CN109284086B

CN109284086B - Demand-oriented adaptive dynamic evolution method for Web service

Info

Publication number: CN109284086B
Application number: CN201810939197.3A
Authority: CN
Inventors: 陆佳炜; 吴涵; 卢成炳; 赵伟; 周焕; 徐俊; 肖刚
Original assignee: Zhejiang University of Technology ZJUT
Current assignee: Zhejiang University of Technology ZJUT
Priority date: 2018-08-17
Filing date: 2018-08-17
Publication date: 2021-05-18
Anticipated expiration: 2038-08-17
Also published as: CN109284086A

Abstract

A demand-oriented adaptive Web service dynamic evolution method comprises the following steps: firstly, calculating semantic similarity between two concepts A and B in a domain ontology; second step, providing service S₁And service S₂Input similarity Sim_inputThe calculation method of (1); third step, providing service S₁And service S₂Output similarity Sim_outputThe calculation method of (1); fourth step, calculation service S₁And service S₂Functional similarity functional sim (S) of₁,S₂) (ii) a Step five, providing a service clustering algorithm by combining the service function similarity measurement mode in the step four; sixthly, obtaining the center service of each cluster according to the clustering result in the fifth step; seventhly, respectively calculating the functional similarity between the RCT node requirement description and each cluster center service description; eighthly, screening candidate service sets CandidateServices according to the QoS threshold in the user request; and ninthly, performing comprehensive service scoring and ranking on the candidate service sets CandidateServices filtered in the eighth step. The invention can select the best service meeting the user requirements.

Description

Demand-oriented adaptive dynamic evolution method for Web service

Technical Field

The invention relates to the field of Web service evolution, in particular to a demand-oriented adaptive dynamic Web service evolution method.

Background

A Web service is a software system intended to support cross-network interactions between machines. There are currently two main types of Web services: one is based on SOAP and the other is based on REST. The difference between the two is the use of different interfaces, SOAP-based Web services use the SOAP interface to pass messages and use the Web Services Description Language (WSDL) to describe Web services, which specifies a protocol-and code-independent mechanism for Web service providers, which is an XML vocabulary that describes services accessible on the network and maps them into a collection of communication endpoints with messaging capabilities. While Web services using REST interfaces use generic HTTP methods (GET, DELETE, POST, and PUT) to describe, publish, and use relevant resources.

Current research efforts aim at providing semantic descriptions of Web services by using conceptualized knowledge called ontologies. An ontology is a vocabulary that describes a set of concepts within a domain (a domain may be defined as a particular subject domain or knowledge domain) and the relationships that exist between those concepts. It is applied to attribute reasoning within a domain, or to the definition of the domain itself. In the context of Web services, ontologies play an important role as a way to provide semantic descriptions of Web services. The enhancement of Web services descriptions has prompted the development of semantic Web services, which are described in a machine-understandable manner, which will have a significant impact on areas such as e-commerce and application integration, since it can enable dynamic, extensible and efficient collaboration between different systems and organizations.

With the continuous development of Web services, in order to adapt to changes in environment and changes in user requirements, the Web services in the internet need to evolve continuously. Therefore, Web services have evolved as one of the important research points in the field of service computing. Meanwhile, the Web service is an important technology for constructing the software service, and important research significance and application value are provided for enabling the software system to operate in a self-adaptive mode and supporting dynamic evolution of the service.

Web service evolution generally refers to a process of performing a series of changes after a service is released and operated in order to adapt to environmental changes and continuously meet user requirements. According to the evolution implementation mode, the service evolution can be divided into static evolution and dynamic evolution, the static evolution must pass through the processes of suspending service, updating software and restarting service, and the dynamic evolution refers to the evolution of software during operation, namely, updating while running.

Currently, a lot of research has been conducted on the dynamic evolution of Web services by many researchers. Seung Hwan Ryu et al, university of New Nanwelshi, focuses on dynamic evolution of service composition at the service orchestration level and proposes a method for determining the migratability of a service composition instance. Vasillios Andrikopoulos et al, the university of thelberg, the netherlands, introduced the concept of service evolution management and studied the consistency problem of various versions of services (including composite services) generated in the evolution process. Sony, wei, et al, at the university of tokyo, classified dynamic evolution into two categories: ad-hoc (ad-hoc) evolution and evolutionary (evolutionary) evolution. Wherein the ad hoc evolution affects only specific process instances, which are typically caused by process instances encountering an anomaly during operation; the evolutionary evolution is caused by the modification and the upgrade of the process model, the modification of the model necessarily affects all the in-operation process instances, and the evolution of the process instances caused by the upgrade of the process model is called the evolutionary evolution.

Disclosure of Invention

In order to adapt to the rapidly changing user requirements, the Web application needs to effectively understand and manage the requirements in a flexible and self-adaptive manner, and then a reusable software unit is constructed to enhance the software evolution capability. The invention provides a dynamic service evolution method facing to demand self-adaptation. The method utilizes an improved DBSCAN algorithm to cluster services, calculates the distance between samples by combining the functional similarity between the services in the clustering process, and provides a service dynamic matching method based on an RCT model to realize service matching on the basis of service clustering, wherein the process comprises service filtering, service comprehensive scoring and service ranking.

In order to solve the technical problems, the invention adopts the technical scheme that:

a demand-oriented adaptive Web service dynamic evolution method comprises the following steps:

the method comprises the following steps of firstly, calculating semantic similarity between two concepts A and B in a domain ontology, wherein the process is as follows: the functional similarity measurement considers the domain ontology concepts of input and output between services, the matching between the input and the output refers to the matching of concepts related to the input and the output, and in order to calculate the similarity of the two concepts A and B, the relationship between the two concepts in the domain ontology needs to be considered;

a second step of providing a service S by combining the concept similarity calculation method of the first step₁And service S₂Input similarity Sim_inputThe calculation method of (1);

step three, combining the concept similarity calculation method of the step one, and providing a service S₁And service S₂Output similarity Sim_outputThe calculation method of (1);

the fourth step combines the service input similarity Sim obtained in the second step and the third step_inputSimilarity with service output Sim_OutputComputing service S₁And service S₂Functional similarity functional sim (S) of₁,S₂) The calculation formula is as follows:

FunctionalSim(S₁,S₂)＝w₁×Sim_input+w₂×Sim_Outputwherein the weight w₁And w₂Are real numbers between 0 and 1 and sum to 1, which represent the importance of the service consumer to the input similarity and output similarity determinations;

step five, a service clustering algorithm is given by combining the service function similarity measurement mode of the step four, firstly, an unallocated core object is selected arbitrarily as a seed, and then all sample sets with reachable density of the core object are searched to form a cluster; then, continuously selecting another unallocated core object to search a sample set with a reachable density to obtain another cluster until all the core objects are allocated; search radius Eps is given by

Determining, wherein n represents the dimension size of the service, i.e. the average attribute number of all services, k represents the minimum number of services required to form a cluster, and r represents the hypersphere hemisphere composed of n-dimensional servicesThe diameter m is the number of n-dimensional services contained in the hypersphere; while the minimum number MinPts follows empirical heuristics: MinPts is round (m/25), where round is the rounding function and m is the number of samples in the sample set;

sixthly, according to the clustering result obtained in the fifth step, obtaining the center service of each cluster, wherein the cluster center service can represent the typical characteristics of the clustering service;

seventhly, in the service matching process of the system, functional similarity between RCT node requirement description and each cluster center service description needs to be calculated respectively, wherein the RCT nodes are constituent units of an RCT model and are described as Web services needed for completing current Web subtasks, and an iSar framework defines four dependency relationships in a policy dependency model, which are: target dependency, meaning that one role depends on another role to accomplish a certain goal; resource dependence, meaning that one role depends on another role to provide it with physical resources and information; task dependency, meaning that one role depends on another role to complete a certain task; non-functional demand related soft target dependencies, similar to target dependencies; the atomic concepts of roles, targets, tasks and resources of the iSAR are expanded to support mapping relations and related description specifications among user requirements, Web tasks, services and service sequences, and the RCT model is instantiated by an OWL-S format file so as to ensure normal execution of a service matching process; in the RCT model, an ellipse is used for representing user requirements, namely RCT requirement nodes, a hexagon is used for representing Web subtasks, well numbers above the left side of the hexagon nodes are used for representing serial numbers of the Web subtasks, a circle is used for representing services, and a rectangle is used for representing information resources;

because OWL-S is a language designed for describing Web services and is suitable for describing the requirements of services, the requirements of RCT nodes are also described by adopting OWL-S, the functional similarity measurement method of the fourth step is also suitable for calculating the similarity between the RCT nodes and the cluster center services, the cluster where the cluster center services with the maximum functional similarity are located is obtained by calculation, and the services in the cluster are candidate services Candidateservice which can meet the requirements of current users;

eighthly, screening candidate service sets CandidateServices according to the QoS threshold value in the user request, filtering out services which do not meet the requirements of the user, and enabling a service consumer to designate the QoS threshold value in the service request to indicate that Web services with the QoS lower than or higher than the threshold value are rejected;

ontology-based OWL-S extension, namely adding non-functional description QoS into Web service description;

