CN105787072A - Field knowledge extracting and pushing method oriented to progress - Google Patents

Abstract

A field knowledge extracting and pushing method oriented to a progress includes the steps of extracting knowledge from a knowledge resource, expressing the knowledge through a knowledge vector, extracting a progress vector from a progress task, extracting a user property vector from user description, matching and calculating the progress vector, the user property vector and the knowledge vector to obtain the relevancy of knowledge demand and knowledge, calculating a knowledge value through a user behavior variable, and correcting the knowledge relevancy calculation result through the knowledge value to obtain a knowledge sequencing result closely related to the progress task, thereby more accurately pushing the knowledge to the progress task. It is thought that as the essence of knowledge pushing, correct knowledge is transmitted to correct people at correct time in a correct form; the field knowledge extracting and pushing system and method oriented to the progress are high in work efficiency and high in accuracy.

Description

The domain knowledge of a kind of Process-Oriented extracts and method for pushing

Technical field

The present invention relates to a kind of computer literacy application system, especially for extracting the knowledge relevant to customer service activity, and push the knowledge method and system to related service activity according to actual needs intelligently.

Background technology

Under environment of knowledge-based economy, knowledge is just becoming the most important resource of more and more enterprise, its essence is and knowledge is considered as most important ERM, by the acquisition of knowledge and share, appropriate user is given by appropriate Knowledge delivery, thus being effectively improved the knowledge innovation of Company Knowledge utilization rate, promotion employee, enterprise is made to obtain competitive advantage.

Information management and integrated be improve Company Knowledge reuse level, it is prevented that Organizational Knowledge Loss Caused, it is achieved operation flow automatization, intelligentized effective means.Enterprise have accumulated substantial amounts of experience and knowledge in process of production, but these knowledge exist with the form of implicit knowledge in data base, knowledge base, paper document and human brain mostly, people obtain the mode of knowledge mainly through manual information retrieval, it is impossible to ensure the promptness of knowledge acquisition, accuracy and comprehensive.So research is by existing knowledge and the moving phase association of product design, carry out integrated with design process so that the participant of R & D design process obtains, in the appropriate time, the knowledge needed and seems necessary.

R & D design process is knowledge intensive active set, and the substantial amounts of rule knowledge of design activity application, Case knowledge, empirical data carry out decision-making and design on the one hand；Design activity is also the source that knowledge produces and obtains on the other hand.The target of Knowledge Aggregation is exactly be combined with design process by Design Knowledge Management, by the description to attributes such as design activity role, object, target, demands, mate with relevant knowledge, thus reducing manual knowledge search work, it is achieved rapidly and accurately design knowledge is pushed to designer.

Company Knowledge propelling movement has been studied and has been attempted by some scholar and tissue both at home and abroad at present.In based on flow process and event driven knowledge supplying system research, main book has: at document: Wang Shengfa, Gu Xinjian, the write paper such as Guo Jianfeng: the knowledge active push research of used for products design, it is loaded in " computer integrated manufacturing system ", 2007,13 (2): 234-239, for the shortcoming that knowledge actively can not be passed to suitable designer by existing system in due course, it is proposed that the management of a kind of knowledge based, with the product-design knowledge active push architecture of workflow logic nets and mode；But its rule involved by knowledge coupling system and object are relatively more, thus have impact on knowledge coupling and the efficiency pushed and accuracy；At document: Liu Xinyu is write paper: based on the theoretical frame of knowledge supplying system and the applied research of flow process.Shenyang: Northeastern University, 2005.To the knowledge supplying system based on flow process, mainly to it, theoretical, framework and application have carried out systematic study；To knowledge push technology, mainly describe the method for pushing towards the operation knowledge in post and the project knowledge towards employee；But still not accounting for the impact of ken, its method exists that retrieval data volume is big, push the problem that precision is not high；At document: Chen Hao, Liu Nian, Hu Yanjun writes paper: based on the knowledge Modeling research of workflow, it is loaded in " automated manufacturing ", 2005,27 (5): 8-12, it is proposed that based on knowledge Modeling and the system framework thereof of workflow, discuss the integrated technology supporting knowledge with knowledge, knowledge and people and knowledge and process, and knowledge is applied to the method in the operation flow of enterprise；But this article only considered the interaction of workflow and knowledge base, it does not have consider knowledge agent people impact wherein, and simply establish conceptual model and simply describe correlation technique.The common drawback of the studies above is that the knowledge coupling proposed exists the problems such as low, the poor accuracy of efficiency with method for pushing.

Summary of the invention

It is an object of the invention to: the domain knowledge of a kind of new Process-Oriented proposed for above-mentioned background technology Problems existing extracts and supplying system and method.It is considered herein that the essence that knowledge pushes is in the correct time, with correct form, correct knowledge is passed to correct people.Only knowledge, business events flow path and people's triplicity are got up, effective knowledge could be realized and push.

