and time.

As shown in Figure 4, the step of time-based bunch of page rank value of described calculating comprises: page rank value calculation procedure S201, and it is used for the time-based page rank value of compute cluster document; Author's rank value calculation procedure S202 is for time-based author's rank value of compute cluster document; Document library rank value calculation procedure S203 is for the time-based document library rank value of compute cluster document; With weighting step S204, the weighted sum of the time-based document library rank value of calculating by time-based author's rank value of calculating the time-based page rank value calculated at described page rank value calculation procedure S201, calculating at described author's rank value calculation procedure S202 with at described document library rank value calculation procedure S203, and for bunch in the described weighted sum of all documents accumulations produce described time-based bunch of page rank value.Below, will be described further as follows to above each step.

In the present invention, page rank value calculation procedure S201 passes through in all time T _iThe weighted sum of time-based page rank changing value of document calculate the document in time T _nTime-based page rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iThe weight of time-based page rank changing value of document be time T _iTo T _nThe aging function of mistiming, and in time T _iThe time-based page rank changing value of document be that all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, and described all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that obtains at search step S101.Provided the example of the time-based page rank value of document in the compute cluster in above-mentioned description about page rank value computing unit 201, therefore do not repeated them here.

In addition, remove outside said method of the present invention, can also use as disclosed method in US20050234877A1, or as the page rank value of single document in disclosed method compute cluster in FutureRank:Ranking Scientific Articles by Predicting their FuturePageRank one literary composition.

Time-based author's rank value of the document that author's rank value calculation procedure S202 calculates is time-based author's rank value sum of all authors of this document, wherein, and by in all time T _iAuthor's the weighted sum of time-based author's rank changing value calculate this author in time T _nTime-based author's rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iAuthor's the weight of time-based author's rank changing value be time T _iTo T _nThe aging function of mistiming, and in time T _iTime-based author's rank value of author write for this author all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, described author write all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that obtains at search step S101.Provided the example of time-based author's rank value of document in the compute cluster in above-mentioned description about author's rank value computing unit 202, therefore do not repeated them here.

In addition, as the time-based page rank value of document of calculating in page rank value calculation procedure S201, remove outside said method of the present invention, can also use as disclosed method in US20050234877A1, or as author's rank value of single document in disclosed method compute cluster in FutureRank:Ranking Scientific Articles by Predicting their FuturePageRank one literary composition.

The time-based document library rank value of document that document library rank value calculation procedure S203 calculates is the time-based document library rank value of the document place document library, and document library is in time T _nThe time time-based document library rank value be in all time T _iThe time-based document library rank changing value sum of document library, i=1 wherein ..., n, T _i≤ T _n,, and in time T _iThe time-based document library rank changing value of document library be that in the document storehouse, all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, in described document library, all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that search step S101 obtains.Provided the example of the time-based document library rank value of document in the compute cluster in above-mentioned description about document library rank value computing unit 203, therefore do not repeated them here.

And, as the time-based page rank value of document of page rank value calculation procedure S201 calculating and the time-based author's rank value of document of author's rank value computing unit S202 calculating, remove outside said method of the present invention, can also use as disclosed method in US20050234877A1, or as the document library rank value of single document in disclosed method compute cluster in FutureRank:Ranking Scientific Articles by Predicting their Future PageRank one literary composition.

Subsequently, at weighting step S204, the weighted sum of the time-based document library rank value that the time-based page rank value that calculating page rank value calculation procedure S201 calculates, time-based author's rank value that author's rank value calculation procedure S202 calculates and document library rank value calculation procedure S203 calculate, and for bunch in the described weighted sum of all documents accumulation, thereby obtain time-based bunch of page rank value.Provided the example that calculates time-based bunch of page rank value by weighted sum above-mentioned in about the description of weighted units 204, therefore do not repeated them here.

In addition, those skilled in the art also are appreciated that, disclosed method in using as US20050234877A1, or as in FutureRank:Ranking Scientific Articles by Predicting their FuturePageRank one literary composition during disclosed method, by accumulation calculate bunch in page rank value, author's rank value and the document library rank value of single document, can obtain the time-based document links value of described document, and, by the document links value of all documents in accumulation bunch, also can obtain a bunch page rank value.

