CN109471982B - Web service recommendation method based on QoS (quality of service) perception of user and service clustering - Google Patents

Web service recommendation method based on QoS (quality of service) perception of user and service clustering Download PDF

Info

Publication number
CN109471982B
CN109471982B CN201811394933.8A CN201811394933A CN109471982B CN 109471982 B CN109471982 B CN 109471982B CN 201811394933 A CN201811394933 A CN 201811394933A CN 109471982 B CN109471982 B CN 109471982B
Authority
CN
China
Prior art keywords
service
user
clustering
qos
similar
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201811394933.8A
Other languages
Chinese (zh)
Other versions
CN109471982A (en
Inventor
暴建民
温韬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN201811394933.8A priority Critical patent/CN109471982B/en
Publication of CN109471982A publication Critical patent/CN109471982A/en
Application granted granted Critical
Publication of CN109471982B publication Critical patent/CN109471982B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention provides a Web service recommendation method based on QoS (quality of service) perception of user and service clustering, which is based on QoS (quality of service) record of user calling service and uses cosine similarity calculation to cluster users with similar service preference; then, extracting and normalizing feature words of a WSDL file of a description document of the service, and clustering the services with similar functional features by adopting a K-means + + algorithm; and finally integrating the similar user set and the similar service set into a collaborative filtering algorithm to calculate the missing QoS value of the service invocation matrix and generate recommendation. The result of a simulation experiment performed by using a data set provided by a WSDream open source project shows that the Web service recommendation method based on user and service clustering has a better result on two main evaluation indexes of average absolute error and root mean square error compared with the prior art.

Description

Web service recommendation method based on QoS (quality of service) perception of user and service clustering
Technical Field
The invention relates to the field of service recommendation in data mining, in particular to a Web service recommendation method based on QoS (quality of service) perception of user and service clustering.
Background
With the rapid development of the current internet and the explosive growth of web services with various functions available thereon, it is increasingly difficult for users to select appropriate web services, and thus there is a strong need for a related method for automatically and efficiently selecting appropriate web services for recommendation to users. In the face of massive service types and service information, how to efficiently and accurately recommend relevant services to users according to user specificity is becoming a research problem in the industry. Web service recommendations provide a direction to address this dilemma.
The web service recommendation is to establish a recommendation model for service recommendation according to the basic information, the history call record, the preference attribute of the user and the like of the user. The web service recommendation becomes a research hotspot in the field of service computing in recent years, the current academic community has matured the research on the web service recommendation, wherein the prediction and recommendation of related services and variables are realized on the basis of QoS (quality of service), and on the basis, the recommendation precision is improved by taking other factors into consideration, such as introduction of geographic factors or service function factors, which all achieve better results. However, the influence of comprehensive consideration of the user neighborhood and the service function attribute on the recommendation result under the condition that the QoS matrix data is sparse is not mentioned. The performance of the web service recommendation system depends on the recommendation method used by the system, and the basic assumption for the web service recommendation method is that: 1. the user likes a certain service, which also likes a service similar to the service; 2. users like the used services of users with similar background and preference; 3. a user likes a service with a certain feature, and the user also likes other services with the feature. The current web service recommendation method mainly comprises recommendation based on IF/THEN rules, recommendation based on contents, collaborative filtering recommendation, mixed recommendation and the like. Among them, the Collaborative Filtering Recommendation (CFR) is a relatively mature and mainstream Recommendation at present. Compared with other recommendation algorithms, the algorithm has the following advantages: firstly, no special requirement is required on a recommended object, and the recommendation can be realized for complex and abstract resources; second, only explicit or implicit user historical rating data is needed, without attribute knowledge about the user himself, and without any negative impact on the user's recommendation experience. Although the collaborative filtering recommendation algorithm has achieved many important research results, there are still many key problems to be solved urgently, which mainly include the data sparsity problem, the cold start problem, the scalability problem, and so on. The sparsity problem is mainly embodied in that the user has fewer ratings on related web services in the current data set, and potential factors influencing the preference of the user on the web services are fewer, but each potential factor greatly influences the recommendation result, so that the accuracy of web service recommendation is greatly reduced. Many attempts are made by many researchers at home and abroad at present to improve the accuracy and performance of a web service recommendation algorithm, and abundant results are obtained. Zheng et al mainly introduces the factor of the geographical field, alleviates sparsity of a prediction matrix by a matrix decomposition method, adopts a random gradient descent algorithm to quickly converge and give a prediction