OWL-S is called totally on-load Web Language for Services, it is Web service Ontology description Language based on OWL, use a complete set of description mechanism to carry on the semantic description to the Web service, and the description file that is produced can be understood and processed by the computer;

OWL-S includes three basic components:

(1) the Service Profile describes what the Service does, and the basic attributes comprise a Service name (serviceName), a text description (textDescription) and the like; the functional attributes comprise four parts of input information (hasInput), output information (hasOutput), preconditions (hasRecondition) and expected effects (hasEffect);

(2) service Model describes how services work, OWL-S treats each Service as a Process (Process) and divides it into three categories: atomic processes (Atomic processes), Simple processes (Simple processes), and Composite processes (Composite processes); the atomic process has no subprocess and can be directly called; the simple process can not be directly called, and a corresponding Service group description file does not exist; the combined process is formed by combining a plurality of sub-processes through a certain control structure, and in OWL-S, each process body can be semantically described through IOPE;

(3) service grouping describes Service calling information, OWL-S defines a specific message format by using WSDL as the basis of a Service grouping mechanism, and encapsulates input information, output information and a Service address;

constructing an ontology-based OWL-S extended semantic service description model based on a Web service description language OWL-S;

the semantic service description is represented by a quadruplet: WS ═ N, D, I, O >, where:

n represents the Name of the service, D represents the text Description of the service, and the N and D form the basic attribute of the service;

the functional attributes of the service include I and O, which represent the attributes of the Input (Input) and Output (Output) interface parameters of the service WS, respectively, each interface is a set of parameter sets, and I ═ I₁,…,i_m}, element i₁,…,i_mRepresenting semantic concepts corresponding to input parameters of the service WS in the domain ontology; o is the output set of the service, O ═ O₁,…,o_n}, element o₁,…,o_nRepresenting semantic concepts corresponding to the output parameters of the service WS in the domain ontology;

in the OWL-S Service Profile, a group of Service parameters (ServiceParameter) is used, the Service parameters comprise a name (serviceParameterName) and a parameter value (Sparamer), after the OWL-S is connected with a QoS body, QoProperty becomes a subclass of the OWL-S serviceParameter, and qoPropertyName and qProperty are respectively the sub-attributes of the OWL-S serviceParameterName and the Sparamer attribute;

each QoS attribute QoSProperty is defined by a name in the form of a string, qoospropertyname, and a set of characteristics qoschharacteristic, described as follows:

value: a value representing a QoS attribute, which represents one QoS attribute of a provided service to a service provider; but for the service consumer it represents the threshold for the QoS attribute;

type: this feature is used to distinguish between two types of QoS:

revenue-type benefit, in which case the user-specified QoS attribute value represents the minimum value to consider;

cost type cost, in which case the user specified QoS attribute value represents the maximum value to be considered;

unit: a unit of measure of the QoS attribute;

weight: a QoS weight for specifying the importance of the QoS attribute;

confidence: representing a confidence level of the user's preference for it;

ninthly, performing comprehensive service scoring and ranking on the candidate service sets CandidateServices filtered in the eighth step, wherein the service with the highest score is the best service; before service scoring, consistency verification needs to be performed on the QoS weight determined by the user, if the obtained consistency ratio CR is smaller than 0.1, scoring and ranking steps are performed, otherwise, the QoS weight needs to be revised again by the user, and the calculation is finished.

Further, in the first step, the step of calculating the semantic similarity between the two concepts a and B in the domain ontology is as follows:

step (1.1) if the concepts A and B are identical or they are declared equivalent classes, the similarity Sim of the concepts A, B _concept1, otherwise, performing the step (1.2);

step (1.2) if the concept A is directly or indirectly a subclass of the concept B, the similarity Sim of the concepts A and B_conceptThe calculation formula is as follows:

wherein prop (A) and prop (B) represent the attribute sets of concept A and concept B, respectively, and Size (prop (B)) and Size (prop (A)) represent the attribute numbers of concept B and concept A, respectively, otherwise, the step (1.3) is carried out

Step (1.3) if the concept B is directly or indirectly a subclass of the concept A, the similarity Sim of the concepts A and B_conceptThe calculation formula is as follows:

otherwise, performing the step (1.4);

step (1.4) if the concept A and the concept B have no parent-child relationship, but the two concepts directly or indirectly have a common parent class concept C, a word semantic similarity measurement method based on a naive Bayes model is adopted, firstly, all attributes of the concept A and the concept B are respectively traversed, the attribute names of the concept A and the concept B are subjected to feature extraction through a computeFeature function, and then, a conditional probability distribution list and an adjustment factor after sample training are adopted to calculate the similarity Sim between the concept attributes_wordAccording to the similarity Sim between the concept attributes_wordIs likeComparing the judgment factor eta, judging whether the two attributes are the same attribute, counting, and finally calculating the similarity Sim of the concepts A and B_concept；

Step (1.5) if the relationship between the concept A and the concept B does not meet the above condition, the similarity Sim of the concepts A and B_conceptIs set to 0.

Still further, the process of step (1.4) is as follows:

step (1.4.1) setting variable i to represent the same number of attributes in concept A and concept B and setting the initial value to be 0, and performing step (1.4.2);

step (1.4.2) if the traversal of the attribute set prop (A) of the concept A is completed, then step (1.4.7) is performed, otherwise, the next prop (A) is taken out from prop (A)_jAnd removing it from prop (A) and carrying out step (1.4.3);

step (1.4.3) if the traversal of the attribute set prop (B) of the concept B is completed, returning to the step (1.4.2), otherwise, taking out the next prop (B) from the prop (B)_kAnd removing it from prop (B) and carrying out step (1.4.4);

step (1.4.4) is based on naive Bayes model, combines WorkNet English dictionary, and pairs of prop (A) through ComputeFeature function_jAnd prop (B)_kThe attribute name of (2) is subjected to feature extraction to obtain L (prop (A)_j,prop(B)_k) And D (prop (A)_j,prop(B)_k) The process is as follows:

calculating the word semantics of each attribute name, wherein each word corresponds to one or more semantics, each word pair also corresponds to one or more semantic pairs, and the semantic node distance with the shortest distance among all the semantic pairs corresponding to the word pair is defined as the word pair distance L (prop (A)_j,prop(B)_k) And defining the semantic pair depth with the shortest distance between semantic nodes as word pair depth D (prop (A)_j,prop(B)_k) Known attribute names prop (A)_jPresence in semantic node v_j1,v_j2,…，v_jnIn the synonymous phrase of (1), attribute name prop (B)_kPresence in semantic node v_k1,v_k2，…v_kmIn a synonymous phrase of (1), then prop (A)_jAnd prop (B)_kThe distance calculation formula and the depth calculation formula are as follows:

wherein L (v)_ja,v_kb) Representing semantic nodes v_jaAnd semantic node v_kbDistance of D (v)_ja,v_kb) Representing semantic pairs (v)_ja,v_kb) Depth of (d);

further, mean functions LW (i) and DW (o) are generated according to a training set of the naive Bayes model, and then a conditional probability distribution column P (L (prop (A)) is calculated by utilizing the mean functions LW (i) and DW (o)_j,prop(B)_k) I C and P (D (prop (A))_j,prop(B)_k) I C), wherein C is a word class classification, the value range is { U, N }, wherein U represents 'consistent' and N represents 'inconsistent', and finally, the adjustment factors alpha and beta are calculated according to the following calculation formula:

then, the step (1.4.5) is carried out;

step (1.4.5) based on naive Bayes model, the characteristic L (prop (A) of ontology concept attribute is processed_j,prop(B)_k) And D (prop (A)_j,prop(B)_k) Injecting the mixture into the conditional probability distribution list obtained in the step (1.4.4), and sequentially extracting the conditional probability V₁＝P(L(prop(A)_j,prop(B)_k)＝i|C＝U)，V₂＝P(D(prop(A)_j,prop(B)_k)＝o|C＝U),

V₃＝P(L(prop(A)_j,prop(B)_k) I | C ═ N) and V₄＝P(D(prop(A)_j,prop(B)_k) O | C ═ N), finally combining the adjustment factors α and β in step (1.4.4) to calculate prop (a)_jAnd prop (B)_kSimilarity between them Sim_wordThe calculation formula is as follows:

Sim_word(prop(A)_j,prop(B)_k)＝(αV₁×V₂)/(αV₁×V₂+βV₃×V₄) Carrying out the step (1.4.6);

step (1.4.6) if Sim_wordGreater than or equal to the similarity criterion factor eta, prop (A)_jAnd prop (B)_kAdding 1 to the variable i for the same attribute, returning to the step (1.4.2), and otherwise, returning to the step (1.4.3);

step (1.4.7) calculating similarity Sim of concepts A and B_conceptThe calculation formula is as follows:

wherein i represents the number of attributes in concept A and concept B, prop (A) and prop (B) represent the attribute sets of concept A and concept B, respectively, and Size (prop (B)) and Size (prop (A)) represent the number of attributes in concept B and concept A, respectively, and step (1.5) is performed.