The technical scheme is that the domain knowledge of a kind of Process-Oriented extracts and supplying system, including such as lower module:

Line module, user is the target that knowledge pushes, and the knowledge that user's reference system pushes performs flow tasks, and carries out other information management activities；

Knowledge resource module, provides a memory space for Company Knowledge, and it is carried out Classification Management, and responding to knowledge pushes request, and the employee of auxiliary line module completes flow tasks；

Knowledge pushing module, is calculated reconstruction of layer by representation of knowledge abstraction, layer and knowledge and constitutes, be used for reducing knowledge matching range, and a matching symbol closes the domain knowledge of flow characteristic, and is pushed to user, improves matching efficiency and precision；

Process module, is the source of knowledge propelling movement, and flow process is produced by the real needs of enterprise, it is possible to be divided into several relatively independent sub-process task, and each sub-process can arrange the attribute of knowledge according to its relevant knowledge related to.

Other information management activities that described line module carries out include the knowledge pushed is evaluated and is fed back, issuing the implicit knowledge of oneself, customize knowledge interested, recommend knowledge etc. to other users.

Described knowledge resource module is a knowledge space of the knowledge composition by aspects such as Company Knowledge storehouse, enterprise staff implicit knowledge, customer knowledge, the open networks of enterprise external.

Described knowledge pushing module is calculated reconstruction of layer by representation of knowledge abstraction, layer and knowledge and constitutes.

Described representation of knowledge abstraction, layer includes Knowledge Extraction layer and representation of knowledge layer.

Described Knowledge Extraction layer refers to that employing text message extractive technique extracts required knowledge content from knowledge resources numerous in enterprise.

Described text message extractive technique refers to a certain class extracting regulation from one section of text or a few class customizing messages, and is carried out the process of structuring process；Its processing procedure includes, the rule set that first tectonic information extracts, and recycles these rules and extracts the information specified from text.

Described representation of knowledge layer is that the result extracted by text message carries out information norm with forms such as the tables of data of data base, and representation of knowledge layer is corresponding with knowledge vector.

Described knowledge calculates reconstruction of layer and refers on the basis of the representation of knowledge, according to the flow tasks demand to knowledge, by the matching operation to knowledge, obtains qualified knowledge, and obtained knowledge is pushed to the user performing flow tasks.

The matching operation of described knowledge comprises the steps:

First, correlation calculations, refer to according to operation flow the demand of knowledge (i.e. flow process vector), and the feature of knowledge self (i.e. knowledge vector) and the specialty of user, preference and experience (i.e. user property vector), the knowledge meeting knowledge requirement is carried out correlation calculations；

Second, knowledge score value calculates and refers to based on the score calculation to knowledge that user behavior vector produces；

3rd, knowledge sort refers on the basis of correlation calculations, knowledge score value result of calculation intervenes the correlation calculations sequence to knowledge, knowledge score after being corrected, according to score order from high to low, knowledge is ranked up, result is pushed to the user of a certain particular task, thus realizing the propelling movement with flow tasks with the knowledge being closely connected of the present invention.

The domain knowledge of a kind of Process-Oriented extracts and method for pushing, it is characterised in that comprise the steps:

First, from knowledge resource, extract knowledge and represent by knowledge vector, from flow tasks, extracting flow process vector, from user profile, extracting user property vector；

Second, by carrying out matching primitives between flow process vector, user property vector sum knowledge vector, draw the degree of association of knowledge requirement and knowledge；

3rd, calculated by user behavior variable and obtain knowledge score value, finally with knowledge score value, the result of knowledge relatedness computation is corrected, obtains knowledge sort result closely-related with flow tasks, it is achieved knowledge more accurately pushing to flow tasks.

The described knowledge that extracts from knowledge resource refers to that employing text message extractive technique extracts required knowledge content from knowledge resources numerous in enterprise；Described text message extractive technique refers to a certain class extracting regulation from one section of text or a few class customizing messages, and is carried out the process of structuring process.

Described knowledge vector is used to the vector space of Description of Knowledge.

The source of described knowledge vector (KNOWLEDGEVECTORS, WV) includes the theme of text, author, key word, summary and time；Described knowledge vector is a five-tuple:

KV=SF, AF, KF, ABF, TIF

Wherein:

SF is the theme the factor (TOPICFACTORS), SF={sf_i: I=1,2 ..., | SF | }, represent the affiliated theme of text, by the theme lexical representation of finite number, same section document can belong to one or more theme；

AF is author's factor (AUTHORFACTORS), AF={AFJ:J=1,2 ..., | AF | }, represent the author of knowledge；

KF is the key word factor (KEYWORDFACTORS), KF={KFK:K=1,2 ..., | KF | }, represent the key word of text；

ABF is the summary factor (ABSTRACTFACTORS), ABF={ABFM:M=1,2 ..., | ABF | }, the vocabulary from text snippet of finite number represent；

TIF is time factor (TIMEFACTORS), by six coded representations of YY-MM-DD, represents the year, month and day that document produces respectively.