result, and a collaborative filtering algorithm improved by the scheme has a more accurate prediction result compared with other algorithms. Duncylin and the like propose a collaborative filtering algorithm based on project score prediction, the algorithm fills score null values in a user score item union set by adopting a project-based collaborative filtering method, and calculates similarity among users on the filled union set. Zhang et al indicate that most nearest neighbors searched by a user-based collaborative filtering method have no score on target items, so that the accuracy of score prediction is seriously influenced, an iteration threshold is set for the nearest neighbors, the user-based collaborative filtering method is executed in a circulating mode to fill up the unverified items of the nearest neighbors, and after scores of all the nearest neighbor pair target items are obtained, a final prediction score is generated by adopting a traditional collaborative filtering method; guosheng Kang et al calculate ranking scores of Web service candidates based on QoS preferences and diversity characteristics of users on service potential, and by using interests and QoS preferences of users in historical invocation of Web services, provide an innovative diversity-aware Web service ranking algorithm, construct a Web service graph based on functional similarity between Web services, and evaluate Web service candidate entries according to their scores and diversity derived from the Web service graph; yan Hu et al propose an improved time-aware collaborative filtering method that integrates time information into similarity measurements and QoS predictions for high-performance Web service recommendations.
The researches have better performance in the aspect of realizing prediction and recommending related services and variables based on QoS, but the great influence of service functions and user similarity on the precision of a recommendation result is mostly ignored, many recommendation models are deeply influenced by the high sparsity of QoS matrix data, and the problem of low precision of the recommendation result caused by the factors is urgently solved.
Disclosure of Invention
In order to solve the above problems, the present invention provides a Web service recommendation method based on user and service clustering QoS sensing, which first performs context feature extraction according to WSDL documents of Web services to obtain function description thereof, and performs clustering. And then similarity clustering is carried out on the QoS attributes of the historical calling services of the user, the QoS prediction is carried out on the cooperative service function by adopting an improved matrix decomposition recommendation algorithm, and recommendation is generated, so that the problem of low similarity caused by the sparsity of recommendation matrix data is solved. And finally, the algorithm is adopted for testing, 80% of data sets are used for training the algorithm to obtain the optimal parameters, and the rest data are used for testing. Compared with the other three mainstream web service recommendation algorithms, the test result has better results on two error evaluation parameters of MAE and RMSE.
The invention provides a Web service recommendation method based on QoS (quality of service) perception of user and service clustering, which comprises the following steps:
step 1: preprocessing the service call matrix by the users in the data set, clustering each similar user by adopting a modified cosine similarity algorithm to obtain N user sets U which are most similar to each user(i)The calculation formula (1) is:
Figure GDA0003620711830000041
wherein, IijRepresenting a set of items that have been jointly scored by users I, j, IiRepresented is a set of items scored by user I, IjIs a set of items scored by user j, Ri,kIs the rating of user i on rating item k, Rj,kIs the score of the user j for the item k,
Figure GDA0003620711830000042
respectively averaging the scores of the user i and the user j on the respective scoring items;
step 2: extracting the characteristics of related function description of a Web service description document WSDL file, carrying out Google distance normalization, clustering the services with similar function characteristics by adopting a K-means + + algorithm to obtain a similar service set S(i)The calculation formula (2) is:
Figure GDA0003620711830000043
wherein, FVs1，FVs2Respectively web service S1，S2Is given by the feature word vector, | FVs1The | is a vector base number (mode), Normalized Google Distance, NGD for short, and is a correlation representation of two words obtained by carrying out standardized calculation on data obtained by a Google search engine;
and step 3: solving the characteristic vectors of the user matrix U and the service matrix S of the service call matrix Rpi,qj；
And 4, step 4: let r bei,jThe QoS value invoked for service j for user i, the missing QoS value in R is noted as the element of matrix R
Figure GDA0003620711830000051
Is provided with
Figure GDA0003620711830000052
And 5: to calculate
Figure GDA0003620711830000053
For each non-null ri,jE.g. R, calculating
Figure GDA0003620711830000054
Step 6: for each j e [1, n],k∈U(i)Calculating
Figure GDA0003620711830000055
Wherein e1Is piUpdate factor of (I)i,jAnd the calling condition of the user i to the service j is shown, the calling is 1 if the calling is available, and the calling is 0 if the calling is not available. Alpha is a weight factor, and large alpha indicates that the influence of adjacent users on the current predicted QoS is larger;
and 7: for each i ∈ [1, m ∈ >],d∈S(j)Calculating
Figure GDA0003620711830000056
Wherein e2Is qjBeta is a weight factor, and the larger beta indicates that the functional characteristics of the service have larger influence on the current predicted QoS;
and 8: calculating pi＝pi-γe1，qj＝qj-γe2Wherein gamma is an iteration factor used for controlling the number and speed of iteration;
and step 9: iterate S5 through S8 until all non-empty ri,jCalculating or meeting the iteration termination condition;
step 10: obtaining missing QoS values
Figure GDA0003620711830000057
And recommend and
Figure GDA0003620711830000058
and (5) the similar service is provided for the relevant user to complete the recommendation.