Further, in the second step, the service S₁And service S₂Input similarity Sim_inputThe process of the calculation method of (1) is as follows:

step (2.1) establishing a service input parameter similarity maximum matching array InSim and initializing, and performing step (2.2);

step (2.2) service S₁Is subtracted by the service S₂Obtaining a parameter quantity difference d by inputting the parameter quantity, and performing the step (2.3);

step (2.3) if d is less than or equal to 0, then service S₁Is set to S_shortService S₂Is set to S_longOtherwise, it will serve S₂Is set to S_shortService S₁Is set to S_longCarrying out the step (2.4);

step (ii) of(2.4) traverse S_longIf the traversal is completed, go to step (2.8), otherwise go from S_longIn the next input parameter long is taken out_iCarrying out the step (2.5);

step (2.5) traverse S_shortIf the traversal is completed, returning to the step (2.4), otherwise, returning to the step S_shortGet the next input parameter short_jCarrying out the step (2.6);

step (2.6) calculating parameter Long according to the concept similarity calculation method of the first step_iAnd parameter short_jSimilarity Sim of_ijCarrying out the step (2.7);

step (2.7) with Sim_ijAnd InSim [ i ]]By comparison, if Sim_ijGreater than InSim [ i]Then, will InSim [ i ]]Is set to Sim_ijOtherwise InSim [ i ]]Returning to the step (2.5) when the value is the original value;

step (2.8) calculation service S₁And service S₂Input similarity Sim_inputThe calculation formula is as follows:

wherein Size (S)_longInput) and Size (S)_shortInput) represents the service S, respectively_longNumber of input parameters and service S_shortThe number of input parameters, | d | represents the difference between the number of input parameters of two services, and InSim is the maximum matching array of the similarity of the input parameters.

In the third step, service S₁And service S₂Output similarity Sim_outputThe process of the calculation method of (1) is as follows:

step (3.1) creating a service output parameter similarity maximum matching array OutSim and initializing, and performing step (3.2);

step (3.2) service S₁Minus the service S₂Obtaining a parameter quantity difference d according to the output parameter quantity, and performing the step (3.3);

step (3.3) if d is less than or equal to 0,will serve S₁Is set to S_shortService S₂Is set to S_longOtherwise, it will serve S₂Is set to S_shortService S₁Is set to S_longAnd (4) performing the step (3.4);

step (3.4) traverse S_longIf the traversal is completed, go to step (3.8), otherwise go from S_longIn the next output parameter long_iAnd (5) carrying out the step (3.5);

step (3.5) traverse S_shortIf the traversal is completed, the step (3.4) is returned, otherwise, the step S is started_shortIn the next output parameter short_jAnd (4) carrying out the step (3.6);

step (3.6) calculating the parameter Long according to the concept similarity calculation method of the first step_iAnd parameter short_jSimilarity Sim of_ijCarrying out the step (3.7);

step (3.7) with Sim_ijAnd OutSim [ i ]]By comparison, if Sim_ijGreater than OutSim [ i ]]Then OutSim [ i ] is set]Is set to Sim_ijElse OutSim [ i ]]Returning to the step (3.5) when the value is the original value;

step (3.8) calculation service S₁And service S₂Output similarity Sim_OutputThe calculation formula is as follows:

wherein Size (S)_long.Output) and Size (S)_shortOutput) respectively represent the services S_longNumber of output parameters and service S_shortThe number of output parameters, | d | represents the difference value of the number of output parameters of two services, and OutSim is the maximum matching array of the similarity of the output parameters.

In the fifth step, the process of the service clustering algorithm is as follows:

step (5.1) initializing an unvisited sample set into an original sample set T, dividing a cluster set C into empty sets, and performing step (5.2);

step (5.2) traverse the originalStarting a sample set T, if traversal is completed, ending, otherwise, taking out the next service S from T_iAnd removing it from T, proceeding to step (5.3);

step (5.3) combines the service function similarity measurement mode defined by step four to search service S_iThe Eps neighborhood of (c);

step (5.4) creation of a New Cluster C_kIf S is_iIf the number of Eps neighborhood samples is not less than MinPts, then S is determined_iAdding the mixture into the cluster, performing the step (5.5), and otherwise, returning to the step (5.2);

step (5.5) traversal service S_iNeighborhood set N of_Eps(S_i) If the traversal is completed, the step (5.13) is carried out, otherwise the set N is taken out_Eps(S_i) Of (2) next service S'_kAnd (5.6) carrying out the step;

step (5.6) creating service S'_kNeighborhood set N of_Eps(S’_k) Carrying out the step (5.7);

step (5.7) if the original sample set T contains service S'_kThen service S'_kRemoving from T, performing step (5.8), otherwise, returning to step (5.5);

step (5.8) traverses the original sample set T, if the traversal is finished, the step (5.11) is carried out, otherwise, the next service S 'is taken out from the T'_lCarrying out the step (5.9);

step (5.9) adopts the service function similarity measurement mode defined in step four to calculate service S'_kAnd service S'_lFunctional similarity functional Sim (S'_k,S’_l) Go to step (5.10)

Step (5.10) if service S'_kAnd service S'_lFunctional similarity functional Sim (S'_k,S’_l) Is less than or equal to the search radius Eps, then service S'_lAdd service S'_kNeighborhood set N of_Eps(S’_k) Returning to the step (5.8), otherwise, directly returning to the step (5.8);

step (5.11) if service S'_kIf the number of Eps neighborhood samples is not less than MinPts, the service in the neighborhood is added to the serviceS_iEps neighborhood set N_Eps(S_i) And (5.12) performing the step (5.12), otherwise, returning to the step (5.5);

step (5.12) if the cluster set C does not contain service S'_kCluster of (2), then service S'_kAdding the cluster C created in the step (5.4)_kReturning to the step (5.5), otherwise, directly returning to the step (5.5);

step (5.13) clustering the clusters C_kAnd adding the cluster set C and returning to the step (5.2).

In the step (5.3), the search service S_iThe Eps neighborhood of (c) is as follows:

step (5.3.1) Create service S_iNeighborhood set N of_Eps(S_i) Carrying out the step (5.3.2);

step (5.3.2) traverses the original sample set T, if the traversal is completed, the step (5.4) is carried out, otherwise, the next service S is taken out from the T_jCarrying out the step (5.3.3);

step (5.3.3) uses the functional similarity defined in step four to calculate service S_iAnd service S_jFunctional similarity functional sim (S)_i,S_j) And (5.3.4) carrying out the step;

step (5.3.4) if service S_iAnd service S_jFunctional similarity functional sim (S)_i,S_j) Less than or equal to the search radius Eps, the service S will be provided_jJoining service S_iNeighborhood set N of_Eps(S_i) And (5.3.2) returning, otherwise, directly returning to the step (5.3.2).

In the sixth step, the calculation process of the cluster center service is as follows:

step (6.1) creation of the list S_centerRecording the service of each cluster center, and performing the step (6.2);

step (6.2) traversing the cluster set C created in the fifth step, if the traversal is completed, ending the process, otherwise, taking out the next cluster C from the cluster set C_kCarrying out the step (6.3);

step (6.3) of establishing a variable in.num and out.num pair cluster C_kAll input parameters and output parameters ofLine statistics is carried out, initial values of two variables are set to be 0, and a cluster center service S is created_kcenterAnd (6.4) carrying out the step;

step (6.4) traverse clustering C_kIf the traversal is completed, go to step (6.6), otherwise take out the next service S_kAnd (6.5) carrying out the step;

step (6.5) statistical service S_kNumber of input parameters S_kInnum and the number of output parameters S_kOutnum, and accumulating to the in.num and out.num respectively, and returning to the step (6.4);

step (6.6) traverse clustering C_kIf traversal is complete, then S_kcenterJoin list S_centerAnd returning to the step (6.2), otherwise, taking out the next service S_kCarrying out step (6.7);

step (6.7) calculation service S_kIn, num if p1 is greater than a given threshold, p1

The input parameter is considered to have higher importance in the current cluster and added to the input parameter list S of the cluster center service_kcenterInput, performing step (6.8);

step (6.8) calculation service S_kOut. num if p2 is greater than a given threshold, p2

The output parameter is considered to have a higher importance in the current cluster and added to the output parameter list S of the cluster center service_kcenterOutput, return to step (6.6).

