CN105930443A - Goal-oriented RESTful Web service discovery method - Google Patents

Abstract

The invention discloses a goal-oriented RESTful Web service discovery method. The method comprises the steps of firstly collecting information (particularly functional text description) of RESTful Web services, and performing domain division on Web service sets; secondly constructing a domain knowledge base in the Web service set of each domain, wherein the domain knowledge base comprises a Web service-service goal assignment matrix, service goal clusters and the like; and finally based on the constructed domain knowledge base, recommending service goals with semantic similarity for a given user query, and then matching a service goal set selected by a user with a service goal set of the Web services to obtain a candidate Web service set. According to the method, the RESTful Web services meeting a user requirement goal can be accurately discovered; the method has very high practicality; and by recommending the service goals with the semantic similarity for an initial query of the user, the user can be assisted in making a high-quality query meeting the user demand.

Description

A kind of object-oriented RESTful Web service finds method

Technical field

The invention belongs to service computing technique field, take particularly to a kind of object-oriented RESTful Web Business discovery method.

Background technology

Web service is basic as service-oriented computing (Service-Oriented Computing, SOC) Component, is to encapsulate self-contained, the self-described of specific calculation or business function, the software module of platform independence, Can issue on the internet and call.Utilize existing Web service resource, the effect of software development can be improved Rate and quality, reduce development cost [document 1] simultaneously.Along with the fast development of SOC, service-oriented software Exploitation (Service-Oriented Software Development, SOSD) is increasingly becoming on the Internet soft The main flow of part exploitation, has been widely used in many fields, such as ecommerce, Workflow Management etc. [document 2]. Under this situation, the Web service resource of interconnection Web realease presents the trend of quickly growth.

At present, on the Internet, announced Web service can be largely classified into two classes: follows simple object access association The Web service of view (Simple Object Access Protocol, SOAP) (is called for short SOAP-based Web Service) and follow sign state transfer (REpresentational State Transfer, REST) agreement RESTful Web service.SOAP-based Web service must use the WSDL of standard (Web Services Description Language, WSDL) is described, to the exploitation of Web service Person causes certain restriction.Although RESTful Web service also possesses Web application describes language (Web Application Description Language, WADL), the structurized description language such as WSDL 2.0, But most of developers tend to the RESTful Web service using simple natural language text to be developed it It is described [document 7].In recent years, RESTful Web service is favored by more and more developers, its Growth trend also becomes apparent from [document 8] than traditional SOAP-based Web service.Such as, by the end of 2016 On March 23, in, famous Web service programming website ProgrammableWeb (http://www.programmableweb.com/ is called for short PWeb) Web service of upper registration reaches 14,836, wherein, the ratio of RESTful Web service is about 62%, the ratio of SOAP-based Web service Example is about 16%.

Web service finds one of key support technology as SOC, it is intended to help user from numerous Web Service Source is excavated the Web service that disclosure satisfy that its demand, promotes reusing of Web service.Although it is existing big The Web service of amount finds that method (such as [document 3], [document 4], [document 5], [document 6]) is suggested, But current Web service finds method and still there is the following problem:

(1) existing Web service finds that method is mainly for the SOAP-based Web clothes using WSDL to describe Business, the RESTful Web service to mainly describing with natural language text is paid close attention to less.

(2) when carrying out Web service and finding, user would generally use and can accurately express the high-level of its demand Target, such as " planning stroke (plan a trip) ", " searching hotel (find hotels) ", as inquiry Condition.But, the Web service registration center (such as PWeb) of current main flow still uses based on keyword The Web service discovery mechanism joined, performance is the highest, it is difficult to meet the demand of user.

(3) in addition to the matching mechanisms between user's inquiry and Web service, the quality of user's inquiry is also impact Web service finds the key factor of result.One inquiry that can accurately reflect user's request contributes to obtaining more Relevant Web service.But, for most of users, owing to lacking and desired Web service function Relevant knowledge, is difficult to formulate high-quality inquiry.Such as, due to Web service developer/supplier's statement The difference of mode, the Web service function that can there is semantic similitude describes, such as " get hotels ", " find Hotels ", " search accommodations " etc..When formulating inquiry, user is difficult to consider comprehensively To the function of these semantic similitude, and then cause omitting many Web services that disclosure satisfy that its demand.The most right The concern of this problem is less.

[document 1] M.Bano, D.Zowghi, N.Ikram, et al.What makes service oriented requirements engineering challenging？a qualitative study.IET Software, 2014,8(4),pp.154-160.

[document 2] L.Chen, L.Hu, Z.Zheng, et al.WTCluster:Utilizing Tags for Web Services Clustering.International Conference on Service-Oriented Computing,2011,pp.204-218.

[document 3] P.Plebani and B.Pernici.URBE:Web Service Retrieval Based on Similarity Evaluation.IEEE Transactions on Knowledge and Data Engineering,2009,21(11),pp.1629-1642.

[document 4] F.Liu, Y.Shi, J.Yu, et al.Measuring Similarity of Web Services Based on WSDL.IEEE International Conference on Web Services,2011, pp.155-162.

[document 5] G.Cassar, P.Barnaghi, and K.Moessner, Probabilistic Matchmaking Methods for Automated Service Discovery.IEEE Transactions on Services Computing,2013,7(4),pp.654-666.

[document 6] Z.Cong, A.Fernandez, H.Billhardt, et al.ServiceDiscovery Acceleration with Hierarchical Clustering.Information Systems Frontiers, 2014,17(4),pp.799-808.

[document 7] W.Jiang, D.Lee, and S.Hu.Large-Scale Longitudinal Analysis of SOAP-Based and RESTful Web Services.IEEE 19th International Conference on Web Services,2012,pp.218-225.

[document 8] M.Maleshkova, C.Pedrinaci, and J.Domingue.Investigating Web APIs on the World Wide Web.European Conference on Web Services,2010, pp.107-114.

