CN103400227B - Excavate based on figure and the flow process of map distance recommends method - Google Patents

Abstract

The present invention relates to process automation field, disclose a kind of excavation based on figure and the flow process recommendation method of map distance, specifically comprise the following steps that pre-treatment step: by abstract for the flow process collection of the input form being numbered directed graph, obtain flow process subgraph；Mode discovery step: subgraph digs the intersection that described pre-treatment step exports by evidence and decomposing module and decomposes, obtain upstream subgraph, both candidate nodes collection and confidence level, gained upstream subgraph, both candidate nodes collection and confidence level are registered as the Data Entry in pattern table；Flow process recommendation step: recommending module obtains reference flowchart, reference flowchart is compared with the upstream subgraph in pattern table, selects matched data entry, the both candidate nodes collection corresponding to matched data entry is output as recommended flowsheet.It is an advantage of the current invention that recommendation efficiency is high, the computation complexity of algorithm is less, it is recommended that precision is high, supports the process of labyrinth flow process, has higher using value.

Description

Excavate based on figure and the flow process of map distance recommends method

Technical field

The present invention relates to process automation field, excavate and the flow process recommendation side of map distance based on figure particularly to a kind of Method.

Background technology

Can Business Process Modeling rapidly and efficiently be to weigh modern enterprise tackle the major criterion of changeable corporate environment.So And, Business Process Modeling is an extremely complex and time-consuming job, and it requires that the field that modeling personnel not only possess specialty is known Know, in addition it is also necessary to be familiar with the execution process of each operational action, execution sequence and abnormality processing situation.At present, based on business intelligence The technology of (business intelligence, BI), as digging flow and flow process are retrieved, is used to assist process modeling.Flow process By data mining technology, digging technology finds that from flow process storehouse or event log flow process is as modeling reference；Flow process retrieval skill Art is then to retrieve similar flow process fragment from workflow warehouse and assist modeling with this.But these technology automaticities are low, Remaining a need for more artificial participation, modeling efficiency fall is the highest, and modeling accuracy can not meet requirement.Flow process recommended technology based on Existing process mode and modeling fragment, can recommend to model the follow-up possible flowage structure of fragment for modeling personnel, therefore automatically Receive much concern as the important supplementary means of process modeling process.

Current most of operation flow is all to be modeled with the form of graph structure, such as Petri network, event driven process Chain (Event-driven Process Chains, EPC), Business Process Modeling and mark (Business Process Model And Notation, BPMN) etc., therefore, existing procedure recommends to develop on the algorithm that method is all excavated at figure, figure mates. Flow process based on figure recommends method to be carried out in three steps: first process flow process storehouse with figure mining algorithm gSpan, obtains frequency Numerous subgraph；Then decompose subgraph and obtain pattern table (including ion path and corresponding active node), and be stored in data In storehouse；Finally by the upstream subgraph compared in current process fragment and pattern table, it is recommended that go out most suitable active node.So After, existing flow process recommends method to fail to support the flow process of loop structure, and its practicality is extremely restricted.

In order to support to comprise the Complicated Flow of loop structure, improve high efficiency and the effectiveness of commending system simultaneously, We construct a commending system using the matching algorithm improved, and can effectively support to include the stream of the labyrinths such as circulation Journey also has preferably performance in efficiency and recommendation accuracy.

Summary of the invention

The present invention is directed to existing flow process recommends the practicality of method to limit bigger, it is impossible to support to comprise loop structure The shortcoming such as Complicated Flow, it is provided that a kind of novel excavate based on figure and the flow process of map distance recommends method.