The further improvement lies in that: the step 1 clusters similar users, and the clustering method comprises the following steps:
step 1.1: input Ii,Ij,Iij,Ri,k,Rj,k；
Step 1.2: computing
Figure GDA0003620711830000061
Step 1.3: for each user i, i is more than 0 and less than or equal to m, each service j is not equal to i, j is more than 0 and less than or equal to m-1, and SimU (i, j) is calculated and stored in the temp mapping table;
step 1.4: taking the first N corresponding users in the temp mapping table to form a neighbor set U of the user i(i)＝{Uk|Uk∈Top-N(Ui) I ≠ k }, completing clustering.
The further improvement lies in that: the step 2 clusters similar services, and the clustering method is as follows:
step 2.1: TF-IDF calculation is carried out on the WSDL file, a service function feature description word set is extracted and is recorded as FVsi；
Step 2.2: and (3) carrying out Google distance normalization processing on the feature word set, wherein the calculation formula (3) is as follows:
Figure GDA0003620711830000062
m is the total web page number searched in Google by using the characteristic words x and y, and is the logf (x), the logf (y) is the hit number searched by using the characteristic words x and y respectively, and f (x and y) is the number of web pages simultaneously appearing by using x and y;
step 2.3: performing K-means + + clustering based on the feature word vectors, and selecting the clustering number K;
step 2.4: feature word set vector FV from servicessiRandomly selecting one as an initial clustering center C1；
Step 2.5: calculating the distance D between each vector and the current nearest cluster center(x)Calculating the probability of each vector becoming the next cluster center
Figure GDA0003620711830000063
Determining the next center according to the probability wheel disc;
step 2.6: repeating the step 2.5 until K clustering centers are selected;
step 2.7: FV for each servicesiCalculating the normalized Google distance from the Google distance to K clustering centers, and classifying the Google distance to the class to which the center with the minimum distance belongs;
step 2.8: for each class C(i)Recalculating centers of all the elements in (1);
step 2.9: repeating the step 2.7 and the step 2.8 until the clustering center is not changed any more, and obtaining a web service cluster S(i)＝{s|s∈C(i),i∈[1,k]And finishing clustering.
The further improvement lies in that: the range of NGD in step 2.2 is NGD (x, y) e [0, ∞). The invention has the beneficial effects that: some important features of the service itself, such as the functional description, are feature extracted, and these features play a crucial role in the user preference. And then carrying out similarity clustering on the users based on the QoS of the user calling service. Factors influencing the preference of the user to the web service are fully considered in the aspects of the server and the client, and the problem that the recommendation precision is influenced due to the fact that the service is not fully considered to the recommendation result in the prior art is solved. Aiming at the problems that data of a user service calling matrix is sparse and accuracy of a service recommendation result is greatly influenced, a matrix decomposition model is provided for predicting a missing QoS value, a gradient descent iteration algorithm is adopted for the model, and the result can be obtained efficiently by combining an actual data quantity control iteration factor.
Drawings
FIG. 1 is a scene framework diagram of the present invention.