In the eighth step, the QoS threshold-based service filtering method includes the steps of:

step (8.1) traverses the services in the candidate service set CandidateServices, if the traversal is completed, the operation is ended, otherwise, the next service S is taken out from the candidate service set CandidateServices_cAnd (4) carrying out the step (8.2);

step (8.2) traversal service S_cIf the traversal is completed, the step (8.1) is returned, otherwise, the slave service S_cThe next QoS attribute QoSPropity is taken out, and the step (8.3) is carried out;

step (8.3), if the QoS constraint QoSProperty specified by the user comprises QoSProperty, the step (8.4) is carried out, otherwise, the step (8.2) is returned;

step (8.4) judging whether the Confidence coefficient Confidence of QoSconstruments to QoSproperty of QoS constraint QoSconstruments is not 0, wherein the Confidence coefficient represents the Confidence of a user to a specified threshold consValue, the value of the coefficient is in the range of [0,1], and 1 represents that a filtering algorithm must strictly adhere to the specified threshold; 0 represents that the filtering algorithm must ignore the threshold, if Confidence is not 0, then go to step (8.5), otherwise return to step (8.2);

step (8.5), if the Type of QoSProperty belongs to benefit Type benefit and the attribute Value of QoSProperty is less than the product of Confidence and consValue, performing step (8.6), otherwise, if the Type belongs to cost Type cost and the attribute Value of QoSProperty is greater than the result of the ConsValue divided by the Confidence, also performing step (8.6), otherwise, returning to step (8.2);

step (8.6) will serve S_cFiltering out and removing from candidate service set CandidateServices, and returning to the step (8.2).

The ninth step comprises the following steps:

step (9.1) of carrying out weight calculation on QoS attribute set by a user, wherein a judgment matrix for pair comparison is constructed by combining the mechanism of an Analytic Hierarchy Process (AHP), wherein each element scores each other element through a predefined score (from 1 to 9) to represent the relative importance of each element; the scale method of 1-9 conforms to the psychological habits of people in making judgments, and table 1 gives the corresponding absolute scale:

TABLE 1

The steps of QoS attribute weight calculation are as follows:

step (9.1.1) determining an index to be used by the AHP method, and performing step (9.1.2);

step (9.1.2) establishing an n-order pairwise comparison matrix A [ n ] [ n ] based on user preference, and performing step (9.1.3);

step (9.1.3) calculates importance weight of each QoS attribute based on the pairwise comparison matrix a, specifically the calculation steps are as follows:

step (9.1.3.1) establishing vector W [ n ] and temporary vector M [ n ] recording importance weight of each QoS attribute and temporary matrix P [ n ] [ n ], and proceeding step (9.1.3.2);

step (9.1.3.2) sets up temporary variable j and initial value 0, proceed step (9.1.3.3);

step (9.1.3.3) if j is less than n, proceeding to step (9.1.3.4), otherwise proceeding to step (9.1.3.5);

step (9.1.3.4) the sum of jth column of the matrix A is assigned to M [ j ], the value of j plus 1 is assigned to j, and the step (9.1.3.3) is returned;

step (9.1.3.5) sets temporary variables k, l and initial value 0, and proceeds step (9.1.3.6);

step (9.1.3.6), if k is less than n, performing step (9.1.3.7), otherwise performing step (9.1.3.9);

step (9.1.3.7) if l is less than n, then step (9.1.3.8) is carried out, otherwise, the value of adding 1 to k is given to k, and the step (9.1.3.6) is returned;

step (9.1.3.8) assigning P [ l ] [ k ] the value obtained by dividing A [ l ] [ k ] by M [ k ], assigning l the value of l plus 1 to l, and returning to step (9.1.3.7);

step (9.1.3.9) sets temporary variable m and initial value 0, and proceeds step (9.1.3.10);

step (9.1.3.10), if m is smaller than n, performing step (9.1.3.11), otherwise performing step (9.1.3.12);

step (9.1.3.11) assigning the sum of the mth row of the matrix P to W [ m ], assigning the value of m plus 1 to m, and returning to step (9.1.3.10);

step (9.1.3.12) sets temporary variable i and initial value 0, and proceeds step (9.1.3.13);

step (9.1.3.13) if i is less than n, then step (5.3.14) is carried out, otherwise, the calculation is finished;

step (9.1.3.14) of dividing W [ i ] by n and then assigning W [ i ], and then assigning i with a value of i plus 1 and returning to step (9.1.3.13);

step (9.2) verifies the consistency of the user judgment according to the pair comparison matrix A constructed in step (9.1) and the weight vector W obtained in step (9.1), in the AHP method, if the corresponding consistency ratio CR is less than 0.1, the user judgment is considered to have satisfactory consistency, otherwise, the user judgment needs to be subjectively corrected until the consistency requirement is met, CR needs to be obtained by dividing the CI value by the average random consistency index RI, and the calculation steps are as follows:

step (9.2.1) of establishing a temporary vector M [ n ]]And a temporary matrix P [ n ]][n]Setting up the variable CI to represent the index of conformity, λ_maxThe element with the largest median value in the vector M is represented, and the step (9.2.2) is carried out;

step (9.2.2) setting temporary variables i, j and setting initial values 0, and performing step (9.2.3);

step (9.2.3), if j is less than n, performing step (9.2.4), otherwise, performing step (9.2.6);

step (9.2.4) if i is less than n, then step (9.2.5) is carried out, otherwise, the value of adding 1 to i is given to i, and then step (9.2.3) is returned;

step (9.2.5) assigning the value obtained by multiplying A [ i ] [ j ] by W [ j ] to P [ i ] [ j ], then assigning the value obtained by adding 1 to j, and returning to the step (9.2.4);

step (9.2.6) sets temporary variable k and initial value 0, and proceeds step (9.2.7);

step (9.2.7), if k is less than n, performing step (9.2.8), otherwise performing step (9.2.9);

step (9.2.8) assigning the sum of the k-th row of the matrix P to M [ k ], assigning the value of k plus 1 to k, and returning to step (9.2.7);

step (9.2.9) sets temporary variable l and initial value 0, and proceeds step (9.2.10);

step (9.2.10), if l is less than n, performing step (9.2.11), otherwise performing step (9.2.12);

step (9.2.11) of assigning M [ l ] to a value obtained by dividing M [ l ] by W [ l ], assigning l to a value obtained by adding 1 to l, and returning to step (9.2.10);

step (9.2.12) assigns the element of the vector M with the largest median to λ_maxCarrying out the step (9.2.13);

step (9.2.13) is based on the formula CI ═ λ_max-n)/(n-1) calculating a consistency index, and proceeding to step (9.2.14);

step (9.2.14) dividing the CI value by the corresponding average random consistency index RI to obtain a consistency ratio CR, if CR is less than 0.1, the user is judged to have satisfactory consistency, otherwise, the user is judged to have no satisfactory consistency;

and (9.3) carrying out comprehensive scoring and ranking on the services, wherein the steps are as follows:

step (9.3.1), if the final result of step (9.2) is that the user is judged to have satisfactory consistency, step (9.3.2) is carried out, otherwise, the result that the QoS weight needs to be revised again by the user is fed back, and the calculation is finished;

step (9.3.2) of establishing a vector of service composite score, namely, compScore [ services.size ], QoS score vector, qossscore [ services.size ], and QoSMtx matrix recording standardized QoS values, wherein services.size represents the number of services of a candidate service set, rows in QoSMtx represent candidate Web services, and columns represent QoS attributes, and step (9.3.3) is carried out;

step (9.3.3) traverses candidate service sets CandidateServices, if the traversal is completed, step (9.3.8) is carried out, otherwise, the next service S of the candidate service sets CandidateServices is taken out_iCarrying out the step (9.3.4);

step (9.3.4) traversal service S_iQoS attribute set S in (1)_iQsproperties, if traversal is complete, proceed to step (9.3.7), otherwise from S_iThe next property S is fetched from the qsProperties_i.P_jPerforming step (9.3.5);