Summary of the invention

In order to solve above-mentioned technical problem, the invention provides a kind of object-oriented RESTful Web service Discovery method, it is possible to find to meet the RESTful Web service of user's request target exactly, and have very well Practicality.

The technical solution adopted in the present invention is: 1. an object-oriented RESTful Web service discovery side Method, it is characterised in that comprise the following steps:

Step 1: collect the information of RESTful Web service, including Web service title, art and merit The text of energy property describes, and obtains Web service collection；

Step 2: Web service collection is carried out pretreatment；

Step 3: for pretreated Web service collection, it may be judged whether there is field and divide (i.e. Web service Whether comprise the information of art)；

The most then perform following step 4；

If it is not, then perform following step 5；

Step 4: judge field divides whether there is overlap；

The most then perform following step 5；

If it is not, then perform following step 6；

Step 5: use Web service sorting technique that Web service collection is carried out field division；

The basic process of Web service sorting technique is as follows: first, it is intended that need the field carrying out classifying (permissible It is that existing field or the field of user's setting are concentrated in Web service), Web service collection is divided into appointment neck The unrelated Web service collection of Web service collection that territory is relevant and field；Then, relevant from field respectively Web The unrelated Web service of services set and field is concentrated and is chosen the same number of Web service structure training set, except training Web service outside collection constitutes test set；Then, vector is built for the Web service in training set and test set empty Between model, specifically, each Web service is expressed as a vector, the dimension of vector is all Web The sum of the different words that services package contains, vector one word of every one-dimensional representation, each word each Web service to Weights in amount use TF-IDF (Term Frequency Inverse Document Frequency) meter Calculate；Finally, use the support vector machine (Support Vector Machine, the SVM) vector to building empty Between classify, obtain the Web service collection of designated field.

Step 6: carry out the structure of domain knowledge base for the Web service collection in each field, including: domain term Converge sequencing table (Ranked Domain Word List, RDWL), field verb collection (Domain Verb Set, DV), field core name word set (Domain Core Noun Set, DCN), Web service-service goal association Matrix (Web Service-Service Goal Assignment Matrix, WSSGAM) and service goal bunch Collection (Service Goal Clusters, SGC)；

Step 7: given user is inquired about q and carries out Web service discovery.

As preferably, described in step 2, Web service collection is carried out pretreatment, including participle, lemmatization, Remove stop words and word frequency statistics, implement and include following sub-step:

Step 2.1: resolve the information of each Web service, obtains all words wherein comprised；

Step 2.2: all words are reduced into its basic original shape, such as " retrieves ", " retrieved " The basic prototype of " retrieving " is " retrieve "；

Step 2.3: remove some words without practical significance according to disabling vocabulary, as " can ", " it ", " the " etc.；

Step 2.4: add up each word frequency in Web service.

As preferably, in step 6, first to each field, add up the pretreated Web service in this field collection In all words of comprising and each word frequency of occurrence in this field (the most each word is at all Web in this field Frequency of occurrence sum in service)；Then, carrying out domain knowledge base structure for specific field d, it is concrete Realization includes following sub-step:

Step 6.1: calculate each word importance to this field in d, and according to importance to words all in d Descending, obtains the Field Words sequencing table of d, is designated as RDWL_d；

Step 6.2: select RDWL_d150 words of top in noun, constitute d field core noun Collection, is designated as DCN_d；

Step 6.3: the text of each Web service from d carries out service goal extraction in describing, and obtains d's Web service-service goal incidence matrix, is designated as WSSGAM_d；

Service goal (Service Goal) is for representing the function that Web service can be provided by.One service mesh Mark sg is expressed as: sg=＜ sgv, sgn, sgp ＞, and wherein, sgv is verb or verb phrase, represents sg Operation to be performed；Sgn is noun or noun phrase, represents the operation object of sg；Sgp is optional ginseng Manifold, for sg is remarked additionally, the mode of operation of such as sg and constraint etc..

The basic process carrying out service goal extraction from the text of Web service describes is as follows: first, obtains literary composition The statement set that this description comprises；Then, Stanford Parser is utilized (http://nlp.stanford.edu/software/lex-parser.shtml) carries out language to every statement Method resolves, and obtains embodying Stanford Dependency (SD) of grammer dependence between vocabulary in sentence Set；Then, according to 3 kinds of SD pattern: dobj (a, b) →<a,b,null>, nsubjpass (a, b) → <a,b,null>with prep (a, b)+nsubj (a, c) →<a,b,null>, concentrate from the SD of every statement and obtain Take the skeleton (referred to as initial target) of some critical services targets that this statement comprises；Then, every language is utilized In Ju other SD relation (such as amod, nn, conj and prep) to from this statement obtain each at the beginning of Beginning target carries out information expansion, identifies hiding service goal simultaneously, obtains the candidate service object set of this statement； Finally, the candidate service target obtained from all statements carried out lemmatization and goes stop words to process, obtaining The service goal collection that the text of Web service comprises in describing.