Described flow process vector is from flow tasks, and flow process vector (WORKFLOWVECTORS, WV) is a tlv triple:

WV=FF, WTF, WPF

Wherein:

FF is the field factor (FIELDFACTORS), FF={FF_I: I=1,2 ..., | FF | }, represent the finite aggregate of business scope；WTF is the flow type factor (WORKFLOWTYPEFACTORS), WTF={WTF_J: J=1,2 ..., | WTF | }, represent in the orchestration instance operation phase, according to the flow type finite aggregate that concrete example tasks and situation are proposed；WPF is the saddlebag factor (WORKPACKAGEFACTORS), WPF={WPF_K: K=1,2 ..., | WPF | }, the finite aggregate of the characteristic item that the saddlebag at expression flow tasks place comprises；

Field factor FF refers in the Business Process Modeling stage, and based on tissue experience and best practices, the classification of the business scope specified by its composition activity, is the finite aggregate being made up of domain name；Flow type factor WTF refers to the affiliated type of flow process, such as program planning class flow process, calculates class flow process, test class flow process etc.；The source of saddlebag factor WPF can include the title of saddlebag, keyword, input, output, binding target and resource；

The field factor in flow process vector, the flow type factor, its weights can be specified by user in advance as required；

The weights of the characteristic item in saddlebag, calculate by the calculation of Features weight in knowledge vector.

Described user property vector refers to characterize the vector space of individual consumer's feature.

Described user property vector is a tlv triple:

UV=MF, UIF, EF

Wherein:

MF is the specialty factor (MAJORFACTORS), MF={MFI:I=1,2 ..., | MF | }, by the major name lexical representation of finite number；

UIF is the user interest factor, UIF={UIFI:I=1,2 ..., | UIF | }, by several feature lexical representations；

EF is experience factor (EXPERIENCEFACTORS), EF={EFI:I=1,2 ..., | EF | }, extract from the passing project experiences of user, the professional technicality of finite number represent；Characteristic item in user property vector, can be specified its weights by user previously according to needs.

Described knowledge requirement (is represented by KR) that the vector set being made up of flow process vector WV, user property vector UV is characterized, it may be assumed that KR=WV ∪ UV.

The relatedness computation method of described knowledge requirement and knowledge is:

$S_{IM}(KR,KV)=COS\mathrm{\θ}=\frac{\underset{k=1}{\overset{n}{\Σ}}{W}_{1k}\×W_{2k}}{\sqrt{\left(\underset{k=1}{\overset{n}{\Σ}}{W_{1k}}^2\right)\×\left(\underset{k=1}{\overset{n}{\Σ}}{W_{2k}}^2\right)}}$

Wherein, W1K, W2K represent knowledge requirement KR and the weights of knowledge vector KV k-th characteristic item, 1≤K≤N respectively.

Described calculating by user behavior variable is obtained knowledge score value method and is:

Based on the score calculation to knowledge that user behavior vector produces.Correlativity calculation result is had corrective action by the result that knowledge score value calculates, and it acts in knowledge sort and embodies.

Described user behavior vector is browsed by user based on specific knowledge, downloads, collects, recommending data calculates, and represents with FL, FD, FF, FR respectively.

The formula that knowledge score value (KNOWLEDGESCORE, KS) calculates is:

$KS=\mathrm{\α}\×\frac{\frac{{\mathrm{FD}}_i}{FD}+\frac{{\mathrm{FL}}_i}{FL}}{2}+(1-\mathrm{\α})\×\frac{\frac{{\mathrm{FF}}_i}{FF}+\frac{{\mathrm{FR}}_i}{FR}}{2}$

Wherein, FD_IThe download time being downloaded maximum knowledge in the number of times and the A to Z of that this knowledge is downloaded, FL is represented respectively with FD_IRepresent respectively in the number of times and the A to Z of that this knowledge browsed with FL and browsed the number of visits of maximum knowledge, FF_IWith the collection number of times that FF represents knowledge maximum by collection in the number of times and the A to Z of that this knowledge collected respectively, FR_IThe recommendation number of times of recommended maximum knowledge in the recommended number of times of this knowledge and the A to Z of is represented respectively with FR.By above-mentioned definition it can be seen thatIn like manner, $0\≤\frac{{\mathrm{FL}}_i}{FL}\≤1,0\≤\frac{{\mathrm{FF}}_i}{FF}\≤1,0\≤\frac{{\mathrm{FR}}_i}{FR}\≤1.$

The computational methods of described knowledge sort result are:

On the basis of correlation calculations, knowledge score value result of calculation intervenes the correlation calculations sequence to knowledge, knowledge score after being corrected, according to score order from high to low, knowledge is ranked up, result is pushed to the employee of a certain particular task, thus realizing the propelling movement with flow tasks with the knowledge being closely connected of the present invention；Its computing formula is:

KCS=β × Sim (KR, KV)+(1-β) × KS

Wherein, KCS (KNOWLEDGECORRECTIONSCORE, KCS) represents the knowledge score after correcting, SIM (KR, KV) is correlativity calculation result, and KS is knowledge score value result of calculation, by suing for peace to after the two ranking operation, the knowledge score KCS after being corrected；β is used to regulate the operator of correlativity calculation result and knowledge score value result of calculation.