in the step (9.3.5), since the measurement intervals of each QoS attribute value are different, the QoS values need to be normalized to make the value range between 0 and 1, so as to ensure that each QoS attribute is evaluated by using the same span, and therefore, the QoS attribute S needs to be calculated_i.P_jNormalized value of (n) v (p)_ij) And assigns this value to QoSMtx [ i ]][j]The concrete calculation steps are as followsThe following:

step (9.3.5.1) if the attribute S_i.P_jIf the attribute is a profit type attribute, the step (9.3.5.2) is performed, and if the attribute S is the profit type attribute_i.P_jIf the attribute is a cost attribute, the step (9.3.5.3) is performed;

step (9.3.5.2) nv (p)_ij) The calculation formula of (a) is as follows:

wherein, v (p)_ij) Presentation service S_iIn the attribute S_i.P_jIs taken as the original value of v (p)_j)_maxRepresentation of candidate service set CandidateServices and S in each service_i.P_jMaximum values belonging to the same attribute, v (p)_j)_minRepresentation of candidate service set CandidateServices and S in each service_i.P_jA minimum value belonging to the same attribute, and a step (9.3.6) is performed;

step (9.3.5.3) nv (p)_ij) The calculation formula of (a) is as follows:

step (9.3.6) if the attribute S_i.P_j(ii) satisfies the user-defined QoS constraints QoSContriants, then calculates the attribute S_i.P_jQoS score of (q) QoS score of [ i ]]And then, accumulating and recording the QoS scores, wherein a calculation formula in the whole process is as follows:

qosScore[i]+＝(QoSMtx[i][j]×QoSConstraint[j].Weight)

wherein QoSConstratent [ j ]]Weight represents the user vs. attribute S_i.P_jReturning to the step (9.3.4) according to the defined weight, otherwise, directly returning to the step (9.3.4);

step (9.3.7) combines the function similarity calculation method of the fourth step to calculate the current RCT node requirement S_RCTAnd service S_iFunctional similarity functional sim (S) of_i,S_RCT) And combining the steps of (a)9.3.6) the resulting QoS score qosScore [ i]To service S_iThe composite score is weighted and the final result is assigned to the ComScore [ i ]]The calculation formula is as follows:

compScore[i]＝w1*FunctionalSim(S_i,S_RCT)+w2*qosScore[i]

wherein w1 and w2 respectively represent the function similarity weight and the QoS scoring weight, and the step (9.3.3) is returned;

step (9.3.8) ranks the services that completed the composite score and returns the service with the highest score, i.e., the best service.

The invention has the advantages that the improved DBSCAN algorithm is utilized to cluster the services, the distance between samples is calculated by combining the functional similarity between the services in the clustering process, and then the RCT model-based dynamic service matching method is carried out on the basis of service clustering, including the algorithm execution of service filtering, service comprehensive scoring and ranking. The server program can thus finally select the best service that meets the user's needs.

Drawings

Figure 1 shows a service dynamic evolution process diagram.

Fig. 2 shows an RCT model diagram.

Fig. 3 shows an OWL-S extension diagram supporting QoS.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

Referring to fig. 1, a dynamic evolution method of a demand-oriented adaptive Web service is described, where a main flow of the dynamic evolution method of the service is described as follows:

in the system initialization stage, the service system clusters the Web services in the repository to reduce the search space of the services (step (r), corresponding to the first step to the sixth step of the inventive content). Meanwhile, a context-aware Web task (RCT) model facing the user requirements needs to be constructed in advance (step two).

After the initialization work is finished, when a user sends a request, the system can be matched with a corresponding RCT demand node in an RCT model according to the request, and the demand node is connected by a series of Web subtask nodes through a certain structure. The system reads each Web subtask node from the demand node according to a defined flow structure. And when a node is positioned, reading the OWL-S description information of the node to match services meeting the requirements of the node for execution. In the matching process, the functional similarity between the RCT node requirement description and each cluster center service description needs to be calculated respectively, and then the cluster where the cluster center service with the largest functional similarity is located is obtained (step three, corresponding to the seventh step of the invention content). And then, screening the services according to the QoS threshold value in the user request, and filtering the services which do not meet the requirements of the user (step (c), corresponding to the eighth step of the invention content). And finally ranking the candidate services according to the comprehensive score formed by the QoS attribute score and the function similarity, and returning the service with the highest score (step five, corresponding to the ninth step of the invention content), namely the best service.

The RCT model in the above description is an extension and redefinition of the issar framework of the present invention, supporting evolving user requirements, and specification of Web interactions and related contexts. The iSAR framework is a target-oriented requirement modeling method which is proposed by professor Eric YU of Toronto university and the like to understand and redesign a business process according to strategy dependency and principles thereof, and defines four dependencies in the strategy dependency model, namely: target dependency, meaning that one role depends on another role to accomplish a certain goal; resource dependence, meaning that one role depends on another role to provide it with physical resources and information; task dependency, meaning that one role depends on another role to complete a certain task; non-functional requirements are related to soft target dependencies, similar to target dependencies. In particular, the invention extends the atomic concepts of the roles, targets, tasks and resources of the iStar to support the mapping relationships and relevant description specifications among user requirements, Web tasks, services and service sequences, and the RCT model is instantiated by an OWL-S format file to ensure the normal execution of the service matching process. Fig. 2 is an RCT model diagram provided by the present invention, where an oval node represents a user requirement (RCT requirement node), a hexagon node represents a Web subtask, a well number on the upper left of the hexagon node represents a Web subtask sequence number, a circle represents a service, and a rectangle represents an information resource (e.g., order information in a hotel reservation scenario).

The dynamic service evolution method of the present invention is specifically introduced below, and the steps are as follows:

the method comprises the following steps of firstly, calculating semantic similarity between two concepts A and B in a domain ontology, and specifically comprising the following steps:

the functional similarity measure takes into account domain ontology concepts of inputs and outputs between services, and a match between inputs (outputs) mainly refers to a match of concepts related to inputs (outputs). To calculate the similarity of two concepts a and B, the relationship between the two concepts in the domain ontology needs to be considered.

A domain ontology is a specialized ontology describing knowledge of a given domain, where a "domain" is established according to the needs of an ontology builder, and may be a subject domain, a combination of several domains, or a small area within a domain. If two concepts in a domain ontology have different names but the same set of individuals, they are called equivalence classes (equivalent classes).

otherwise, performing the step (1.4);

step (1.4) if the concept A and the concept B have no parent-child relationship, but the two concepts directly or indirectly have a common parent class concept C, a word semantic similarity measurement method based on a naive Bayes model is adopted, firstly, all attributes of the concept A and the concept B are respectively traversed, the attribute names of the concept A and the concept B are subjected to feature extraction through a computeFeature function, and then, a conditional probability distribution list and an adjustment factor after sample training are adopted to calculate the similarity Sim between the concept attributes_wordAccording to the similarity Sim between the concept attributes_wordComparing with similarity judgment factor eta, judging whether two attributes are the same attribute, counting, and finally calculating similarity Sim of concepts A and B_conceptThe method comprises the following steps:

step (1.4.4) is based on naive Bayes model, combines WorkNet English dictionary, and pairs of prop (A) through ComputeFeature function_jAnd prop (B)_kThe attribute name of (2) is subjected to feature extraction to obtain L (prop (A)_j,prop(B)_k) And D (prop (A)_j,prop(B)_k). The naive Bayes model is one of the two most extensive classification models, and WorkNet is an English dictionary established and maintained by the university of Princeton, cognition, science and laboratory under the guidance of Miller in the teaching of psychology. The specific calculation process is as follows:

calculating the word semantics of each attribute name, since each word corresponds to oneMultiple semantics, each word pair may correspond to one or more semantic pairs. Defining the semantic node distance with the shortest distance in all semantic pairs corresponding to the word pair as the word pair distance L (prop (A)_j,prop(B)_k) And defining the semantic pair depth with the shortest distance between semantic nodes as word pair depth D (prop (A)_j,prop(B)_k) Known attribute names prop (A)_jPresence in semantic node v_j1,v_j2,…，v_jnIn the synonymous phrase of (1), attribute name prop (B)_kPresence in semantic node v_k1,v_k2，…v_kmIn a synonymous phrase of (1), then prop (A)_jAnd prop (B)_kThe distance calculation formula and the depth calculation formula are as follows:

wherein L (v)_ja,v_kb) Representing semantic nodes v_jaAnd semantic node v_kbDistance of D (v)_ja,v_kb) Representing semantic pairs (v)_ja,v_kb) Of the depth of (c).