For achieving the above object, the present invention can take following technical proposals:

Excavate based on figure and the flow process of map distance recommends method, specifically comprise the following steps that

Pre-treatment step: by abstract for the flow process collection of the input form being numbered directed graph, obtain flow process subgraph, described flow process Collection includes technological process, operation flow and flow of transactions, uses Frequent Subgraph Mining that flow process subgraph is dug evidence, defeated Go out to comprise all intersections including flow process subgraph and the subgraph frequency of occurrences thereof；

Mode discovery step: the intersection that subgraph digs evidence and described pre-treatment step is exported by decomposing module (21) is carried out point Solve, obtain having an impact upstream subgraph, both candidate nodes collection and confidence level, gained is had an impact upstream subgraph, both candidate nodes collection with And confidence level is registered as the Data Entry in pattern table (3), the end node of described flow process subgraph is as both candidate nodes, remainder It is allocated as upstream subgraph, selects both candidate nodes to concentrate the confidence level upstream subgraph more than threshold value, be and have an impact upstream subgraph；

Flow process recommendation step: recommending module (4) obtains reference flowchart, and described reference flowchart is defeated by subscriber interface module (1) Enter, reference flowchart is compared with the upstream subgraph that has an impact in pattern table (3), select matched data entry, will Join the both candidate nodes collection corresponding to Data Entry and be output as recommended flowsheet.

As preferably, in flow process recommendation step, the step of described comparison specifically includes:

1) each in pattern table (3) is had an impact upstream subgraph p and the process of reference flowchart R, obtain described p and R Minimum public hypergraph MCSub and maximum public subgraph MCSup, be calculated MM distance, described MM distance be minimum public surpasses The difference of the size of figure MCSub and maximum public subgraph MCSup, i.e. MMDist=| MCSup |-| MCSub |；

2) according to the node rearward position of reference flowchart R, the positional distance Lo of described p and R, described node position backward are obtained It is set to: make R=(N, E, L, n, α) represent reference flowchart,Represent R set of node to be recommended, In (x) and Out (x) represents input node collection and the output node collection of node x respectively, and Num (N) represents the number of set N interior joint, x, y, z ∈(NNRR)；The node rearward position of node x is:

$\mathrm{Lop}\left(x\right)=\left\{\begin{array}{c}0,x\∈\mathrm{RR}\\ \mathrm{Lop}\left(y\right)+1,x\∈i\left(y\right),\mathrm{mum}\left(o\left(i\left(y\right)\right)\right)=1\\ \max (\mathrm{Lop}\left(z\right)+1),z\∈o\left(i\left(x\right)\right),\mathrm{num}\left(o\left(i\left(x\right)\right)\right)>1\end{array}\right.;$

The calculation procedure of described positional distance Lo includes: 1 ') find out the loop structure in described reference flowchart R, respectively by difference Replace the most independent loop structure without articulare, described is the node independent of active node without articulare, if Common node is there is, then by said two or plural loop structure between two or more loop structure Replace without articulare with same respectively, obtain the figure without circulation of reference flowchart R；2 ') obtain in figure without circulation figure according to gained The node rearward position of node；3 ') by without circulation figure reverts to step 1 without articulare ') replace it before loop structure, And the positional distance without articulare is assigned to the node of loop structure, obtain in reference flowchart R the node of all nodes backward Position；4 ') according to step 3 ') obtained by node rearward position, obtain described positional distance Lo；

3) obtain total distance of described p and R according to described MM distance and positional distance Lo, and by both candidate nodes collection, always away from From and confidence level add in CNS；

4) entry in CNS is always ranked up apart from ascending according to described, if apart from identical, then according to confidence Degree is ranked up from high to low, then elects multiple results forward for position as matched data entry.

As preferably, also include concrete steps: described flow process recommendation step also includes obtained recommended flowsheet Adding flow process storehouse, described flow process storehouse is used for the Data Source of the flow process collection in pre-treatment step as input.

As preferably, the flow process collection in described pre-treatment step also includes the flow process inputted by subscriber interface module (1) Collection.

As preferably, circulation performs described pre-treatment step, mode discovery step and flow process recommendation step.

As preferably, described Frequent Subgraph Mining includes step in detail below: Land use models increases strategy, uses deep The convenient pattern search space of degree mode of priority, on the basis of known Frequent tree mining pp, extension produces the son frequently son of described pp Figure, described sub-Frequent tree mining is the child node of Frequent tree mining, and calculates the support of described sub-Frequent tree mining；To described in each The sub-Frequent tree mining of pp, continues extension, in the way of depth-first till finding all of Frequent tree mining.