In sum, by bunch page ranking equipment and the method based on clustering/classification and time of the present invention, can automatically find sub-field and can automatic Prediction field trend, find following popular sub-field.And, come analytic trend owing to adopting based on page rank, can calculate objectively and analytic trend, when having avoided adopting quoting method due to the past by a large amount of quoting, older document ranking can be higher, and because the less new document of quoting can the very low problem of rank, thereby improved the accuracy of trend analysis.

The sequence of operations that illustrates in instructions can be carried out by the combination of hardware, software or hardware and software.When carrying out this sequence of operations by software, can be installed to computer program wherein in the storer in the computing machine that is built in specialized hardware, make computing machine carry out this computer program.Perhaps, can be installed to computer program in the multi-purpose computer that can carry out various types of processing, make computing machine carry out this computer program.

For example, can be pre-stored in hard disk or ROM (ROM (read-only memory)) as recording medium computer program.Perhaps, can be temporarily or for good and all storage (record) computer program in removable recording medium, such as floppy disk, CD-ROM (compact disc read-only memory), MO (magneto-optic) dish, DVD (digital versatile disc), disk or semiconductor memory.Can so removable recording medium be provided as canned software.

The present invention has been described in detail with reference to specific embodiment.Yet clearly, in the situation that do not deviate from spirit of the present invention, those skilled in the art can carry out change and replace embodiment.In other words, the present invention is open with the form of explanation, rather than explains with being limited.Judge main idea of the present invention, should consider appended claim.

Claims (9)

1. bunch page ranking equipment based on clustering/classification and time comprises:

Searcher is configured to receive the given query statement of user, according to the document correlation of described query statement from data centralization search relevant documentation and the calculating document of searching for, thus the set of relevant documents that obtains sorting, and with described set of relevant documents output;

Cluster and grow up to be a useful person, be configured to receive from the set of relevant documents of described searcher output, obtain bunch thereby described set of relevant documents is carried out cluster or classification, and with described bunch of output;

Time-based bunch of page rank counter, be configured to from described cluster grow up to be a useful person receive export bunch, based on a bunch time-based bunch of page rank value of calculating, and export described time-based bunch of page rank value, described time-based bunch of page rank value is the combination of the time-based document links value of all documents in described bunch, and as the combination of time-based page rank value, time-based author's rank value and the time-based document library rank value of all documents in described bunch;

Bunch trend maker, be configured to receive described time-based bunch of page rank value from described time-based bunch of page rank counter, and according to the time-based bunch of page rank value in future of described time-based bunch of page rank value compute cluster, and export the time-based bunch of page rank value in described future; With

Bunch trend rank device is configured to receive from bunch trend maker the time-based bunch of page rank value in described future, thereby and the time-based bunch of page rank value in described future sorted obtain trend,

Wherein, described time-based bunch of page rank counter further comprises:

Page rank value computing unit is configured to the time-based page rank value of document in compute cluster;

Author's rank value computing unit is configured to time-based author's rank value of document in compute cluster;

Document library rank value computing unit is configured to the time-based document library rank value of document in compute cluster; With

Weighted units, by calculate time-based page rank value from described page rank value computing unit, from time-based author's rank value of described author's rank value computing unit with from the weighted sum of the time-based document library rank value of described document library rank value computing unit, and in accumulation bunch, the weighted sum of all documents produces described time-based bunch of page rank value.

2. equipment as claimed in claim 1, wherein, described page rank value computing unit is by in all time T _iThe weighted sum of time-based page rank changing value of document calculate the document in time T _nTime-based page rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iThe weight of time-based page rank changing value of document be time T _iTo T _nThe aging function of mistiming, and in time T _iThe time-based page rank changing value of document be that all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, and described all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that described searcher obtains.

3. equipment as claimed in claim 1, wherein, time-based author's rank value of the document that described author's rank value computing unit calculates is time-based author's rank value sum of all authors of this document, wherein, by in all time T _iAuthor's the weighted sum of time-based author's rank changing value calculate this author in time T _nTime-based author's rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iAuthor's the weight of time-based author's rank changing value be time T _iTo T _nThe aging function of mistiming, and in time T _iTime-based author's rank value of author write for this author all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, described author write all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that described searcher obtains.