Further, mean functions LW (i) and DW (o) are generated according to a training set of the naive Bayes model, and then a conditional probability distribution column P (L (prop (A)) is calculated by utilizing the mean functions LW (i) and DW (o)_j,prop(B)_k) I C and P (D (prop (A))_j,prop(B)_k) I C), wherein C is a word class classification with a value range of { U, N }, wherein U represents "consistent" and N represents "inconsistent". Finally, adjusting factors alpha and beta are calculated, and the calculation formula is as follows:

then, the step (1.4.5) is carried out;

step (1.4.5) based on naive Bayes model, the characteristic L (prop (A) of ontology concept attribute is processed_j,prop(B)_k) And D (prop (A)_j,prop(B)_k) Injecting the mixture into the conditional probability distribution list obtained in the step (1.4.4), and sequentially extracting the conditional probability V₁＝P(L(prop(A)_j,prop(B)_k)＝i|C＝U)，V₂＝P(D(prop(A)_j,prop(B)_k)＝o|C＝U),V₃＝P(L(prop(A)_j,prop(B)_k) I | C ═ N) and V₄＝P(D(prop(A)_j,prop(B)_k) O | C ═ N). Finally, the pro (A) is calculated by combining the regulatory factors alpha and beta in step (1.4.4)_jAnd prop (B)_kSimilarity between them Sim_wordThe calculation formula is as follows:

wherein i represents the number of attributes in concept A and concept B, prop (A) and prop (B) represent the attribute sets of concept A and concept B, respectively, and Size (prop (B)) and Size (prop (A)) represent the number of attributes in concept B and concept A, respectively, and the step (1.5) is performed;

step (1.5) if the relationship between the concept A and the concept B does not meet the above condition, the similarity Sim of the concepts A and B_conceptSet to 0;

second step, concept of combining the first stepSimilarity calculation method, providing service S₁And service S₂Input similarity Sim_inputThe specific process of the calculation method is as follows:

step (2.4) traverse S_longIf the traversal is completed, go to step (2.8), otherwise go from S_longIn the next input parameter long is taken out_iCarrying out the step (2.5);

wherein Size (S)_long.Input and Size: (S_shortInput) represents the service S, respectively_longNumber of input parameters and service S_shortThe number of input parameters, | d | represents the difference value of the number of the input parameters of two services, and InSim is the maximum matching array of the similarity of the input parameters;

step three, combining the concept similarity calculation method of the step one, and providing a service S₁And service S₂Output similarity Sim_outputThe specific process of the calculation method is as follows:

step (3.3) if d is less than or equal to 0, service S is carried out₁Is set to S_shortService S₂Is set to S_longOtherwise, it will serve S₂Is set to S_shortService S₁Is set to S_longAnd (4) performing the step (3.4);

wherein Size (S)_longOutput) and Size (S)_shortOutput) respectively represent the services S_longNumber of output parameters and service S_shortThe number of output parameters, | d | represents the difference value of the number of the output parameters of the two services, and OutSim is a maximum matching array of the similarity of the output parameters;

FunctionalSim(S₁,S₂)＝w₁×Sim_input+w₂×Sim_Outputwherein the weight w₁And w₂Is a real number between 0 and 1 and sums to 1. They represent the importance of the service consumer to input similarity and output similarity determinations. By default, w₁And w₂Are all set to 0.5.

And fifthly, giving a service clustering algorithm by combining the service function similarity measurement mode in the fourth step, firstly, arbitrarily selecting an unallocated core object as a seed, and then searching all sample sets with reachable densities of the core object to form a cluster. And then continuously selecting another unallocated core object to search a sample set with a reachable density to obtain another cluster until all the core objects are allocated. Formula of search radius Eps in the invention

Determining, wherein n represents the dimension size of the service, i.e. the average attribute number of all services, k represents the minimum number of services required to form a cluster, r represents the radius of a hypersphere composed of n-dimensional services, and m is the n-dimensional radius contained in the hypersphereThe number of services; while the minimum number MinPts follows empirical heuristics: MinPts is round (m/25), where round is the rounding function and m is the number of samples in the sample set.

step (5.2) traversing the original sample set T, if the traversal is finished, ending, otherwise, taking out the next service S from the T_iAnd removing it from T, proceeding to step (5.3);

further, step (5.3) said finding service S_iThe Eps neighborhood of (c) is as follows:

step (5.3.4) if service S_iAnd service S_jFunctional similarity functional sim (S)_i,S_j) Less than or equal to the search radius Eps, the service S will be provided_jJoining service S_iNeighborhood set N of_Eps(S_i) Returning to the step (5.3.2), otherwise, directly returning to the step (5.3.2);

step (5.11) if service S'_kIf the number of Eps neighborhood samples is not less than MinPts, the service in the neighborhood is added to the service S_iEps neighborhood set N_Eps(S_i) And (5.12) performing the step (5.12), otherwise, returning to the step (5.5);

step (5.13) clustering the clusters C_kAdding the cluster set C, and returning to the step (5.2);

and sixthly, obtaining the central service of each cluster according to the clustering result of the fifth step. The cluster center service can represent the typical characteristics of the cluster service, and in general, the cluster center service is not a specific existing service but a virtual service constructed by a certain algorithm. The specific calculation process of the cluster center service is as follows:

step (6.3) of establishing a variable in.num and out.num pair cluster C_kAll the input parameters and the output parameters are counted, initial values of the two variables are set to be 0, and a cluster center service S is established_kcenterAnd (6.4) carrying out the step;

The output parameter is considered to have a higher importance in the current cluster and added to the output parameter list S of the cluster center service_kcenterOutput, return to step (6.6);

and seventhly, in the service matching process of the system, the functional similarity between the RCT node requirement description and each cluster center service (obtained in the sixth step) description needs to be calculated respectively. The RCT node is a constituent unit of the RCT model, and can be described as a Web service required for completing a current Web subtask, and the RCT model is an extension and redefinition of the itar framework proposed by professor Eric YU of toronto-the-world of the present invention, and the itar framework defines four dependencies in the policy dependency model, which are: target dependency, meaning that one role depends on another role to accomplish a certain goal; resource dependence, meaning that one role depends on another role to provide it with physical resources and information; task dependency, meaning that one role depends on another role to complete a certain task; non-functional requirements are related to soft target dependencies, similar to target dependencies. In particular, the present invention extends the atomic concept of roles, goals, tasks, and resources of iStar to support mapping relationships and related description specifications between user requirements, Web tasks, services, and service sequences. The RCT model will be instantiated in an OWL-S format file to ensure normal execution of the service matching process. In the RCT model, an ellipse is used to represent a user demand (RCT demand node), a hexagon is used to represent a Web subtask, a well number above and to the left of the hexagon represents a sequence number of the Web subtask, a circle represents a service, and a rectangle represents an information resource.

Because OWL-S is a language designed for describing Web services and is suitable for describing the requirements of services, the requirements of RCT nodes are also described by adopting OWL-S, the functional similarity measurement method in the fourth step is also suitable for calculating the similarity between the RCT nodes and cluster center services, the cluster where the cluster center services with the maximum functional similarity are located is obtained through calculation, the services in the cluster are candidate services CandidateServices which can meet the requirements of current users, and the calculation step is the same as the fourth step, so that repeated description is not needed.

And eighthly, screening candidate service sets CandidateServices (obtained in the seventh step) according to the QoS threshold in the user request, and filtering out services which do not meet the requirements of the user. The service consumer may specify a QoS threshold in the service request indicating rejection of Web services having a QoS below or above the threshold.

Before introducing the QoS threshold based service filtering method, the ontology based OWL-S extension proposed by the present invention is introduced, i.e. non-functional description (QoS) is added to the Web service description.

OWL-S (original DAML-S) is called as Ontology Web Services for Services, is a Web service Ontology description Language based on OWL (Web Ontology Web), semantically describes Web Services by using a whole set of description mechanism, and generates a description file which can be understood and processed by a computer.

OWL-S includes three basic components:

(4) the Service Profile describes what the Service does, and the basic attributes comprise a Service name (serviceName), a text description (textDescription) and the like; the functional attributes include four parts of input information (hasInput), output information (hasOutput), preconditions (hasRecondition) and expected effects (hasEffect).

(5) Service Model describes how services work, OWL-S treats each Service as a Process (Process) and divides it into three categories: atomic processes (Atomic processes), Simple processes (Simple processes), and Composite processes (Composite processes). The atomic process has no subprocess and can be directly called; the simple process can not be directly called, and a corresponding grouping description file does not exist; the combined process is formed by combining a plurality of sub-processes through a certain control structure, and in OWL-S, each process body can be semantically described through IOPE.

(6) Service grouping describes the calling information of the Service, OWL-S uses WSDL as the basis of the Service grouping mechanism, defines a specific message format, and encapsulates input information, output information, Service addresses and the like.