Make S_dRepresent the Web service collection of d, SG_dRepresent the service mesh extracted from all Web services of d Mark collection, thenWSSGAM_dEach element WSSGAM_d(s_i,sg_j) ∈ { 0,1}, WSSGAM_d(s_i,sg_j)=1 represents Web service s_i∈S_dComprise Service goal sg_j∈SG_d；

Step 6.4: all service goals (i.e. SG in statistics d_d) verb (i.e. sgv part) that comprises, Obtain the field verb collection of d, be designated as DV_d；

Step 6.5: calculate the semantic similarity between service goal in d, build service goal semantic similarity Matrix, every a line of matrix represents a service goal in d, and every string also represents a service in d Target, the value of each unit is the semantic similitude between two service goals that the row and column at this unit place is corresponding Degree；

Step 6.6: based on service goal semantic similarity matrix, service goal each in d is expressed as one Vector, the dimension of vector is the sum of service goal in d, a service in every one-dimensional representation d of vector Target, value the most one-dimensional in the vector of each service goal sg is between the service goal that sg is corresponding with this dimension Semantic similarity；Then, utilize K-Means algorithm that the vector of service goals all in d is clustered, To the service goal gathering of d, it is designated as SGC_d。

As preferably, each word w importance to this field in d described in step 6.1, computational methods are such as Under:

$r_{d,w}=\frac{N(w,d)}{\underset{w_i\&Element;d}{\max }N(w_i,d)}\&times;(\mathrm{\&alpha;}\&times;(1-\frac{\left|\right\{d:w\&Element;d\left\}\right|}{\left|D\right|})+(1-\mathrm{\&alpha;})\&times;\frac{N(w,d)}{{\mathrm{\&Sigma;}}_{d_i\&Element;D}N(w,d_j)}),$

Wherein, N (w, d) represents w frequency of occurrence in d,Represent the maximum that in d, word occurs The frequency, | and d:w ∈ d} | representing the field number comprising w, | D | represents that the field belonging to whole Web service is total Number,Represent w frequency of occurrence sum in all spectra；α is little between 0 to 1 Number, acquiescence takes 0.6, can be adjusted by user.

As preferably, semantic similarity between service goal in d described in step 6.5, computational methods are as follows:

First, the word that service goal sg each in d comprises is divided into 3 subsets, i.e.

V_d(sg)=W (sg) ∩ DV_d,CN_d(sg)=W (sg) ∩ DCN_d,

Oth_d(sg)=W (sg)-V_d(sg)-CN_d(sg),

Wherein, W (sg) represents the word set that sg comprises；DV_dAnd DCN_dRepresent respectively d field verb collection and Field core name word set；V_d(sg)、CN_dAnd Oth (sg)_d(sg) represent respectively verb collection that sg comprises, Field core name word set and other word sets in addition to verb with field core noun are (such as adjective, field non-core Heart noun etc.)；For service goal, the importance priority between these 3 kinds of words is: field core noun > Verb > other words；

Then, any two service goal sg in d is calculated_iAnd sg_jBetween semantic similarity, use as follows Formula:

$\begin{array}{c}gSim({\mathrm{sg}}_i,{\mathrm{sg}}_j)={\mathrm{\&lambda;}}_1\&times;WSim\left(V_d\right({\mathrm{sg}}_i),V_d\left({\mathrm{sg}}_j\right))+{\mathrm{\&lambda;}}_2\&times;WSim\left({\mathrm{CN}}_d\right({\mathrm{sg}}_i),{\mathrm{CN}}_d\left({\mathrm{sg}}_j\right))\\ +{\mathrm{\&lambda;}}_3\&times;WSim\left({\mathrm{Oth}}_d\right({\mathrm{sg}}_i),{\mathrm{Oth}}_d\left({\mathrm{sg}}_j\right))\end{array},$

Wherein, λ₁、λ₂And λ₃For weight factor；Acquiescence takes λ₁=0.3, λ₂=0.6, λ₃=0.1, can be by User adjusts；WSim(W₁,W₂) represent two word sets W₁And W₂Between semantic similarity, computing formula is:

$WSim(W_1,W_2)=\left\{\begin{array}{cc}\frac{1}{\left|W_1\right|}\&times;\underset{w_i\&Element;W_1}{\&Sigma;}\underset{w_j\&Element;W_2}{max}\left\{wsim(w_i,w_j)\right\},& if\left|W_1\right|\&le;\left|W_2\right|\\ \frac{1}{\left|W_2\right|}\&times;\underset{w_i\&Element;W_2}{\&Sigma;}\underset{w_j\&Element;W_1}{max}\left\{wsim(w_i,w_j)\right\},& if\left|W_1\right|>\left|W_2\right|\end{array}\right.,$

$wsim(w_i,w_j)=\left\{\begin{array}{cc}1,& if{w}_{i}equalsto{w}_j\\ WNSim(w_i,w_j),& otherwise\end{array}\right.,$

Wherein, | W₁| and | W₂| represent word set W respectively₁And W₂In the number of word that comprises；wsim(w_i,w_j) represent Two word w_iAnd w_jBetween semantic similarity；Represent word set W₂In all words with Word w_iBetween maximum semantic similarity；WNSim(w_i,w_j) represent two word w_iAnd w_jAt WordNet Semantic similarity in (http://wordnet.princeton.edu/).

As preferably, implementing of step 7 includes following sub-step:

Step 7.1: given user is inquired about q and carries out pretreatment, including participle, lemmatization, goes to disable Word and word frequency statistics；

Step 7.2: calculate the matching degree between q and each field, obtain the field the highest with q matching degree, It is designated as md；

Step 7.3: calculate the semantic similarity between each service goal bunch in q and md, obtain and q language The service goal gathering that justice is similar, is designated as mC；

Step 7.4: calculate the semantic similarity between each service goal and q in mC, and according to semantic phase Like degree to service goal descendings all in mC, obtain the service goal recommendation list of q, therefrom user's choosing Select the service goal that can embody its demand, be designated as SG_q, as new inquiry；

Step 7.5: by SG_qThe service goal collection of Web service each with md mates, and is met Candidate's Web service collection of user's request.