Fig. 2 is a flow chart of the similar user clustering algorithm of the present invention.
FIG. 3 is a flow chart of a similar service clustering algorithm of the present invention.
Fig. 4 is a flow chart of the gradient descent algorithm of the present invention predicting missing QoS values and generating a recommendation algorithm.
FIG. 5 is a graph comparing recommendation accuracy on MAE under the same data set with the mainstream web service recommendation algorithm.
FIG. 6 is a graph comparing recommendation accuracy on RMSE for the same data set as the mainstream web service recommendation algorithm.
Detailed Description
For the purpose of enhancing understanding of the present invention, the present invention will be further described in detail with reference to the following examples, which are provided for illustration only and are not to be construed as limiting the scope of the present invention.
The data for the embodiment is provided by the open source item WSDream, which contains 339 user calls to 5,825 Web services in the real world, resulting in over one million and over one-half million call records. Data cleaning and preprocessing are carried out on the basis of original data, model training is carried out by using 80% of data in a data set to determine optimal parameters, and the remaining 20% of data are used for verifying the effectiveness of an algorithm model.
The architecture of the web recommendation algorithm based on user and service clustering is shown in FIG. 1. Firstly, similarity calculation and clustering are carried out on users; then, extracting functional description text features of the web service description document WSDL in the data set, and clustering by using a k-means + + algorithm after the Google distance is normalized; and finally integrating the similar users and services into a collaborative filtering recommendation model to predict the QoS value, and comparing the predicted value with the QoS of the related service to generate recommendation.
To describe the algorithm principle, there are the following definitions:
definition 1: the service scores of the user i and the user j are represented on the QoS value of service calling, and the similarity of the user i and the user j is defined on the basis of the service QoS scores of the user, and the similarity is as follows:
Figure GDA0003620711830000091
Iijrepresenting a set of items that have been jointly scored by users I, j, IiRepresented is a set of items scored by user I, IjIs a set of items scored by user j, Ri,kIs the rating of user i on rating item k, Rj,kIs the score of the user j for the item k,
Figure GDA0003620711830000092
respectively, are the average values of the scores of the user i and the user j on the respective scoring items.
Definition 2: for the WSDL service description Document of each web service, a Term-Inverse Document Frequency, abbreviated as TF-IDF method is used to obtain a feature word vector of a service description text, and the feature word vector is marked as FVsiThe feature word vector for the service si is represented. Finally, the feature word set is converted into a numerical value to facilitate similarity calculation. NGD processing is used. NGD is a relevance representation of two words normalized using data from the Google search engine, defined as follows:
Figure GDA0003620711830000093
m is the total web page number searched in Google by using the characteristic words x and y, log f (x), log f (y) is the hit number searched by using the characteristic words x and y respectively, and f (x and y) is the number of web pages appearing by using x and y simultaneously. NGD (x, y) is ∈ [0, ∞ ].
Definition 3: the similarity of two web services is defined as:
Figure GDA0003620711830000094
FVs1，FVs2respectively a Web service S1，S2Is given by the feature word vector, | FVs1And | is the cardinality of the vector. Definition 4: let the scoring matrix of user calls to service be expressed as R ═ UTS, similar user characteristic matrix U belongs to Rl×mThe similar service characteristic matrix S is formed by Rl ×mDefining the feature vector in the user feature matrix as piService feature vector is qj. The missing value in the scoring matrix for service j for user i
Figure GDA0003620711830000101
Is defined as:
Figure GDA0003620711830000102
definition 5: similar users and similar services are integrated into a matrix decomposition model, and are expressed mathematically as follows:
Figure GDA0003620711830000103
U(i)representing a set similar to user i, S(j)Representing a set of services with similar functional descriptions as service j. I isi,jThe call of the user i to the service j is shown, and the call is 1 in some cases and 0 in other cases.
Figure GDA0003620711830000104
Is the square of the F-norm. The latter two terms are regularization terms, λ, used to mitigate the over-fit user service matrixi,λjAre parameters used to control the overfitting constraints. SimU (i, k) represents the similarity between user i and user k, and SimS (j, d) represents the similarity between Web service j and service d. And alpha and beta are weight coefficients for controlling the user characteristics and the service characteristics, wherein the large alpha indicates that similar users have larger influence on the currently predicted QoS, and the large beta indicates that the functional characteristics of the service have larger influence on the currently predicted QoS.