Advantages of the present invention: it is considered herein that the essence that knowledge pushes is in the correct time, with correct form, correct knowledge is passed to correct people.Only knowledge, business events flow path and people's triplicity are got up, effective knowledge could be realized and push.The invention described above provides domain knowledge extraction and the supplying system of Process-Oriented and method work efficiency is high, accuracy is good.

Accompanying drawing explanation

Fig. 1 represents the extraction of Process-Oriented domain knowledge and the framework of supplying system

Fig. 2 represents that knowledge pushes the implementation of layer

Fig. 3 represents knowledge computing and the sequencer procedure of Process-Oriented

Detailed description of the invention

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described:

The present invention proposes the domain knowledge supplying system of a kind of Flow driving, including such as lower module:

Line module, user is the target that knowledge pushes, and the knowledge that user's reference system pushes performs flow tasks, and carries out other information management activities；

Knowledge resource module, provides a memory space for Company Knowledge, and it is carried out Classification Management, and responding to knowledge pushes request, and the employee of auxiliary line module completes flow tasks；

Knowledge pushing module, is calculated reconstruction of layer by representation of knowledge abstraction, layer and knowledge and constitutes, be used for reducing knowledge matching range, and a matching symbol closes the domain knowledge of flow characteristic, and is pushed to user, improves matching efficiency and precision；

Process module, is the source of knowledge propelling movement, and flow process is produced by the real needs of enterprise, it is possible to be divided into multiple relatively independent sub-process task, and each sub-process can arrange the attribute of knowledge according to its relevant knowledge related to.

Meanwhile, the domain knowledge of its Process-Oriented extracts and method for pushing, comprises the steps:

First, from knowledge resource, extract knowledge and represent by knowledge vector, from flow tasks, extracting flow process vector, from user profile, extracting user property vector；

Second, by carrying out matching primitives between flow process vector, user property vector sum knowledge vector, draw the degree of association of knowledge requirement and knowledge；

3rd, calculated by user behavior variable and obtain knowledge score value, finally with knowledge score value, the result of knowledge relatedness computation is corrected, obtains knowledge sort result closely-related with flow tasks, it is achieved knowledge more accurately pushing to flow tasks.

Organization Chart in Fig. 1 describe at all levels between logical relation.Major function at all levels describes as follows:

(1) client layer

User mostlys come from the employee of enterprises, is the main body of knowledge creation and application, is also the target of knowledge propelling movement.User is except the knowledge execution flow tasks that reference system pushes, other information management activity can also be carried out, as the knowledge pushed being evaluated and fed back, issuing the implicit knowledge of oneself, customizing knowledge interested, recommend the activities such as knowledge to other users.

(2) knowledge resource layer

Knowledge resource layer be one by a knowledge space of the knowledge composition of the aspects such as Company Knowledge storehouse, enterprise staff implicit knowledge, customer knowledge, the open network of enterprise external.The main task of knowledge resource layer is to provide a memory space for Company Knowledge, and it is carried out Classification Management, and responding to knowledge pushes request, and the employee of auxiliary client layer completes flow tasks.

(3) knowledge pushes layer

It is realize being mutually related between client layer, knowledge resource layer and flow process layer an intermediate level that knowledge pushes layer.Knowledge pushes layer can reduce knowledge matching range, and a matching symbol closes the domain knowledge of flow characteristic, and is pushed to user, improves matching efficiency and precision.Knowledge pushes the emphasis place that layer is the present invention.

(4) flow process layer

Flow process layer is the starting point that knowledge pushes, and is the source of knowledge propelling movement.Flow process is produced by the real needs of enterprise, it is possible to is divided into several relatively independent sub-process task, distributes to each employee and undertake and complete task.Each sub-process can arrange the attribute of knowledge according to its relevant knowledge related to.

2, knowledge pushes layer

It is made up of as in figure 2 it is shown, described knowledge pushes layer representation of knowledge abstraction, layer and knowledge calculating reconstruction of layer.

Described representation of knowledge abstraction, layer includes Knowledge Extraction layer, representation of knowledge layer.

Knowledge Extraction layer refers to that employing text message extractive technique extracts required knowledge content from knowledge resources numerous in enterprise.

Described text message extractive technique refers to a certain class extracting regulation from one section of text or a few class customizing messages, and is carried out the process of structuring process.The present invention adopts rule-based text message extraction model to realize Knowledge Extraction.Its processing procedure includes, the rule set that first tectonic information extracts, and recycles these rules and extracts the information specified from text.

Representation of knowledge layer is that the result extracted by text message carries out information norm with forms such as the tables of data of data base.Representation of knowledge layer is corresponding with knowledge vector.About knowledge vector, it is discussed further below.

Knowledge calculates reconstruction of layer and refers on the basis of the representation of knowledge, according to the flow tasks demand to knowledge, by the matching operation to knowledge, obtains qualified knowledge, and obtained knowledge is pushed to the employee performing flow tasks.