4. equipment as claimed in claim 1, wherein, the time-based document library rank value of document that described document library rank value computing unit calculates is the time-based document library rank value of the document place document library, and document library is in time T _nThe time time-based document library rank value be in all time T _iThe time-based document library rank changing value sum of document library, i=1 wherein ..., n, T _i≤ T _n, and in time T _iThe time-based document library rank changing value of document library be that in the document storehouse, all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, in described document library, all point to the document of the document in time T _iThe weight of time-based page rank changing value be directly proportional to the document correlation that described searcher obtains.

5. equipment as claimed in claim 1, wherein, described bunch of trend maker comes the time-based bunch of page rank value in the future of compute cluster according to the changing value of time-based bunch of page rank value or the rate of change of time-based bunch of page rank value.

6. bunch page rank method based on clustering/classification and time comprises:

Search step, it is according to the document correlation of being searched for relevant documentation and calculating the document of searching for from data centralization by the given query statement of user, thus the set of relevant documents that obtains sorting;

Cluster into step, thereby it carries out cluster or classification to described set of relevant documents and obtains bunch;

Time-based bunch of page rank calculation procedure, it is based on the described bunch time-based bunch of page rank value of calculating that obtains in step of clustering into, described time-based bunch of page rank value is the combination of the time-based document links value of all documents in described bunch, and as the combination of time-based page rank value, time-based author's rank value and the time-based document library rank value of all documents in described bunch;

Bunch trend generates step, and it is according to the time-based bunch of page rank value in the future of described time-based bunch of page rank value compute cluster; With

Bunch trend rank step, thus its time-based bunch of page rank value to described future sorts and obtains trend,

Wherein, described time-based bunch of page rank calculation procedure further comprises:

Page rank value calculation procedure, the time-based page rank value of document in its compute cluster;

Author's rank value calculation procedure, time-based author's rank value of document in its compute cluster;

Document library rank value calculation procedure, the time-based document library rank value of document in its compute cluster; With

The weighting step, its by calculate time-based page rank value from described page rank value calculation procedure, from time-based author's rank value of described author's rank value calculation procedure with from the weighted sum of the time-based document library rank value of described document library rank value calculation procedure, and in accumulation bunch, the weighted sum of all documents produces described time-based bunch of page rank value.

7. method as claimed in claim 6, wherein, by in all time T _iThe weighted sum of time-based page rank changing value of document calculate the document in time T _nTime-based page rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iThe weight of time-based page rank changing value of document be time T _iTo T _nThe aging function of mistiming, and in time T _iThe time-based page rank changing value of document be that all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, and described all point to the document of the document in time T _iWeight and the described search step of time-based page rank changing value in the document correlation that obtains be directly proportional.

8. method as claimed in claim 6, wherein, time-based author's rank value of described document is time-based author's rank value sum of all authors of this document, wherein, by in all time T _iAuthor's the weighted sum of time-based author's rank changing value calculate this author in time T _nTime-based author's rank value, i=1 wherein ..., n, T _i≤ T _n, and described in all time T _iAuthor's the weight of time-based author's rank changing value be time T _iTo T _nThe aging function of mistiming, and in time T _iTime-based author's rank value of author write for this author all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, described author write all point to the document of the document in time T _iWeight and the described search step of time-based page rank changing value in the document correlation that obtains be directly proportional.

9. method as claimed in claim 6, wherein, the time-based document library rank value of document that described document library rank value computing unit calculates is the time-based document library rank value of the document place document library, and document library is in time T _nThe time time-based document library rank value be in all time T _iThe time-based document library rank changing value sum of document library, i=1 wherein ..., n, T _i≤ T _n, and in time T _iThe time-based document library rank changing value of document library be that in the document storehouse, all point to the document of the document in time T _iThe weighted sum of time-based page rank changing value, in described document library, all point to the document of the document in time T _iWeight and the described search step of time-based page rank changing value in the document correlation that obtains be directly proportional.

Images