Based on Web service description language OWL-S, the invention constructs a semantic service description model as shown in the following.

The invention adopts quadruplets to express semantic service description: WS ═ N, D, I, O >, where:

n represents the Name (Name) of the service, D represents the text Description (Description) of the service, and the N and the D form the basic attribute of the service;

the functional attributes of the service include I and O, which represent the attributes of the Input (Input) and Output (Output) interface parameters of the service WS, respectively, each interface is a set of parameter sets, and I ═ I₁,…,i_m}, element i₁,…,i_mRepresenting semantic concepts corresponding to input parameters of the service WS in the domain ontology; o is the output set of the service, O ═ O₁,…,o_n}, element o₁,…,o_nAnd representing semantic concepts corresponding to the output parameters of the service WS in the domain ontology.

Fig. 3 shows a new Service Profile model. In the figure, a dotted arrow (subcolasiof) represents an inheritance relationship between a subclass and a parent class, a solid arrow (hasProperty) represents a membership relationship of an attribute, and in OWL-S Service Profile, the invention uses a group of Service parameters (ServiceParameter) comprising a name (serviceParameteName) and a parameter value (Sparamer), wherein Thing represents the specific content of the parameter value (Sparamer) & xsd; # string indicates the attribute of the type string, & process; # Input and & process; # Output indicates the input information content and the Output information content of the service, respectively. After the OWL-S and the QoS body are connected, QoSProperty becomes a subclass of the OWL-S ServiceParameter, and qoPropertyName and qProperty are child attributes of the OWL-S serviceParameterName and the sParameter attribute respectively.

Each QoS attribute (QoSProperty) is defined by a name in the form of a string (qoospropertyname) and a set of characteristics (qoschharacteristic), described as follows:

value: a value representing a QoS attribute. For a service provider, it represents a QoS attribute of the service provided; but for the service consumer it indicates the threshold for the QoS attribute.

Type: this feature is used to distinguish between two types of QoS:

revenue-type benefit (e.g., reliability, quality improves as attribute values increase). In this case, the user-specified QoS attribute value represents the minimum value to be considered.

Cost type costs (e.g., response time, quality decreases as the attribute value increases). In this case, the user-specified QoS attribute value represents the maximum value to be considered.

Unit: the metric unit (e.g., meta, sec) of the QoS attribute.

Weight: QoS weights are used to specify the importance of QoS attributes.

Confidence: representing the confidence level of the user's preference for it.

Where Weight and Confidence are two attributes of the qoschharacteristic subclass RequesterQoS, RequesterQoS representing QoS settings of a service requester and qoschharacteristic another subclass ProviderQoS representing QoS settings of a service provider.

Next, a QoS threshold based service filtering method will be described, which includes the following steps:

and (8.4) judging whether the Confidence coefficient Confidence of the QoSContractions to QoSproperty is not 0 or not. The confidence coefficient represents the user's confidence in the designated threshold consValue, and should be in the range of 0, 1.1 indicates that the filtering algorithm must strictly adhere to the specified threshold; 0 represents that the filtering algorithm must ignore this threshold. If the Confidence is not 0, performing the step (8.5), otherwise, returning to the step (8.2);

step (8.6) will serve S_cFiltering and removing from candidate service sets CandidateServices, and returning to the step (8.2);

and ninthly, performing comprehensive service scoring and ranking on the candidate service sets CandidateServices filtered in the eighth step, wherein the service with the highest score is the best service. Before service scoring, consistency verification needs to be performed on the QoS weight determined by the user, if the obtained consistency ratio CR is smaller than 0.1, scoring and ranking steps are performed, otherwise, the QoS weight is fed back, the result of revising the user is required, and the calculation is finished, specifically, the following steps are performed:

and (9.1) performing weight calculation on the QoS attribute set by the user, wherein the Analytic Hierarchy Process (AHP) mechanism is combined, and the analytic hierarchy process is a decision-making method which decomposes elements always related to decision into a target, a standard, a scheme and the like and performs qualitative and quantitative analysis on the basis. A decision matrix is constructed for pairwise comparison, where each element scores every other element by a predefined score (from 1 to 9), indicating their relative importance. The scale method of 1-9 conforms to the psychological habit of people in making judgment, and the corresponding absolute scale is given in table 1.

TABLE 1

Further, the QoS attribute weight calculation steps are as follows:

step (9.1.1) determining an index (specified by a user) to be used by the AHP method, and performing step (9.1.2);

and (9.2) verifying the consistency of the judgment of the user according to the pair of comparison matrixes A constructed in the step (9.1) and the weight vector W obtained in the step (9.1). In the AHP method, if the corresponding consistency ratio CR is less than 0.1, it is considered that the user has a satisfactory consistency, otherwise, the user needs to subjectively correct until the consistency requirement is satisfied. CR is obtained by dividing the CI value by the average random consistency index RI, and the calculation steps are as follows:

step (9.2.13) is based on the formula CI ═ λ_maxCalculating a consistency index of-n)/(n-1), and performingA step (9.2.14);

to further understand step (9.1) and step (9.2), step (9.1) and step (9.2) are described herein with reference to the following example:

(1) an index to be used by the method is determined. As an example, the present invention selects four QoS attributes as indicators, namely: response time, execution price, reliability, and availability.

(2) A pairwise comparison matrix based on user preferences is established. Since there are four QoS attributes in the example, the matrix will contain 16 elements, and the matrix described in Table 2 represents the corresponding decisions of the user in the pairwise comparison.

TABLE 2

(3) And (4) calculating each QoS attribute weight in combination with the calculating step of the step (9.1). The RI values corresponding to different values of n are given in table 3, where n is the number of QoS attributes.

TABLE 3

In this example, n is 4, so the calculation step according to step (9.1) can derive each QoS attribute weight (as in table 4).

TABLE 4

(4) And finally, verifying the consistency of the judgment of the user by combining the calculation step of the step (9.2). The following results were calculated: lambda [ alpha ]_max＝4.203，CI＝0.068，CR ═ 0.075. The CR value is less than 0.1 and thus the weight is accepted and the user judges to have satisfactory consistency.

And (9.3) carrying out comprehensive scoring and ranking on the services, which comprises the following specific steps:

step (9.3.2) establishes a vector of service composite scores, comp [ services.size ], QoS score vector, QoS score [ services.size ], and QoSMtx matrix recording normalized QoS values, where services.size represents the number of services in the candidate service set, rows in QoSMtx represent candidate Web services, and columns represent QoS attributes, and step (9.3.3) is performed.

step (9.3.5) because the measurement interval of each QoS attribute value is different, the QoS values need to be normalized to have a value range between 0 and 1, so as to ensure that each QoS attribute is evaluated with the same span. It is therefore necessary to calculate the QoS attribute S_i.P_jNormalized value of (n) v (p)_ij) And assigns this value to QoSMtx [ i ]][j]The specific calculation steps are as follows:

step (9.3.5.2) nv (p)_ij) The calculation formula of (a) is as follows:

step (9.3.5.3) nv (p)_ij) The calculation formula of (a) is as follows:

qosScore[i]+＝(QoSMtx[i][j]×QoSConstraint[j].Weight)

step (9.3.7) combines the function similarity calculation method of the fourth step to calculate the current RCT node requirement S_RCTAnd service S_iFunctional similarity functional sim (S) of_i,S_RCT) And combining the QoS score qOSScore [ i ] obtained in the step (9.3.6)]To service S_iThe composite score is weighted and the final result is assigned to the ComScore [ i ]]The calculation formula is as follows:

compScore[i]＝w1*FunctionalSim(Si,SRCT)+w2*qosScore[i]

Claims

1. A dynamic evolution method of demand-oriented adaptive Web services is characterized by comprising the following steps:

step five, a service clustering algorithm is given by combining the service function similarity measurement mode of the step four, firstly, an unallocated core object is selected arbitrarily as a seed, and then all sample sets with reachable density of the core object are searched to form a cluster; then, another unallocated core object is continuously selected to search for a sample set with a reachable density, and another sample set is obtainedClustering until all core objects are distributed; search radius Eps is given by

Determining, wherein n represents the dimension size of the service, namely the average attribute number of all the services, k represents the minimum number of services required for forming a cluster, r represents the radius of a hypersphere composed of n-dimensional services, and m is the number of n-dimensional services contained in the hypersphere; while the minimum number MinPts follows empirical heuristics: MinPts is round (m/25), where round is the rounding function and m is the number of samples in the sample set;

ontology-based OWL-S extension, namely adding non-functional description QoS into Web service description; OWL-S is called totally on-load Web Language for Services, it is Web service Ontology description Language based on OWL, use a complete set of description mechanism to carry on the semantic description to the Web service, and the description file that is produced can be understood and processed by the computer;