3, knowledge calculates process

Knowledge calculates process and is made up of steps such as correlation calculations, the calculating of knowledge score value, knowledge sorts.

1) described correlation calculations, refer to according to operation flow the demand of knowledge (i.e. flow process vector), and the feature of knowledge self (i.e. knowledge vector) and the specialty of user, preference and experience (i.e. user property vector), the knowledge meeting knowledge requirement is carried out correlation calculations；

In the present invention, flow process vector, from flow tasks, is the main source of knowledge requirement.Flow process vector (WORKFLOWVECTORS, WV) is a tlv triple:

WV=FF, WTF, WPF

Wherein:

FF is the field factor (FIELDFACTORS), FF={FF_I: I=1,2 ..., | FF | }, represent the finite aggregate of business scope；WTF is the flow type factor (WORKFLOWTYPEFACTORS), WTF={WTF_J: J=1,2 ..., | WTF | }, represent in the orchestration instance operation phase, according to the flow type finite aggregate that concrete example tasks and situation are proposed；WPF is the saddlebag factor (WORKPACKAGEFACTORS), WPF={WPF_K: K=1,2 ..., | WPF | }, the finite aggregate of the characteristic item that the saddlebag at expression flow tasks place comprises.

Field factor FF refers in the Business Process Modeling stage, and based on tissue experience and best practices, the classification of the business scope specified by its composition activity, is the finite aggregate being made up of domain name；Flow type factor WTF refers to the affiliated type of flow process, such as program planning class flow process, calculates class flow process, test class flow process, etc.；The source of saddlebag factor WPF can include the title of saddlebag, keyword, input, output, binding target and resource.(illustrate: described saddlebag refers to the project deliverable achievement of the lowest level of work breakdown structure (WBS) (WORKBREAKDOWNSTRUCTURE, WBS), has its structurized representation.) such as, certain flow process vector WV, its field factor FF={ general field, pneumatic field, construction applications }, flow type factor WTF={ program planning class flow process, calculate class flow process }, saddlebag factor WPF={ title, input, output, resource, binding target }.The characteristic item general field of designated field factor FF, pneumatic field, construction applications, its weights respectively 0.3,0.3,0.2.Specify the characteristic item program planning class flow process of flow type factor WTF, calculate class flow process, its weights respectively 0.25,0.15.Characteristic item title in saddlebag factor WPF, input, output, resource, binding target weights, calculate by the calculation of TF-IDF Features weight in knowledge vector, its weights respectively 0.3,0.2,0.2,0.15,0.1.

Then the vector representation of flow process vector WV is WV={FF (0.3,0.3,0.2), WTF (0.25,0.15), WPF (0.3,0.2,0.2,0.15,0.1) }.

Described knowledge vector (KNOWLEDGEVECTORS, KV) is used to the vector space of Description of Knowledge, and its source includes the theme of text, author, key word, summary and time.The acquisition of knowledge vector is completed by aforesaid Knowledge Extraction process, and namely rule-based text message extracts.As it has been described above, knowledge vector is a five-tuple:

KV=SF, AF, KF, ABF, TIF

Wherein:

SF is the theme the factor (TOPICFACTORS), SF={sf_I: I=1,2 ..., | SF | }, represent the affiliated theme of text, by the theme lexical representation of finite number, same section document can belong to one or more theme；AF is author's factor (AUTHORFACTORS), AF={AF_J: J=1,2 ..., | AF | }, represent the author of knowledge；KF is the key word factor (KEYWORDFACTORS), KF={KF_K: K=1,2 ..., | KF | }, represent the key word of text；ABF is the summary factor (ABSTRACTFACTORS), ABF={ABF_M: M=1,2 ..., | ABF | }, the vocabulary from text snippet of finite number represent；TIF is time factor (TIMEFACTORS), by six coded representations of YY-MM-DD, represents the year, month and day that document produces respectively, such as 13-06-01, represents on June 1st, 2013.

Such as, certain knowledge vector KV, its subject gene SF={ aerofoil designs }, author factors A F={ Wang Li }, key word factor K F={ aerofoil profile }, the method for a summary factors A BF={ aerofoil configuration design } and, time factor TIF={13-06-01}.The weights of the characteristic item aerofoil design in subject gene SF, calculate by the calculation of TF-IDF Features weight, and its weights are 0.3.The weights of the characteristic item Wang Li in author factors A F, calculate by the calculation of TF-IDF Features weight, and its weights are 0.2.The weights of the characteristic item aerofoil profile in key word factor K F, calculate by the calculation of TF-IDF Features weight, and its weights are 0.2.The weights of the method for a kind of aerofoil configuration design of characteristic item in summary factors A BF, calculate by the calculation of TF-IDF Features weight, and its weights are 0.1.The weights of the characteristic item 13-06-01 in time factor TIF, calculate by the calculation of TF-IDF Features weight, and its weights are 0.15.

Then the vector representation of knowledge vector KV is KV={SF (0.3), AF (0.2), KF (0.2), ABF (0.1), TIF (0.15) }.