As preferably, described in step 7.2, calculate the matching degree between q and each field d, use following public Formula:

$match(q,d)=\frac{{\mathrm{\&Sigma;}}_{k=1}^K({\mathrm{\&pi;}}_{q,w_k}\&times;{\mathrm{\&pi;}}_{d,w_k})}{\sqrt{\left({\mathrm{\&Sigma;}}_{k=1}^n{{\mathrm{\&pi;}}_{q,w_k}}^2\right)\&times;\left({\mathrm{\&Sigma;}}_{k=1}^n{{\mathrm{\&pi;}}_{d,w_k}}^2\right)}},$

${\mathrm{\&pi;}}_{q,w_k}=r_{d,w_k}\&times;log(1+f_{q,w_k}),$

${\mathrm{\&pi;}}_{d,w_k}=\left\{\begin{array}{cc}{r}_{d,w_k}\&times;(1+f_{q,w_k}),& if{f}_{q,w_k}\&GreaterEqual;1\\ r_{d,w_k},& otherwise\end{array}\right.,$

Wherein, K represents the Field Words sequencing table RDWL only considering d_dIn K word of top, acquiescence takes 300, can be adjusted by user；Represent word w_kImportance to d；Represent word w_kFrequency in q Secondary.

As preferably, described in step 7.3, calculate the semantic similitude between each service goal bunch in q and md Degree, concrete grammar is as follows:

First, by service goal bunch C each in md_iThe word comprised is divided into 3 subsets, i.e.

$V_{md}\left(C_i\right)={\&cup;}_{{\mathrm{sg}}_k\&Element;C_i}{V}_{md}\left({\mathrm{sg}}_k\right),{\mathrm{CN}}_{md}\left(C_i\right)={\&cup;}_{{\mathrm{sg}}_k\&Element;C_i}{\mathrm{CN}}_{md}\left({\mathrm{sg}}_k\right),$

${\mathrm{Oth}}_{md}\left(C_i\right)={\&cup;}_{{\mathrm{sg}}_k\&Element;C_i}{\mathrm{Oth}}_{md}\left({\mathrm{sg}}_k\right),$

Wherein, V_md(C_i)、CN_md(C_i) and Oth_md(C_i) represent C respectively_iThe verb collection that comprises, field core name Word set and other word sets in addition to verb with field core noun；V_md(sg_k)、CN_md(sg_k) and Oth_md(sg_k) represent service goal sg respectively_kThe verb collection that comprises, field core name word set and except verb and neck Other word sets outside the core noun of territory.

Then, V is obtained respectively_md(C_i)、CN_md(C_i) and Oth_md(C_iWith any word semantic similitude in q in) Vocabulary subset；

V_md(C_iWith the vocabulary subset of any word semantic similitude in q in), it is defined as follows:

Wherein, W (q) represents the word set that q comprises；θ_vFor similarity threshold；wsim(w_i, w_j) represent two word w_i And w_jBetween semantic similarity.

CN_md(C_iWith the vocabulary subset of any word semantic similitude in q in), it is defined as follows:

Wherein, W (q) represents the word set that q comprises；θ_cnFor similarity threshold；wsim(w_i, w_j) represent two words w_iAnd w_jBetween semantic similarity.

Oth_md(C_iWith the vocabulary subset of any word semantic similitude in q in), it is defined as follows:

Wherein, W (q) represents the word set that q comprises；θ_othFor similarity threshold；wsim(w_i, w_j) represent two words w_iAnd w_jBetween semantic similarity.

Finally, q and C is calculated_iBetween semantic similarity, use equation below:

Wherein, V_md(C_i)、CN_md(C_i) and Oth_md(C_i) represent C respectively_iThe verb collection that comprises, field core name Word set and other word sets in addition to verb with field core noun；WithRepresent V respectively_md(C_i)、CN_md(C_i) and Oth_md(C_iWith any word justice phase in q in) As vocabulary subset；N(w_k,C_i) represent word w_kAt C_iIn the frequency；λ₁、λ₂And λ₃For weight factor, Corresponding to step 6.5 calculates the λ in the semantic similarity between service goal₁、λ₂And λ₃。

As preferably, described in step 7.3, mC is defined as follows:

MC={C_i|C_i∈SGC_md∧qCSim(q,C_i)≥θ_c,

Wherein, SGC_mdRepresent the service goal gathering of md；qCSim(q,C_i) represent q and C_iBetween semanteme Similarity；θ_cFor similarity threshold, acquiescence takes 0.4, can be adjusted by user.

As preferably, described in step 7.4, calculate the semantic similarity between each service goal and q in mC, Concrete grammar is as follows:

First, the word that q comprises is divided into 3 subsets, it may be assumed that

V_md(q)=W (q) ∩ DV_md,CN_md(q)=W (q) ∩ DCN_md,

Oth_md(q)=W (q)-V_md(q)-CN_md(q),

Wherein, W (q) represents the word set that q comprises；DV_mdAnd DCN_mdRepresent the field verb collection of md respectively With field core name word set；V_md(q)、CN_md(q) and Oth_md(q) represent respectively verb collection that q comprises, Field core name word set and other word sets in addition to verb with field core noun.

Then, the semantic similarity between each service goal sg and q in mC is calculated:

$\begin{array}{c}qGSim(q,sg)={\mathrm{\&lambda;}}_1\&times;{\mathrm{WSim}}^a\left(V_{md}\right(q),V_{md}\left(sg\right))+{\mathrm{\&lambda;}}_2\&times;{\mathrm{WSim}}^a\left({\mathrm{CN}}_{md}\right(q),{\mathrm{CN}}_{md}\left(sg\right))\\ +{\mathrm{\&lambda;}}_3\&times;{\mathrm{WSim}}^a\left({\mathrm{Oth}}_{md}\right(q),{\mathrm{Oth}}_{md}\left(sg\right))\end{array},$

Wherein, V_md(q)、CN_md(q) and Oth_mdQ () represents verb collection, the field core noun that q comprises respectively Collection and other word sets in addition to verb with field core noun；V_md(sg)、CN_mdAnd Oth (sg)_md(sg) Represent verb collection that sg comprises, field core name word set and its in addition to verb with field core noun respectively His word set；λ₁、λ₂And λ₃For weight factor, corresponding to step 6.5 calculates the semanteme between service goal λ in similarity₁、λ₂And λ₃；WSim^a(W₁,W₂) represent two word sets W₁And W₂Between asymmetric Semantic similarity, calculation is:

${\mathrm{WSim}}^a(W_1,W_2)=\frac{1}{\left|W_1\right|}\&times;\underset{w_i\&Element;W_1}{\&Sigma;}\underset{w_j\&Element;W_2}{max}\left\{wsim(w_i,w_j)\right\},$

Wherein, | W₁| represent word set W₁In the number of word that comprises；wsim(w_i,w_j) represent two word w_iAnd w_jIt Between semantic similarity；Represent word set W₂In all words and word w_iBetween maximum language Justice similarity.