Definition 6: to obtain
Figure GDA0003620711830000105
By devitalizing the equation (5), the elements p of the target feature vectors U and Si,qiThe update iteration mode of (2) is as follows:
Figure GDA0003620711830000106
Figure GDA0003620711830000111
where γ is an iteration factor used to control the number and speed of iterations.
Definition 7: the Mean Absolute Error (MAE) and the Root Mean Square Error (RMSE) of the evaluation factors for evaluating the accuracy of the recommendation result of the recommendation method are defined as follows:
Figure GDA0003620711830000112
Figure GDA0003620711830000113
ri,jis the QoS value actually invoked by user i for Web service j,
Figure GDA0003620711830000114
the model predicts the QoS value, N is the number of predicted values, and the MAE evaluates the relative error of the predicted values to the true values. RMSE emphasizes the relative maximum error. Smaller both represent smaller errors and higher accuracy. The flow of clustering users in the embodiment is shown in fig. 2, and the specific steps are as follows:
step 1.1: input Ii,Ij,Iij,Ri,k,Rj,k；
Step 1.2: computing
Figure GDA0003620711830000115
Step 1.3: for each user i, i is more than 0 and less than or equal to m, each service j is not equal to i, j is more than 0 and less than or equal to m-1, and SimU (i, j) is calculated and stored in the temp mapping table;
step 1.4: taking the first N corresponding users in the temp mapping table to form a neighbor set U of the user i(i)＝{Uk|Uk∈Top-N(Ui),i≠k}；
The process of clustering web services in the embodiment is shown in fig. 3, and the specific steps are as follows:
step 2.1: TF-IDF calculation is carried out on the WSDL file, a service function feature description word set is extracted and is recorded as FVsi；
Step 2.2: and carrying out Google distance normalization processing on the feature word set, wherein the calculation formula is as follows:
Figure GDA0003620711830000121
m is the total web page number searched by Google by using the characteristic words x and y, and logf (x), logf (y) is the hit number searched by using the characteristic words x and y respectively, and f (x and y) is the number of web pages simultaneously appearing by using x and y. NGD (x, y) is E [0, ∞);
step 2.3: performing K-means + + clustering based on the feature word vectors, and selecting the number K of clusters;
step 2.4: feature word set vector FV from servicessiRandomly selecting one as an initial clustering center C1；
Step 2.5: calculating the distance D between each vector and the current nearest cluster center(x)Calculating the probability of each vector becoming the next cluster center
Figure GDA0003620711830000122
Determining the next center according to the probability wheel disc;
step 2.6: repeating the step 2.5 until K clustering centers are selected;
step 2.7: FV for each servicesiCalculate its normalization to K cluster centersGoogle distance, which is classified into the class to which the center with the smallest distance belongs;
step 2.8: for each class C(i)Recalculating centers of all the elements in (1);
step 2.9: repeating the step 2.7 and the step 2.8 until the clustering center is not changed any more, and obtaining the web service cluster S(i)＝{s|s∈C(i),i∈[1,k]}；
The QoS-aware web service recommendation method based on user and service clustering of the embodiment has a flow chart as shown in fig. 4, and specifically includes the following steps:
step 1: preprocessing the service call matrix by the users in the data set, clustering each similar user by adopting a modified cosine similarity algorithm to obtain N user sets U which are most similar to each user(i)The calculation formula is as follows:
Figure GDA0003620711830000131
wherein, IijRepresenting a set of items that have been jointly scored by users I, j, IiRepresented is a set of items scored by user I, IjIs a set of items scored by user j, Ri,kIs the rating of user i on rating item k, Rj,kIs the score of the user j for the item k,
Figure GDA0003620711830000132
respectively averaging the scores of the user i and the user j on the respective scoring items;