Described user property vector refers to characterize the vector space of individual consumer's feature.User property vector is another source of knowledge requirement.User property vector is a tlv triple:

UV=MF, UIF, EF

Wherein:

MF is the specialty factor (MAJORFACTORS), MF={MF_I: I=1,2 ..., | MF | }, by the major name lexical representation of finite number；UIF is the user interest factor, UIF={UIF_I: I=1,2 ..., | UIF | }, by several feature lexical representations；EF is experience factor (EXPERIENCEFACTORS), EF={EF_I: I=1,2 ..., | EF | }, extract from the passing project experiences of user, the professional technicality of finite number represent.

Such as, certain user property vector UV, its specialty factor M F={ master-plan }, user interest factor UIF={ flight dynamics, flight flutter test }, experience factor EF={ aerofoil designs, and improves remodeling }.Specifying the characteristic item master-plan of specialty factor M F, its weights are 0.2.Specify the characteristic item flight dynamics of user interest factor UIF, flight flutter test, its weights respectively 0.1,0.15.Specify the characteristic item aerofoil design of experience factor EF, improve remodeling, its weights respectively 0.15,0.15.

Then the vector representation of user property vector UV is UV={MF (0.2), UIF (0.1,0.15), EF (0.15,0.15) }.

The vector set that knowledge requirement (KNOWLEDGEREQUIREMENT, KR) is made up of flow process vector WV, user property vector UV is characterized, it may be assumed that

KR=WV ∪ UV

Connecting example, the vector representation of flow process vector WV is WV={FF (0.3,0.3,0.2), WTF (0.25,0.15), WPF (0.3,0.2,0.2,0.15,0.1) }, the vector representation of user property vector UV is UV={MF (0.2), UIF (0.1,0.15), EF (0.15,0.15) }.

Then the vector representation of knowledge requirement KR is KR={FF (0.3,0.3,0.2), WTF (0.25,0.15), WPF (0.3,0.2,0.2,0.15,0.1), MF (0.2), UIF (0.1,0.15), EF (0.15,0.15) }.

The weight computing of characteristic item:

A. the field factor in flow process vector, the flow type factor, its weights can be specified by user in advance as required；The weights of the characteristic item in saddlebag, calculate by the calculation of Features weight in knowledge vector；

B. the characteristic item in user property vector, as specialty (major name), interest (technical term), experience (technical term), its weights can be specified by user previously according to needs；

C., in knowledge vector, Features weight calculates, and quotes industry formula, and the present invention refer to TF-IDF function:

$\mathrm{\ψ}={\mathrm{TF}}_{t,d}\×\log \left(\frac{N}{D_t}\right)$

Wherein, TF_{T, D}For the characteristic item T frequency occurred in document D, N is the number of all documents, D_TFor the number of documents containing T.The proposition of this function is based on such a hypothesis: to the difference significant word of document should be those the frequency of occurrences is sufficiently high in a document, but the word that the frequency of occurrences is enough few in other documents of whole collection of document.Such as " undercarriage ", occurring in that 10 times in document D, in document D, total frequency of word is 200, then TF_{T, D}=10/200, namely 0.05；N is 100000, and namely number of files is 100000, and the document containing " undercarriage " is 10, then Then feature weight=the 0.05*4=0.2 of " undercarriage ".

Correlation calculations, is the process trying to achieve dependency between knowledge requirement (KR) and knowledge vector (KV).Particularly as follows:

$S_{IM}(KR,KV)=COS\mathrm{\θ}=\frac{\underset{k=1}{\overset{n}{\Σ}}{W}_{1k}\×W_{2k}}{\sqrt{\left(\underset{k=1}{\overset{n}{\Σ}}{W_{1k}}^2\right)\×\left(\underset{k=1}{\overset{n}{\Σ}}{W_{2k}}^2\right)}}$

Wherein, W_1K、W_2KRepresent the weights of knowledge requirement KR and knowledge vector KV k-th characteristic item, 1≤K≤N respectively.The such as characteristic item of knowledge requirement KR is A, B, C, D, weights respectively 30,20,20,10, the characteristic item of knowledge vector KV is A, C, D, E, weights respectively 40,30,20,10, then the vector representation of KR is KR (30,20,20,10,0), the vector representation of KV is KV (40,0,30,20,10) degree of association of knowledge requirement KR and the knowledge vector KV, then calculated according to above formula is 0.86.

2) described knowledge score value calculates, and refers to based on the score calculation to knowledge that user behavior vector produces.Correlativity calculation result is had corrective action by the result that knowledge score value calculates, and it acts in knowledge sort and embodies.

Described user behavior vector is browsed by user based on specific knowledge, downloads, collects, recommending data calculates, and represents with FL, FD, FF, FR respectively.