As preferably, described mode discovery step also includes pattern table construction step, specifically comprises the following steps that decomposition process Figure, obtains upstream subgraph and candidate's joint；Calculate the confidence level of each upstream subgraph and both candidate nodes, according to upstream subgraph, Both candidate nodes and confidence level forming types table, the Data Entry of described pattern table be tlv triple T=(I, C, f), wherein, I For having an impact the finite aggregate of upstream subgraph, C is the finite aggregate of both candidate nodes, and f:I → C is surjective function.

Related notion and definition:

Distance based on maximum public subgraph and the public hypergraph of minimum (MM distance): two business process map P1 and P2 are Mini-bus hypergraph and the public subgraph of maximum are respectively MCSub and MCSup, and the size of figure P is limit and nodes sum, i.e. and | P |=| N |+|E|.Then its MM distance is minimum public hypergraph and the difference of the size of the public subgraph of maximum, be MMDist=| MCSup |-| MCSub|。

Positional distance: positional distance is to have an impact all nodes minimum in reference model in the subgraph of upstream by calculating Rearward position obtains, and node rearward position is defined as follows:

R=(N, E, L, n, α) is made to represent reference flowchart,Represent the node to be recommended of reference flowchart R Collection.In (x) and Out (x) represents input node collection and the output node collection of node x respectively.Num (N) represents set N interior joint Number.All x, y, z ∈ (N ∪ RR).The rearward position of node x is:

$\mathrm{Lop}\left(x\right)=\left\{\begin{array}{c}0,x\∈\mathrm{RR}\\ \mathrm{Lop}\left(y\right)+1,x\∈i\left(y\right),\mathrm{mum}\left(o\left(i\left(y\right)\right)\right)=1\\ \max (\mathrm{Lop}\left(z\right)+1),z\∈o\left(i\left(x\right)\right),\mathrm{num}\left(o\left(i\left(x\right)\right)\right)>1\end{array}\right..$

The calculating of positional distance is divided into four steps: (1) first finds out the loop structure in reference flowchart, respectively by different nothings Articulare (independent of active node) replaces each the most independent loop structure.If there being two or more circulation Structure has common node, then these circulations replaced without articulare with same, thus obtain without circulation figure；(2) according to node Figure without circulation obtained in the previous step is processed by the definition of rearward position, obtains the rearward position of each node；(3) by unrelated Node revert to before loop structure, and the positional distance without articulare is assigned to each node inside corresponding circulation, this Sample has just obtained the rearward position of all nodes of reference flowchart model；(4) last, according to the node rearward position distance obtained Lo, calculates the positional distance between reference flowchart model and upstream subgraph, specifically, to the node in the word figure of upstream by leaf node Proceed by traversal, the node that first found all occurs in reference flowchart and upstream subgraph, the node of this node to Rear positional distance Lo is the positional distance between reference flowchart and upstream subgraph, if additionally, reference flowchart or upper alien Figure exists circulation, then needs also exist for first eliminating circulation (as described in above-mentioned steps (1)), calculate positional distance the most again.

The substantially running of above-mentioned flow process commending system is:

One) all flow processs represented with various modeling languages are modeled as again oriented labeled graph, generate standardization Flow process storehouse.

Two) utilize figure mining algorithm-gSpan based on depth-first search to excavate flow process storehouse, obtain all flow process subgraphs.

Three) each flow process subgraph is processed, build mould according to upstream subgraph, both candidate nodes, confidence level three category information Formula table.

Four) flow process built obtaining modeling personnel is carried out as reference flowchart and with the upstream subgraph in pattern table Coupling, recommends modeling personnel by both candidate nodes corresponding for the upstream subgraph mated most.

Five) four are repeated, until process modeling completes, the flow process filing that then will complete, store in flow process storehouse.