and 2, step: extracting the characteristics of related function description of a Web service description document WSDL file, carrying out Google distance normalization, clustering the services with similar function characteristics by adopting a K-means + + algorithm to obtain a similar service set S(i)The calculation formula is as follows:
Figure GDA0003620711830000133
wherein, FVs1，FVs2Respectively web service S1，S2Is given by the feature word vector, | FVs1And | is the vector cardinality (modulo). NGD is a relevance representation of two words obtained by carrying out standardized calculation by utilizing data obtained by a Google search engine;
and 3, step 3: solving the characteristic vector p of the user matrix U and the service matrix S of the service call matrix Ri,qj；
And 4, step 4: let r bei,jThe QoS value invoked for user i for service j is the element of the matrix R. The missing QoS value in R is noted
Figure GDA0003620711830000134
Is provided with
Figure GDA0003620711830000135
And 5: to calculate
Figure GDA0003620711830000136
For each non-empty ri,jE.g. R, calculating
Figure GDA0003620711830000137
Step 6: for each j e [1, n],k∈U(i)Calculating
Figure GDA0003620711830000138
Wherein e1Is piUpdate factor of (I)i,jAnd the calling condition of the user i to the service j is shown, the calling is 1 if the calling is available, and the calling is 0 if the calling is not available. Alpha is a weight factor, and large alpha indicates that the influence of adjacent users on the current predicted QoS is larger;
and 7: for each i ∈ [1, m ∈ >],d∈S(j)Calculating
Figure GDA0003620711830000141
Wherein e2Is qjBeta is a weight factor, and the larger beta indicates that the functional characteristics of the service have larger influence on the current predicted QoS;
and 8: calculating pi＝pi-γe1，qj＝qj-γe2Wherein gamma is an iteration factor used for controlling the number and speed of iteration;
and step 9: iterate S5 through S8 until all non-empty ri,jAnd finishing the calculation or meeting the iteration termination condition.
Step 10: obtaining missing QoS values
Figure GDA0003620711830000142
And recommend and
Figure GDA0003620711830000143
similar services are provided for relevant users; to analyze the recommendation accuracy of the QoS-aware web service recommendation algorithm of the embodiment based on user and service clustering, experimental simulations were performed with data sources from real-world collections, data sets hosted on the github open source item WSDream, which contains 5825 web service invocation records from 339 users in 73 countries. In order to verify the recommendation accuracy of the traditional service recommendation method and the method, six mainstream recommendation methods are selected, the same data set is respectively tested, the MAE and the RMSE are respectively calculated, the recommendation accuracy is compared, and the smaller the MAE and the RMSE is, the smaller the representative error is, and the higher the recommendation accuracy is.
And respectively selecting three mainstream web service recommendation methods of IPCC, NIMF and LoNMF for comparison experiments. Each method was run 10 times and the mean values of recorded MAE and RMSE were calculated and finally compared to the results of the invention. The experimental environment is a Linux operating system, the processor is an Intel (R) core (TM) i7-8550U, 8GB memory, and a result simulation experiment is carried out by using Python and C + +.
For the experimental part of the data in the examples, the values of the respective key parameters were determined in a large amount of data training and tuning, in the examples, α is set to 0.4, β is set to 0.5, γ is set to 0.013, and N is set to 10, fig. 5 and 6 are graphs comparing the results of the comparison of the present invention with other conventional six algorithms on MAE and RMSE, the abscissa is the matrix density, and the higher the abscissa represents the more data available, wherein the higher the accuracy of the present invention compared to the conventional algorithm at each matrix density was achieved.