OWL-S includes three basic components:

(1) the Service Profile describes what the Service does, and the basic attributes comprise a Service name serviceName and a text description; the functional attributes comprise four parts of input information hasInput, output information hasOutput, precondition hasRecondition and expected effect hasEffect;

(2) service Model describes how services work, OWL-S treats each Service as a Process, and divides it into three categories: atomic Process Atomic, Simple, and Composite; the atomic process has no subprocess and can be directly called; the simple process can not be directly called, and a corresponding Service group description file does not exist; the combined process is formed by combining a plurality of sub-processes through a certain control structure, and in OWL-S, each process body can be semantically described through IOPE;

the function attribute of the service comprises I and O, which respectively represent the attributes of the Input and output interface parameters of the service WS, each interface is a set of parameter sets, and I ═ { I ═ is₁,…,i_m}, element i₁,…,i_mRepresenting semantic concepts corresponding to input parameters of the service WS in the domain ontology; o is the output set of the service, O ═ O₁,…,o_n}, element o₁,…,o_nRepresenting semantic concepts corresponding to the output parameters of the service WS in the domain ontology;

in the OWL-S Service Profile, using a group of Service parameters, namely, ServiceParameter, including a name ServiceParameterName and a parameter value, Sparamer, connecting an OWL-S and a QoS body, so that QoProperty becomes a subclass of the OWL-S Service Parameter, and qoPropertyName and qProperty are respectively sub-attributes of the OWL-S serviceParameterName and Sparamer attributes;

type: this feature is used to distinguish between two types of QoS:

cost type cost, in which case the user specified QoS attribute value represents the maximum value to be considered; unit: a unit of measure of the QoS attribute;

weight: a QoS weight for specifying the importance of the QoS attribute;

confidence: representing a confidence level of the user's preference for it;

2. The dynamic evolution method of demand-oriented adaptive Web services according to claim 1, wherein in the first step, the step of calculating the semantic similarity between two concepts a and B in the domain ontology is as follows:

step (1.1) if the concepts A and B are identical or they are declared equivalent classes, the similarity Sim of the concepts A, B_concept1, otherwise, performing the step (1.2);

otherwise, performing the step (1.4);

step (1.4) if the concept A and the concept B have no parent-child relationship, but the two concepts have a common parent class directly or indirectlyAnd the concept C adopts a word semantic similarity measurement method based on a naive Bayes model, firstly, the attributes of the concept A and the concept B are respectively traversed, the attribute names of the concept A and the concept B are subjected to feature extraction through a computeFeature function, and then the similarity Sim between the concept attributes is calculated by adopting a conditional probability distribution list and an adjustment factor after sample training_wordAccording to the similarity Sim between the concept attributes_wordComparing with similarity judgment factor eta, judging whether two attributes are the same attribute, counting, and finally calculating similarity Sim of concepts A and B_concept；

3. The dynamic evolution method of demand-oriented adaptive Web services according to claim 2, characterized in that the procedure of step (1.4) is as follows:

calculating the word semantics of each attribute name, wherein each word corresponds to one or more semantics, and each word pair corresponds to one or more semantic pairs, and the word semantics are compared with the semantic pairsThe semantic node distance with the shortest distance among all semantic pairs corresponding to a word pair is defined as the word pair distance L (prop (A)_j,prop(B)_k) And defining the semantic pair depth with the shortest distance between semantic nodes as word pair depth D (prop (A)_j,prop(B)_k) Known attribute names prop (A)_jPresence in semantic node v_j1,v_j2,…，v_jnIn the synonymous phrase of (1), attribute name prop (B)_kPresence in semantic node v_k1,v_k2，…v_kmIn a synonymous phrase of (1), then prop (A)_jAnd prop (B)_kThe distance calculation formula and the depth calculation formula are as follows:

then, the step (1.4.5) is carried out;

step (1.4.5) based on naive Bayes model, the characteristic L (prop (A) of ontology concept attribute is processed_j,prop(B)_k) And D (prop (A)_j,prop(B)_k) Injecting the mixture into the conditional probability distribution list obtained in the step (1.4.4), and sequentially extracting the conditional probability V₁＝P(L(prop(A)_j,prop(B)_k)＝i|C＝U)，V₂＝P(D(prop(A)_j,prop(B)_k)＝o|C＝U),V₃＝P(L(prop(A)_j,prop(B)_k) I | C ═ N) and V₄＝P(D(prop(A)_j,prop(B)_k) O | C ═ N), finally combining the adjustment factors α and β in step (1.4.4) to calculate prop (a)_jAnd prop (B)_kSimilarity between them Sim_wordThe calculation formula is as follows:

4. The dynamic evolution method of demand-oriented adaptive Web services according to any of claims 1 to 3, characterized in that in said second step, service S₁And service S₂Input phaseSimilarity Sim_inputThe process of the calculation method of (1) is as follows:

wherein Size (S)_long.Input) and Size (S)_short.Input) represents services S, respectively_longNumber of input parameters and servicesS_shortThe number of input parameters, | d | represents the difference between the number of input parameters of two services, and InSim is the maximum matching array of the similarity of the input parameters.

5. Requirement-oriented adaptive Web service dynamic evolution method according to any of claims 1 to 3, characterized in that in said third step, service S₁And service S₂Output similarity Sim_outputThe process of the calculation method of (1) is as follows:

step (3.8) calculation service S₁And service S₂Similar outputDegree Sim_OutputThe calculation formula is as follows:

wherein Size (S)_long.Output) and Size (S)_short.Output) respectively represent services S_longNumber of output parameters and service S_shortThe number of output parameters, | d | represents the difference value of the number of output parameters of two services, and OutSim is the maximum matching array of the similarity of the output parameters.

6. The dynamic evolution method of demand-oriented adaptive Web services according to any of claims 1 to 3, characterized in that in the fifth step, the process of the service clustering algorithm is as follows:

7. The dynamic evolution method of demand-oriented adaptive Web services as claimed in claim 6, characterized in that in said step (5.3), said search service S_iThe Eps neighborhood of (c) is as follows:

step (5.3.2) traverses the original sample set T, if the traversal is completed, the step (5.4) is carried out, otherwise, the next service S is taken out from the T_jTo proceed withStep (5.3.3);

8. The dynamic evolution method of demand-oriented adaptive Web services according to any one of claims 1 to 3, characterized in that in the sixth step, the calculation process of the cluster center service is as follows:

9. The dynamic evolution method of demand-oriented adaptive Web services according to any of claims 1 to 3, characterized in that in the eighth step, the QoS threshold based service filtering method comprises the following steps:

10. The dynamic evolution method of demand-oriented adaptive Web services according to any of claims 1 to 3, characterized in that the ninth step comprises the following steps:

step (9.1) weight calculation is carried out on the QoS attribute set by the user, here, a judgment matrix for pair comparison is constructed by combining the mechanism of an Analytic Hierarchy Process (AHP), wherein each element scores each other element through a predefined score, and the predefined score is from 1 to 9 and represents the relative importance of the elements; the scale method of 1-9 conforms to the psychological habits of people in making judgments, and table 1 gives the corresponding absolute scale:

TABLE 1

The steps of QoS attribute weight calculation are as follows:

step (9.2.13) is based on the formula CI ═ λ_maxCalculating a consistency index of-n)/(n-1), and carrying out the stepsA step (9.2.14);

in the step (9.3.5), since the measurement intervals of each QoS attribute value are different, the QoS values need to be normalized to make the value range between 0 and 1, so as to ensure that each QoS attribute is evaluated by using the same span, and therefore, the QoS attribute S needs to be calculated_i.P_jNormalized value of (n) v (p)_ij) And assigns this value to QoSMtx [ i ]][j]The specific calculation steps are as follows:

and (9).3.5.2)nv(p_ij) The calculation formula of (a) is as follows:

wherein, v (p)_ij) Presentation service S_iIn the attribute S_i.P_jV is the original value of_max(p_j) Representation of candidate service set CandidateServices and S in each service_i.P_jMaximum values, v, belonging to the same attribute_min(p_j) Representation of candidate service set CandidateServices and S in each service_i.P_jA minimum value belonging to the same attribute, and a step (9.3.6) is performed;

step (9.3.5.3) nv (p)_ij) The calculation formula of (a) is as follows:

performing step (9.3.6);

qosScore[i]+＝(QoSMtx[i][j]×QoSConstraint[j].Weight)

compScore[i]＝w1*FunctionalSim(S_i,S_RCT)+w2*qosScore[i]