Due to the fact that and have employed above technical scheme that there is significant technique effect:

Recommend efficiency high: the matching strategy that native system is used is based on maximum public subgraph and the calculation of the public hypergraph of minimum Method, mutually than ever based on matching algorithms such as GED, SED, has relatively low time complexity, wherein GED computation complexity lower limit For, and the computation complexity of SED algorithm is, and the complexity of MM algorithm is linear, i.e. therefore the recommendation efficiency of native system is wanted Higher than traditional recommendation method.

Recommend accuracy high: the common portion that the matching algorithm that native system is used considers not only two flow charts is (public Subgraph altogether), it is also considered that the difference (public hypergraph) between two figures, it is possible to preferably calculate the distance of two flow charts, i.e. There is more preferable matching effect, recommendation process has higher accuracy rate.

Support the process to the flow process comprising the labyrinths such as circulation.Native system calculates by improving conventional positional distance Method, supports the labyrinth such as loop structure, parallel branch, has more preferable practicality.

Further, in flow process recommendation step, realize comparing by calculating MM distance and positional distance, thus draw Good recommended flowsheet, its calculation procedure and amount of calculation are less.

By the method that obtained recommended flowsheet is added flow process storehouse, constantly flow process storehouse can be enlarged, expand Open up the Data Source of pre-treatment step, expanded the range of choice of recommendation, improve the accuracy of recommendation.Additionally, user is also Flow process collection can be directly inputted by subscriber interface module.

Use special Frequent Subgraph Mining, improve excavation precision, also reduce further the complicated journey of algorithm Resource overhead required during degree and calculating.

Accompanying drawing explanation

Fig. 1 is the modular structure schematic diagram that flow process recommends method.

Fig. 2 is the schematic flow sheet that flow process recommends method.

Detailed description of the invention

Below in conjunction with embodiment, the present invention is described in further detail.

Embodiment 1

Excavate based on figure and the flow process of map distance recommends method, in order to realize process modeling automatization, system architecture such as figure 1, including with lower module:

User interface 1: this module mainly processes the input and output of user, provides the user interactive function, including flow process literary composition Part (flow process collection) upload and modeling personnel carry out process modeling.

Pretreatment module 2: this module is by abstract for the flow process collection of the input form being numbered directed graph, if flow process collection is master stream The set of journey, flow process mentioned here includes: technological process, i.e. the program of the every procedure arrangement from raw material to manufactured goods；Industry Business flow process, i.e. by two and above business step, completes the process of a complete business conduct, can be referred to as flow process, note Meaning is two and above business step；And the ongoing order of flow of transactions, i.e. things or order layout and arrangement.Should Module also includes a submodule: subgraph digs evidence and decomposing module 21, and this submodule utilizes efficient figure mining algorithm (gSpan) Carry out the excavation of operation flow subgraph, then by decomposition process subgraph, obtain upstream subgraph, both candidate nodes collection and confidence level also Being registered in pattern table, its middle and upper reaches subgraph is for mating with the current reference model that builds, and both candidate nodes collection is then used for Build flow process.Pretreatment module 2 also includes schema extraction step, i.e. finds useful flow process from the subgraph that excavation obtains, and Store in pattern table after certain process, concretely comprise the following steps: 1) decompose and obtain having an impact upstream subgraph and both candidate nodes；2) Calculating the confidence level having an impact upstream subgraph and both candidate nodes, described confidence level is when having an impact upstream subgraph and occurring, under One node is the probability of both candidate nodes, and circular is to have an impact upstream subgraph and probability that both candidate nodes occurs simultaneously Divided by having an impact the probability that upstream subgraph individually occurs；3) obtained is had an impact upstream subgraph, both candidate nodes and confidence Degree stores to pattern table 3.

Pattern table module 3: pattern table module 3 mainly preserves the Data Entry of pretreatment module generation and (has an impact alien Figure, both candidate nodes collection and confidence level), and provide satisfactory entry for recommending module.