As preferably, the collection of candidate's Web service described in step 7.5 is:

Wherein, S_mdRepresent the Web service collection of md；SG_qRepresent user to select from the service goal recommendation list of q The service goal collection selected；SG_sRepresent the service goal collection of Web service s, can be by WSSGAM_mdObtain, I.e. SG_s={ sg_j|WSSGAM_md(s,sg_j)=1}；W(sg_i) and W (sg_j) represent service goal sg respectively_i And sg_jThe word set comprised.

Relative to prior art, the invention has the beneficial effects as follows:

1) can find to meet the RESTful Web service of user's request target exactly, there is good reality The property used；

2) by recommending the service goal of semantic similitude for the initial query of user, user can be helped to formulate energy Enough embody the high-quality inquiry of its demand.

Accompanying drawing explanation

Fig. 1 is the overall framework schematic diagram of the embodiment of the present invention；

Fig. 2 is the schematic flow sheet that in the embodiment of the present invention, domain knowledge base builds；

Fig. 3 is the part log-on message of Web service in PWeb " Tgels "；

Fig. 4 is in the embodiment of the present invention to service the 3rd article of statement in the text description of Web service " Tgels " The process schematic of Objective extraction；

In Fig. 5 embodiment of the present invention in Travel field with inquiry " get hotel " semantic similarity >=θ_c2 The service goal recommendation list of individual service goal bunch and correspondence, wherein the service goal of band underscore represents user The service goal selected.

Detailed description of the invention

Understand and implement the present invention for the ease of those of ordinary skill in the art, below in conjunction with the accompanying drawings and embodiment pair The present invention is described in further detail, it will be appreciated that enforcement example described herein is merely to illustrate reconciliation Release the present invention, be not intended to limit the present invention.

Carry out object-oriented RESTful Web service with the Web service on PWeb website below and be found to be real Execute example, and combine accompanying drawing, describe the implementation process of the present invention in detail.

PWeb is the famous mashup and Web service registration center that can openly access on current the Internet.By On March 23rd, 2016, the Web service of the upper registration of PWeb has reached 14,836, including follow SOAP, REST, All kinds of Web services of the agreements such as XML-RPC, and provide some log-on messages of Web service, such as name, Text description, classification (i.e. art) etc..Fig. 3 illustrates Web service in PWeb " Tgels " The part log-on message of (http://www.programmableweb.com/api/tgels).

Asking for an interview Fig. 1, the present embodiment is first carried out step 1, uses reptile to collect 6 fields from PWeb website: Financial, Music, Media Management, Payment, Travel, Weather, Web clothes The log-on message of business, describes and classification including name, text.

Then, the Web service to collecting carries out pretreatment, specifically includes: utilize Apache Lucene The StandardAnalyzer instrument of (http://lucene.apache.org/) letter to each Web service Breath resolves, and obtains all words that each Web service comprises；Then, utilize WordNet's Lemmatization instrument carries out lemmatization process to all words；Then, remove without real according to disabling vocabulary The word of border meaning；Finally, the frequency of each word in statistics Web service.

Although the Web service collection collected possesses category division, but exists the most overlapping between classification, as " Financial " and " Payment ", " Music " and " Media Management ", " Travel " " Weather ".Therefore, use Web service sorting technique that the category division of Web service collection is adjusted.

Ask for an interview Fig. 2, then perform step 2, be that 6 fields carry out domain knowledge base structure；

First, all words and each word comprised in the pretreated Web service in each field are added up in this field In frequency of occurrence；Then, each specific field d is carried out domain knowledge base structure, specifically includes: meter Calculate each word importance to this field in d, and according to importance to word descendings all in d, obtain Field Words sequencing table RDWL_d；Then, RDWL is selected_d150 words of top in noun, constitute Field core name word set DCN_d。

Then, the text of each Web service from d carries out service goal extraction in describing, obtain Web service- Service goal incidence matrix WSSGAM_d.Fig. 4 is with the 3rd article of language in the text description of Web service " Tgels " Sentence: " API methods support listings of flight options, as well as creating And deleting flight deal packages. " as a example by illustrate service goal extract process.

Then, in statistics d, the sgv part of all service goals, obtains field verb collection DV_d；

Then, calculate the semantic similarity between service goal in d, build service goal semantic similarity matrix； Based on service goal semantic similarity matrix, obtain the vector representation of all service goals in d；Then, use The vector of service goals all in d is clustered by K-Means algorithm, obtains service goal gathering SGC_d。

Finally, perform step 3, as a example by user's inquiry " get hotel " (being designated as q), carry out Web service Find, obtain candidate's Web service collection；

First, q is carried out pretreatment, including participle, lemmatization, remove stop words and word frequency statistics；Then, The matching degree calculated between q and 6 fields (only considers the top 300 of the Field Words sequencing table in each field Individual word, i.e. arranges K=300).Table 1 is the matching degree in 6 fields and q, it can be seen that with q matching degree High field is Travel.

16 fields of table and the matching degree inquiring about " get hotel "

Then, semantic similarity (the threshold value setting between each service goal bunch in q and Travel field is calculated For: θ_v=θ_cn=θ_oth=0.8, θ_c=0.4) semantic similarity >=θ with q, is obtained_cService goal Gathering mC.Then, the semantic similarity between each service goal in q and mC is calculated, and according to semanteme Similarity, to service goal descendings all in mC, obtains service goal recommendation list.From recommendation list, User is selected to the service goal embodying its demand as new inquiry.