The formula that knowledge score value (KNOWLEDGESCORE, KS) calculates is:

$KS=\mathrm{\α}\×\frac{\frac{{\mathrm{FD}}_i}{FD}+\frac{{\mathrm{FL}}_i}{FL}}{2}+(1-\mathrm{\α})\×\frac{\frac{{\mathrm{FF}}_i}{FF}+\frac{{\mathrm{FR}}_i}{FR}}{2}$

Wherein, FD_IThe download time being downloaded maximum knowledge in the number of times and the A to Z of that this knowledge is downloaded, FL is represented respectively with FD_IRepresent respectively in the number of times and the A to Z of that this knowledge browsed with FL and browsed the number of visits of maximum knowledge, FF_IWith the collection number of times that FF represents knowledge maximum by collection in the number of times and the A to Z of that this knowledge collected respectively, FR_IThe recommendation number of times of recommended maximum knowledge in the recommended number of times of this knowledge and the A to Z of is represented respectively with FR.By above-mentioned definition it can be seen thatIn like manner, $0\≤\frac{{\mathrm{FL}}_i}{FL}\≤1,0\≤\frac{{\mathrm{FF}}_i}{FF}\≤1,0\≤\frac{{\mathrm{FR}}_i}{FR}\≤1.$ Download, navigation patterns are merged process by the present invention, and the arithmetical average of the two is assigned to a weights α, collection, recommendation behavior are merged process, the arithmetical average of the two is assigned to weights 1-α.On ordinary meaning, 0.1 α 0.4.That is, the download of user and navigation patterns will lower than collections with recommend the behavior impact on knowledge score value for the impact of knowledge score value.

Such as, FD_I、FD、FL_I、FL、FF_I、FF、FR_I, FR respectively 40,50,80,100,30,40,15,20, if α=0.3, then KS=0.765 in this example.

3) described knowledge sort, refer on the basis of correlation calculations, knowledge score value result of calculation intervenes the correlation calculations sequence to knowledge, knowledge score after being corrected, according to score order from high to low, knowledge is ranked up, result is pushed to the employee of a certain particular task, thus realizing the propelling movement with flow tasks with the knowledge being closely connected of the present invention.Its computing formula is:

KCS=β × Sim (KR, KV)+(1-β) × KS

Wherein, KCS (KNOWLEDGECORRECTIONSCORE, KCS) represents the knowledge score after correcting, SIM (KR, KV) is correlativity calculation result, and KS is knowledge score value result of calculation, by suing for peace to after the two ranking operation, the knowledge score KCS after being corrected.β is used to regulate the operator of correlativity calculation result and knowledge score value result of calculation.Knowledge sort is played a major role by correlation calculations, therefore, on ordinary meaning, and 0.7 β 0.9.Such as, assume the correlativity calculation result of two knowledge respectively 0.86 and 0.80, corresponding respective knowledge score value KS is 0.72 and 0.9, if β is 0.7, the then KCS of the two respectively 0.818 and 0.83, therefore, in final knowledge sort, the sorting position of the latter prior to the former, will give the user performing flow tasks as the knowledge recommendation higher with knowledge requirement matching degree.

Claims (10)

1. the domain knowledge of a Process-Oriented extracts and method for pushing, it is characterised in that comprise the steps:

First, from knowledge resource, extract knowledge and represent by knowledge vector, from flow tasks, extracting flow process vector, from user profile, extracting user property vector；

Second, by carrying out matching primitives between flow process vector, user property vector sum knowledge vector, draw the degree of association of knowledge requirement and knowledge；

3rd, calculated by user behavior variable and obtain knowledge score value, finally with knowledge score value, the result of knowledge relatedness computation is corrected, obtains knowledge sort result closely-related with flow tasks, it is achieved knowledge more accurately pushing to flow tasks.

2. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that the described knowledge that extracts from knowledge resource refers to that employing text message extractive technique extracts required knowledge content from knowledge resources numerous in enterprise；Described text message extractive technique refers to a certain class extracting regulation from one section of text or a few class customizing messages, and is carried out the process of structuring process.

3. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that described knowledge vector is used to the vector space of Description of Knowledge.

4. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that the source of described knowledge vector (KnowledgeVectors, WV) includes the theme of text, author, key word, summary and time；Described knowledge vector is a five-tuple:

KV=SF, AF, KF, ABF, TIF

Wherein:

SF is the theme the factor (TopicFactors), TF={sf_i: i=1,2 ..., | SF | }, represent the affiliated theme of text, by the theme lexical representation of finite number, same section document can belong to one or more theme；

AF is author's factor (AuthorFactors), AF={afj:j=1,2 ..., | AF | }, represent the author of knowledge；

KF is the key word factor (KeywordFactors), KF={kfk:k=1,2 ..., | KF | }, represent the key word of text；

ABF is the summary factor (AbstractFactors), ABF={abfm:m=1,2 ..., | ABF | }, the vocabulary from text snippet of finite number represent；

TIF is time factor (TimeFactors), by six coded representations of YY-MM-DD, represents the year, month and day that document produces respectively.

5. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that described flow process vector is from flow tasks, and flow process vector (WorkflowVectors, WV) is a tlv triple:

WV=FF, WTF, WPF

Wherein:

FF is the field factor (FieldFactors), FF={ff_i: i=1,2 ..., | FF | }, represent the finite aggregate of business scope；WTF is the flow type factor (WorkflowTypeFactors), WTF={wtf_j: j=1,2 ..., | WTF | }, represent in the orchestration instance operation phase, according to the flow type finite aggregate that concrete example tasks and situation are proposed；WPF is the saddlebag factor (WorkpackageFactors), WPF={wpf_k: k=1,2 ..., | WPF | }, the finite aggregate of the characteristic item that the saddlebag at expression flow tasks place comprises；

Field factor FF refers in the Business Process Modeling stage, and based on tissue experience and best practices, the classification of the business scope specified by its composition activity, is the finite aggregate being made up of domain name；Flow type factor WTF refers to the affiliated type of flow process, such as program planning class flow process, calculates class flow process, test class flow process etc.；The source of saddlebag factor WPF can include the title of saddlebag, keyword, input, output, binding target and resource；

The field factor in flow process vector, the flow type factor, its weights can be specified by user in advance as required；

The weights of the characteristic item in saddlebag, calculate by the calculation of Features weight in knowledge vector.

6. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that described user property vector refers to characterize the vector space of individual consumer's feature.

7. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that described user property vector is a tlv triple:

UV=MF, UIF, EF

Wherein:

MF is the specialty factor (MajorFactors), MF={mfi:i=1,2 ..., | MF | }, by the major name lexical representation of finite number；

UIF is the user interest factor, UIF={uifi:i=1,2 ..., | UIF | }, by several feature lexical representations；

EF is experience factor (ExperienceFactors), EF={efi:i=1,2 ..., | EF | }, extract from the passing project experiences of user, the professional technicality of finite number represent；Characteristic item in user property vector, can be specified its weights by user previously according to needs.

8. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that described knowledge requirement (is represented by KR) that the vector set being made up of flow process vector WV, user property vector UV is characterized, it may be assumed that KR=WV ∪ UV.

9. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that the relatedness computation method of described knowledge requirement and knowledge is:

$Sim(KR,KV)=cos\mathrm{\θ}=\frac{\underset{k=1}{\overset{n}{\Σ}}{W}_{1k}\×W_{2k}}{\sqrt{\left(\underset{k=1}{\overset{n}{\Σ}}{W_{1k}}^2\right)\×\left(\underset{k=1}{\overset{n}{\Σ}}{W_{2k}}^2\right)}}$

Wherein, W1k, W2k represent knowledge requirement KR and the weights of knowledge vector KV k-th characteristic item, 1≤k≤N respectively.

10. the domain knowledge of Process-Oriented according to claim 1 extracts and method for pushing, it is characterised in that described calculating by user behavior variable is obtained knowledge score value method and be:

Based on the score calculation to knowledge that user behavior vector produces, correlativity calculation result is had corrective action by the result that knowledge score value calculates, and it acts in knowledge sort and embodies,

Described user behavior vector is browsed by user based on specific knowledge, downloads, collects, recommending data calculates, and represents with FL, FD, FF, FR respectively,

The formula that knowledge score value (KnowledgeScore, KS) calculates is:

$KS=\mathrm{\α}\×\frac{\frac{{\mathrm{FD}}_i}{FD}+\frac{{\mathrm{FL}}_i}{FL}}{2}+(1-\mathrm{\α})\×\frac{\frac{{\mathrm{FF}}_i}{FF}+\frac{{\mathrm{FR}}_i}{FR}}{2}$

Wherein, FD_iThe download time being downloaded maximum knowledge in the number of times and the A to Z of that this knowledge is downloaded, FL is represented respectively with FD_iRepresent respectively in the number of times and the A to Z of that this knowledge browsed with FL and browsed the number of visits of maximum knowledge, FF_iWith the collection number of times that FF represents knowledge maximum by collection in the number of times and the A to Z of that this knowledge collected respectively, FR_iRepresent the recommendation number of times of recommended maximum knowledge in the recommended number of times of this knowledge and the A to Z of respectively with FR, by above-mentioned definition it can be seen that $0\≤$ $\frac{{\mathrm{FD}}_i}{FD}\≤1,$ In like manner, $0\≤\frac{{\mathrm{FL}}_i}{FL}\≤1,0\≤\frac{{\mathrm{FF}}_i}{FF}\≤1,0\≤\frac{{\mathrm{FR}}_i}{FR}\≤1.$

The computational methods of described knowledge sort result are:

On the basis of correlation calculations, knowledge score value result of calculation intervenes the correlation calculations sequence to knowledge, knowledge score after being corrected, according to score order from high to low, knowledge is ranked up, result is pushed to the employee of a certain particular task, thus realizing the propelling movement with flow tasks with the knowledge being closely connected of the present invention；Its computing formula is:

KCS=β × Sim (KR, KV)+(1-β) × KS

Wherein, KCS (KnowledgeCorrectionScore, KCS) represents the knowledge score after correcting, Sim (KR, KV) is correlativity calculation result, and KS is knowledge score value result of calculation, by suing for peace to after the two ranking operation, the knowledge score KCS after being corrected；β is used to regulate the operator of correlativity calculation result and knowledge score value result of calculation.