Recommending module 4: this module has four kinds of proposed algorithms based on MM, GED, SED and NMSF.Recommending module 4 is first Obtain the flow process (reference flowchart) constructed by modeling personnel, and the upstream subgraph that reference flowchart and pattern table provide is compared Relatively, the entry mated most, then according to matching result, corresponding both candidate nodes collection is recommended modeling personnel.Recommending module In 4, various piece constitutes the execution loop of a Guan Bi, triggers recommendation process according to reference flowchart model, by calculating with reference to mould Type and the distance having an impact between the subgraph of upstream, find suitable both candidate nodes collection to recommend modeling personnel, this process meeting one Straight circulation performs until flow process completes to build.Once finishing service process modeling, this model will be placed in flow process storehouse, as The data source of process modeling in the future.

Concrete execution process is as in figure 2 it is shown, mainly comprise the steps that

Pre-treatment step: this stage, by the form of the most abstract for all of flow process label directed graph, then utilizes efficient figure Mining algorithm (gSpan) carries out the excavation of operation flow subgraph, but stronger in order to prevent the relatively low relatedness of frequency ratio Subgraph be eliminated, so subgraph is occurred that frequent degree is set to 0.Export the intersection of all flow process subgraphs and correspondence The frequency of occurrences.

Mode discovery step: this stage based on processing stage result, by decomposition process subgraph, obtain upstream subgraph and Both candidate nodes collection, its middle and upper reaches subgraph is for mating with the current reference model that builds, and both candidate nodes collection is then used for building Flow process.Wherein confidence level is selected more than the upstream subgraph of certain threshold value, i.e. to have an impact upstream subgraph, and upstream subgraph will be had an impact Upstream subgraph, both candidate nodes collection and confidence level are registered in pattern table 3 as Data Entry.

Flow process recommendation step: first this stage obtains the flow process (reference flowchart) constructed by modeling personnel, and by reference stream Journey compares with the upstream subgraph in pattern table 3, and the entry mated most, then according to matching result corresponding candidate Set of node recommends modeling personnel.Specifically, this stage comprises step in detail below:

Upstream subgraph p and the process of reference flowchart R are had an impact for each in pattern table 3, obtains the minimum of p and R Public subgraph and the public hypergraph of maximum, be then calculated the MM distance of correspondence.

Node rearward position according to reference flowchart R, obtains the positional distance Lo of p and R.

According to MM distance and positional distance Lo, obtaining total distance of p and R, total distance is a*MM+b*Lo, generally, a, b Being respectively 1,6, and add in CNS by both candidate nodes collection, distance and confidence level, CNS is both candidate nodes data set, including candidate Set of node, distance and confidence level.

By the entry in CNS according to the ascending sequence of distance value (if apart from equal, then according to confidence level from high to low Sequence), it is then back to front n result and recommends modeling personnel.

In a word, the foregoing is only presently preferred embodiments of the present invention, all equalizations made according to scope of the present invention patent Change and modification, all should belong to the covering scope of patent of the present invention.

Claims (6)

1. one kind is excavated and the flow process recommendation method of map distance based on figure, it is characterised in that specifically comprise the following steps that

Pre-treatment step: by abstract for the flow process collection of the input form being numbered directed graph, obtain flow process subgraph, described flow process Ji Bao Include technological process, operation flow and flow of transactions, use Frequent Subgraph Mining that flow process subgraph digs evidence, output bag Containing all intersections including flow process subgraph and the subgraph frequency of occurrences thereof；Mode discovery step: subgraph digs evidence and decomposing module (21) The intersection exporting described pre-treatment step is decomposed, and obtains having an impact upstream subgraph, both candidate nodes collection and confidence level, will Gained has an impact upstream subgraph, both candidate nodes collection and confidence level and is registered as the Data Entry in pattern table (3), described flow process The end node of figure is as both candidate nodes, and remainder, as upstream subgraph, selects both candidate nodes to concentrate confidence level more than threshold value Upstream subgraph, be and have an impact upstream subgraph；

Flow process recommendation step: recommending module (4) obtains reference flowchart, described reference flowchart is inputted by subscriber interface module (1), will Reference flowchart compares with the upstream subgraph that has an impact in pattern table (3), selects matched data entry, by matched data Both candidate nodes collection corresponding to entry is output as recommended flowsheet；

In flow process recommendation step, the step of described comparison specifically includes:

1) each in pattern table (3) is had an impact upstream subgraph p and the process of reference flowchart R, obtain described p and R Mini-bus hypergraph MCSub and maximum public subgraph MCSup, is calculated MM distance, and described MM distance is minimum public hypergraph The difference of the size of MCSub and maximum public subgraph MCSup, i.e. MMDist=| MCSup |-| MCSub |；

2) according to the node rearward position of reference flowchart R, the positional distance Lop of described p and R, described node rearward position are obtained For: making R=(N, E, Ln, α) represent reference flowchart, RR (RR ∩ N=φ) represents the set of node to be recommended of R, In (x) and Out (x) Representing input node collection and the output node collection of node x respectively, num (N) represents the number of set N interior joint, x, y, z ∈ N ∪ RR；The node rearward position of node x is:

$Lop\left(x\right)=\left\{\begin{array}{c}0,x\∈RR\\ Lop\left(y\right)+1,x\∈In\left(y\right),num\left(Out\right(In\left(y\right)\left)\right)=1\\ max\left(Lop\right(z)+1),z\∈Out\left(In\right(x\left)\right),num\left(Out\right(In\left(y\right)\left)\right)>1\end{array}\right.;$

The calculation procedure of described positional distance Lop includes: 1 ') find out the loop structure in described reference flowchart R, respectively by difference Replace the most independent loop structure without articulare, described is the node independent of active node without articulare, if Common node is there is, then by said two or plural loop structure between two or more loop structure Replace without articulare with same respectively, obtain the figure without circulation of reference flowchart R；2 ') obtain in figure without circulation figure according to gained The node rearward position of node；3 ') by without circulation figure reverts to step 1 without articulare ') replace it before loop structure, And the positional distance without articulare is assigned to the node of loop structure, obtain in reference flowchart R the node of all nodes backward Position；4 ') according to step 3 ') obtained by node rearward position, obtain described positional distance Lop；

3) obtain total distance of described p and R according to described MM distance and positional distance Lop, and by both candidate nodes collection, total distance with And confidence level adds to CNS；

4) entry in CNS is always ranked up apart from ascending according to described, if apart from identical, then according to confidence level by High to Low it is ranked up, then elects multiple results forward for position as matched data entry.

Excavate based on figure the most according to claim 1 and the flow process of map distance recommends method, it is characterised in that also include Concrete steps: described flow process recommendation step also includes obtained recommended flowsheet is added flow process storehouse, and described flow process storehouse is for pre- Process the Data Source as the flow process collection inputted in step.

Excavate based on figure the most according to claim 1 and the flow process of map distance recommends method, it is characterised in that described pretreatment Flow process collection in step also includes the flow process collection inputted by subscriber interface module (1).

Excavate based on figure the most according to claim 2 and the flow process of map distance recommends method, it is characterised in that circulation performs institute State pre-treatment step, mode discovery step and flow process recommendation step.

Excavate based on figure the most according to claim 1 and the flow process of map distance recommends method, it is characterised in that described frequent son Figure mining algorithm includes step in detail below: Land use models increases strategy, uses the convenient pattern search space of depth-first fashion, On the basis of oneself knows Frequent tree mining pp, extension produces the sub-Frequent tree mining of described pp, and described sub-Frequent tree mining is Frequent tree mining Child node, and calculate the support of described sub-Frequent tree mining；To the sub-Frequent tree mining of pp each described, with depth-first Mode continue extension, until find all of frequent sub-because of till.

Excavate based on figure the most according to claim 1 and the flow process of map distance recommends method, it is characterised in that described pattern is sent out Existing step also includes pattern table construction step, specifically comprises the following steps that decomposition process subgraph, obtains upstream subgraph and candidate's joint； Calculate the confidence level of each upstream subgraph and both candidate nodes, according to upstream subgraph, both candidate nodes and confidence level forming types Table, the Data Entry of described pattern table is that (I, C, f), wherein, I is the finite aggregate having an impact upstream subgraph to tlv triple T= Closing, C is the finite aggregate of both candidate nodes, and f:I → C is surjective function.