Fig. 5 is the semantic similarity >=θ with q_c2 service goals bunch and correspondence service goal recommend row Table, wherein the service goal of band underscore represents the service goal that user selects.

Finally, service goal collection user selected and the service goal collection of each Web service in Travel field Mate, be met candidate's Web service collection of user's request.Table 2 be 5 candidate's Web services and they Comprise with user selected by the service goal mated of service goal.

5 candidate's Web services of table 2 " get hotel "

It should be appreciated that the part that this specification does not elaborates belongs to prior art.

It should be appreciated that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered Restriction to scope of patent protection of the present invention, those of ordinary skill in the art is under the enlightenment of the present invention, not Depart under the ambit that the claims in the present invention are protected, it is also possible to make replacement or deformation, each fall within this Within bright protection domain, the scope that is claimed of the present invention should be as the criterion with claims.

Claims (13)

1. an object-oriented RESTful Web service finds method, it is characterised in that include following step Rapid:

Step 1: collect the information of RESTful Web service, including Web service title, art and merit The text of energy property describes, and obtains Web service collection；

Step 2: Web service collection is carried out pretreatment；

Step 3: for pretreated Web service collection, it is judged that whether Web service comprises the letter of art Breath；

The most then perform following step 4；

If it is not, then perform following step 5；

Step 4: judge field divides whether there is overlap；

The most then perform following step 5；

If it is not, then perform following step 6；

Step 5: use Web service sorting technique that Web service collection is carried out field division；

Step 6: carry out the structure of domain knowledge base for the Web service collection in each field, including: domain term Remittance sequencing table, field verb collection, field core name word set, Web service-service goal incidence matrix and service Target gathering；

Step 7: given user is inquired about q and carries out Web service discovery.

Object-oriented RESTful Web service the most according to claim 1 finds method, and its feature exists In: described in step 2, Web service collection is carried out pretreatment, including participle, lemmatization, go stop words and Word frequency statistics, implements and includes following sub-step:

Step 2.1: resolve the information of each Web service, obtains all words wherein comprised；

Step 2.2: all words are reduced into its basic original shape；

Step 2.3: remove some words without practical significance according to disabling vocabulary；

Step 2.4: add up each word frequency in Web service.

Object-oriented RESTful Web service the most according to claim 1 finds method, and its feature exists In, implementing of step 5 includes following sub-step:

Step 5.1: specify the field needing to carry out classifying, is divided into what designated field was correlated with by Web service collection The unrelated Web service collection of Web service collection and field；

Step 5.2: the unrelated Web service of Web service collection relevant from field respectively and field is concentrated and chosen number The Web service structure training set that mesh is identical, the Web service in addition to training set constitutes test set；

Step 5.3: build vector space model for the Web service in training set and test set, will each Web Agent list is shown as a vector, and the dimension of vector is the sum of the different words that all Web services comprise, vector Every one word of one-dimensional representation, each word weights in each Web service vector use TF-IDF to calculate；

Step 5.4: use support vector machine that the vector space built is classified, obtain the Web of designated field Services set.

Object-oriented RESTful Web service the most according to claim 1 finds method, and its feature exists In, in step 6, first to each field, add up the pretreated Web service in this field and concentrate the institute comprised There is word and each word frequency of occurrence sum in all Web services in this field；Then, for specific field D carries out domain knowledge base structure, and it implements and includes following sub-step:

Step 6.1: calculate each word importance to this field in d, and according to importance to words all in d Descending, obtains the Field Words sequencing table of d, is designated as RDWL_d；

Step 6.2: select RDWL_d150 words of top in noun, constitute d field core noun Collection, is designated as DCN_d；

Step 6.3: the text of each Web service from d carries out service goal extraction in describing, and obtains d's Web service-service goal incidence matrix, is designated as WSSGAM_d；

Step 6.4: the verb that in statistics d, all service goals comprise, obtains the field verb collection of d, is designated as DV_d；

Step 6.5: calculate the semantic similarity between service goal in d, build service goal semantic similarity Matrix, every a line of matrix represents a service goal in d, and every string also represents a service in d Target, the value of each unit is the semantic similitude between two service goals that the row and column at this unit place is corresponding Degree；

Step 6.6: based on service goal semantic similarity matrix, service goal each in d is expressed as one Vector, the dimension of vector is the sum of service goal in d, a service in every one-dimensional representation d of vector Target, value the most one-dimensional in the vector of each service goal sg is between the service goal that sg is corresponding with this dimension Semantic similarity；Then, utilize K-Means algorithm that the vector of service goals all in d is clustered, To the service goal gathering of d, it is designated as SGC_d。

Object-oriented RESTful Web service the most according to claim 4 finds method, and its feature exists In, each word w importance to this field in d described in step 6.1, computational methods are as follows:

$r_{d,w}=\frac{N(w,d)}{\underset{w_i\&Element;d}{max}N(w_i,d)}\&times;(\mathrm{\&alpha;}\&times;(1-\frac{\left|\right\{d:w\&Element;d\left\}\right|}{\left|D\right|})+(1-\mathrm{\&alpha;})\&times;\frac{N(w,d)}{{\mathrm{\&Sigma;}}_{d_j\&Element;D}N(w,d_j)})$

Wherein, N (w, d) represents w frequency of occurrence in d,Represent the maximum that in d, word occurs The frequency, | and d:w ∈ d} | representing the field number comprising w, | D | represents that the field belonging to whole Web service is total Number,Representing w frequency of occurrence sum in all spectra, α is little between 0 to 1 Number.

Object-oriented RESTful Web service the most according to claim 4 finds method, and its feature exists In, described in step 6.3, from d, the text of each Web service carries out service goal extraction in describing, its tool Body realizes step: first, obtains text and describes the statement set comprised；Then, Stanford Parser is utilized Every statement is carried out syntax parsing, obtains embodying the SD set of grammer dependence between vocabulary in sentence； Then, according to 3 kinds of SD pattern: dobj (a, b) →<a,b,null>, nsubjpass (a, b) →<a,b,null> With prep (a, b)+nsubj (a, c) →<a,b,null>, concentrate from the SD of every statement and obtain this statement bag The skeleton of some the critical services targets contained, referred to as initial target；Then, other is utilized in every statement SD relation carries out information expansion to each initial target obtained from this statement, identifies hiding service simultaneously Target, obtains the candidate service object set of this statement；Finally, to the candidate service mesh obtained from all statements Mark carries out lemmatization and goes stop words to process, and obtains the service goal collection comprised in the text description of Web service.

Object-oriented RESTful Web service the most according to claim 4 finds method, and its feature exists In, semantic similarity between service goal in d described in step 6.5, computational methods are as follows:

First, the word that service goal sg each in d comprises is divided into 3 subsets, i.e.

V_d(sg)=W (sg) ∩ DV_d,CN_d(sg)=W (sg) ∩ DCN_d,

Oth_d(sg)=W (sg)-V_d(sg)-CN_d(sg),

Wherein, W (sg) represents the word set that sg comprises；DV_dAnd DCN_dRepresent respectively d field verb collection and Field core name word set；V_d(sg)、CN_dAnd Oth (sg)_d(sg) represent respectively verb collection that sg comprises, Field core name word set and other word sets in addition to verb with field core noun；For service goal, this Importance priority between 3 kinds of words is: field core noun > verb > other words；

Then, any two service goal sg in d is calculated_iAnd sg_jBetween semantic similarity, use as follows Formula:

gSim(sg_i,sg_j)=λ₁×WSim(V_d(sg_i),V_d(sg_j))+λ₂×WSim(CN_d(sg_i),CN_d(sg_j))

+λ₃×WSim(Oth_d(sg_i),Oth_d(sg_j)),

Wherein, λ₁、λ₂And λ₃For weight factor；WSim(W₁,W₂) represent two word sets W₁And W₂Between language Justice similarity, computing formula is:

$WSim(W_1,W_2)=\left\{\begin{array}{cc}\frac{1}{\left|W_1\right|}\&times;\underset{w_i\&Element;W_1}{\&Sigma;}\underset{w_j\&Element;W_2}{max}\left\{wsim(w_i,w_j)\right\},& if\left|W_1\right|\&le;\left|W_2\right|\\ \frac{1}{\left|W_2\right|}\&times;\underset{w_i\&Element;W_2}{\&Sigma;}\underset{w_j\&Element;W_1}{max}\left\{wsim(w_i,w_j)\right\},& if\left|W_1\right|>\left|W_2\right|\end{array}\right.,$

$wsim(w_i,w_j)=\left\{\begin{array}{cc}1,& if{w}_{i}equalsto{w}_j\\ WNSim(w_i,w_j),& otherwise\end{array}\right.,$

Wherein, | W₁| and | W₂| represent word set W respectively₁And W₂In the number of word that comprises；wsim(w_i,w_j) represent Two word w_iAnd w_jBetween semantic similarity；Represent word set W₂In all words with Word w_iBetween maximum semantic similarity；WNSim(w_i,w_j) represent two word w_iAnd w_jAt WordNet In semantic similarity.

Object-oriented RESTful Web service the most according to claim 1 finds method, and its feature exists In, implementing of step 7 includes following sub-step:

Step 7.1: given user is inquired about q and carries out pretreatment, including participle, lemmatization, goes to disable Word and word frequency statistics；

Step 7.2: calculate the matching degree between q and each field, obtain the field the highest with q matching degree, It is designated as md；

Step 7.3: calculate the semantic similarity between each service goal bunch in q and md, obtain and q language The service goal gathering that justice is similar, is designated as mC；

Step 7.4: calculate the semantic similarity between each service goal and q in mC, and according to semantic phase Like degree to service goal descendings all in mC, obtain the service goal recommendation list of q, therefrom user's choosing Select the service goal that can embody its demand, be designated as SG_q, as new inquiry；

Step 7.5: by SG_qThe service goal collection of Web service each with md mates, and is met Candidate's Web service collection of user's request.

Object-oriented RESTful Web service the most according to claim 8 finds method, and its feature exists In, calculate the matching degree between q and each field d described in step 7.2, employing equation below:

$match(q,d)=\frac{{\mathrm{\&Sigma;}}_{k=1}^K({\mathrm{\&pi;}}_{q,w_k}\&times;{\mathrm{\&pi;}}_{d,w_k})}{\sqrt{\left({\mathrm{\&Sigma;}}_{k=1}^n{{\mathrm{\&pi;}}_{q,w_k}}^2\right)\&times;\left({\mathrm{\&Sigma;}}_{k=1}^n{{\mathrm{\&pi;}}_{d,w_k}}^2\right)}},$

${\mathrm{\&pi;}}_{q,w_k}=r_{d,w_k}\&times;log(1+f_{q,w_k}),$

${\mathrm{\&pi;}}_{d,w_k}=\left\{\begin{array}{cc}{r}_{d,w_k}\&times;(1+f_{q,w_k}),& if{f}_{q,w_k}\&GreaterEqual;1\\ r_{d,w_k},& otherwise\end{array}\right.,$

Wherein, K represents the Field Words sequencing table RDWL only considering d_dIn K word of top；Generation Table word w_kImportance to d；Represent word w_kThe frequency in q.

Object-oriented RESTful Web service the most according to claim 8 finds method, its feature It is, calculates the semantic similarity between each service goal bunch in q and md described in step 7.3, specifically Method is as follows:

First, by service goal bunch C each in md_iThe word comprised is divided into 3 subsets, i.e.

$V_{md}\left(C_i\right)={\&cup;}_{{\mathrm{sg}}_k\&Element;C_i}{V}_{md}\left({\mathrm{sg}}_k\right),{\mathrm{CN}}_{md}\left(C_i\right)={\&cup;}_{{\mathrm{sg}}_k\&Element;C_i}{\mathrm{CN}}_{md}\left({\mathrm{sg}}_k\right),$

${\mathrm{Oth}}_{md}\left(C_i\right)={\&cup;}_{{\mathrm{sg}}_k\&Element;C_i}{\mathrm{Oth}}_{md}\left({\mathrm{sg}}_k\right),$

Wherein, V_md(C_i)、CN_md(C_i) and Oth_md(C_i) represent C respectively_iThe verb collection that comprises, field core name Word set and other word sets in addition to verb with field core noun；V_md(sg_k)、CN_md(sg_k) and Oth_md(sg_k) represent service goal sg respectively_kThe verb collection that comprises, field core name word set and except verb and neck Other word sets outside the core noun of territory；

Then, V is obtained respectively_md(C_i)、CN_md(C_i) and Oth_md(C_iWith any word semantic similitude in q in) Vocabulary subset；

V_md(C_iWith the vocabulary subset of any word semantic similitude in q in), it is defined as follows:

Wherein, W (q) represents the word set that q comprises；θ_vFor similarity threshold；wsim(w_i,w_j) represent two word w_i And w_jBetween semantic similarity；

CN_md(C_iWith the vocabulary subset of any word semantic similitude in q in), it is defined as follows:

Wherein, W (q) represents the word set that q comprises；θ_cnFor similarity threshold；wsim(w_i,w_j) represent two words w_iAnd w_jBetween semantic similarity；

Oth_md(C_iWith the vocabulary subset of any word semantic similitude in q in), it is defined as follows:

Wherein, W (q) represents the word set that q comprises；θ_othFor similarity threshold；wsim(w_i,w_j) represent two words w_iAnd w_jBetween semantic similarity；

Finally, q and C is calculated_iBetween semantic similarity:

Wherein, V_md(C_i)、CN_md(C_i) and Oth_md(C_i) represent C respectively_iThe verb collection that comprises, field core name Word set and other word sets in addition to verb with field core noun；WithRepresent V respectively_md(C_i)、CN_md(C_i) and Oth_md(C_iWith any word justice phase in q in) As vocabulary subset；N(w_k,C_i) represent word w_kAt C_iIn the frequency；λ₁、λ₂And λ₃For weight factor, Corresponding to the λ in claim 7₁、λ₂And λ₃。

11. object-oriented RESTful Web services according to claim 10 find method, its feature Being, described in step 7.3, mC is defined as follows:

MC={C_i|C_i∈SGC_md∧qCSim(q,C_i)≥θ_c,

Wherein, SGC_mdRepresent the service goal gathering of md；qCSim(q,C_i) represent q and C_iBetween semanteme Similarity, calculation is given the most；θ_cFor similarity threshold.

12. object-oriented RESTful Web services according to claim 8 find method, its feature It is, described in step 7.4, calculates the semantic similarity between each service goal and q, specifically side in mC Method is as follows:

First, the word that q comprises is divided into 3 subsets, it may be assumed that

V_md(q)=W (q) ∩ DV_md,CN_md(q)=W (q) ∩ DCN_md,

Oth_md(q)=W (q)-V_md(q)-CN_md(q),

Wherein, W (q) represents the word set that q comprises；DV_mdAnd DCN_mdRepresent the field verb collection of md respectively With field core name word set；V_md(q)、CN_md(q) and Oth_md(q) represent respectively verb collection that q comprises, Field core name word set and other word sets in addition to verb with field core noun；

Then, the semantic similarity between each service goal sg and q in mC is calculated, employing formula:

QGSim (q, sg)=λ₁×WSim^a(V_md(q),V_md(sg))+λ₂×WSim^a(CN_md(q),CN_md(sg))

+λ₃×WSim^a(Oth_md(q),Oth_md(sg)),

Wherein, V_md(q)、CN_md(q) and Oth_mdQ () represents verb collection, the field core noun that q comprises respectively Collection and other word sets in addition to verb with field core noun；V_md(sg)、CN_mdAnd Oth (sg)_md(sg) Represent verb collection that sg comprises, field core name word set and its in addition to verb with field core noun respectively His word set；λ₁、λ₂And λ₃For weight factor, corresponding to the λ in claim 7₁、λ₂And λ₃； WSim^a(W₁,W₂) represent two word sets W₁And W₂Between asymmetric semantic similarity, calculation is:

${\mathrm{WSim}}^a(W_1,W_2)=\frac{1}{\left|W_1\right|}\&times;\underset{w_i\&Element;W_1}{\&Sigma;}\underset{w_j\&Element;W_2}{max}\left\{wsim(w_i,w_j)\right\},$

Wherein, | W₁| represent word set W₁In the number of word that comprises；wsim(w_i,w_j) represent two word w_iAnd w_jIt Between semantic similarity, calculation is given the most；Represent word Collection W₂In all words and word w_iBetween maximum semantic similarity.

13. object-oriented RESTful Web services according to claim 8 find method, its feature Being, the collection of candidate's Web service described in step 7.5 is:

Wherein, S_mdRepresent the Web service collection of md；SG_qRepresent user to select from the service goal recommendation list of q The service goal collection selected；SG_sRepresent the service goal collection of Web service s, can be by WSSGAM_mdObtain, I.e. SG_s={ sg_j|WSSGAM_md(s,sg_j)=1}；W(sg_i) and W (sg_j) represent sg respectively_iAnd sg_jBag The word set contained.