Claims (4)

1. A Web service recommendation method based on QoS perception of user and service clustering is characterized in that: the method comprises the following steps:
step 1: preprocessing the service call matrix by the users in the data set, clustering each similar user by adopting a modified cosine similarity algorithm to obtain N user sets U which are most similar to each user(i)The calculation formula (1) is:
Figure FDA0003636862590000011
wherein, IijRepresenting a set of items that have been jointly scored by users I, j, IiRepresented is a set of items scored by user I, IjIs a set of items scored by user j, Ri,kIs the rating of user i on rating item k, Rj,kIs the score of the user j for the item k,
Figure FDA0003636862590000012
respectively averaging the scores of the user i and the user j on respective scoring items;
step 2: extracting the characteristics of related function description of a Web service description document WSDL file, carrying out Google distance normalization, clustering the services with similar function characteristics by adopting a K-means + + algorithm to obtain a similar service set S(a)Wherein: the calculation formula (2) is:
Figure FDA0003636862590000013
wherein, FVs1，FVs2Respectively web service S1，S2Is given by the feature word vector, | FVs1The | is a base number model of the vector, and the NGD is the relevance expression of two words obtained by carrying out standardized calculation on data obtained by a Google search engine;
and step 3: solving the characteristic vector p of the user matrix U and the service matrix S of the service call matrix Ri，qn；
And 4, step 4: let r bei,nThe QoS value invoked for service n for user i, the missing QoS value in R is noted as the element of matrix R
Figure FDA0003636862590000021
Is provided with
Figure FDA0003636862590000022
And 5: to calculate
Figure FDA0003636862590000023
For each non-empty ri,nE.g. R, calculating
Figure FDA0003636862590000024
Step 6: for each service n (n is epsilon [1, m)]) User U (U ∈ U)(i)) Computing
Figure FDA0003636862590000025
Wherein e1Is piUpdate factor of (I)i,nRepresenting the calling condition of the user i to the service n, wherein the calling condition is 1 if the calling condition exists, and is 0 if the calling condition does not exist; alpha is a weight factor, and the larger the value of alpha is, the larger the influence of the adjacent user on the current predicted QoS is;
and 7: for each user i, (i e [1, w ]), d is the current service, the calculation is done
Figure FDA0003636862590000026
Wherein e2Is qnBeta is a weight factor, and the larger the value of beta is, the larger the influence of the functional characteristics of the service on the current predicted QoS is;
and 8: calculating pi＝pi-γe1，qn＝qn-γe2Wherein gamma is an iteration factor used for controlling the number and speed of iteration;
and step 9: iterating steps 5-8 until all non-empty ri,nThe iteration termination condition is calculated or met;
step 10: obtaining missing QoS values
Figure FDA0003636862590000027
And recommend and
Figure FDA0003636862590000028
and (5) the similar service is provided for the relevant user to complete the recommendation.
2. The method of claim 1, wherein the method comprises the following steps: the step 1 clusters similar users, and the clustering method is as follows:
step 1.1: input Ii,Ij,Iij,Ri,k,Rj,k；
Step 1.2: computing
Figure FDA0003636862590000031
Step 1.3: for each user i, i is more than 0 and less than or equal to w, j is more than 0 and less than or equal to w-1, calculating SimU (i, j) and storing the SimU (i, j) into a temp mapping table;
step 1.4: taking the first N corresponding users in the temp mapping table to form a neighbor set U of the user i(i)＝{Uj|Uj∈TopN(Uj) And i is not equal to j, and clustering is finished.
3. The method of claim 1, wherein the method comprises the following steps: the step 2 is to cluster similar services, and the clustering method is as follows:
step 2.1: TF-IDF calculation is carried out on the WSDL file, a service function feature description word set is extracted and is recorded as FVsi；
Step 2.2: and (3) carrying out Google distance normalization processing on the feature word set, wherein the calculation formula (3) is as follows:
Figure FDA0003636862590000032
m is the total web page number searched in Google by using the characteristic words x and y, and is the logf (x), the logf (y) is the hit number searched by using the characteristic words x and y respectively, and f (x and y) is the number of web pages simultaneously appearing by using x and y;
step 2.3: performing K-means + + clustering based on the feature word vectors, and selecting the clustering number K;
step 2.4: feature word set vector FV from servicessiRandomly selecting one as an initial clustering center C1；
Step 2.5: calculating the distance D between each vector and the current nearest cluster center(x)Calculating the probability of each vector becoming the center of the next cluster
Figure FDA0003636862590000033
Determining the next center according to the probability wheel disc;
step 2.6: repeating the step 2.5 until K clustering centers are selected;
step 2.7: FV for each servicesiCalculating the normalized Google distance from the Google distance to K clustering centers, and classifying the Google distance to the class to which the center with the minimum distance belongs;
step 2.8: for each class C(i)Recalculating centers of all the elements in (1);
step 2.9: repeating the step 2.7 and the step 2.8 until the clustering center is not changed any more, and obtaining the web service cluster S(a)＝{s|s∈C(a),a∈[1,m]And m is more than or equal to 1}, and clustering is completed.
4. The method of claim 3 for recommending Web services based on QoS awareness of users and service clusters, wherein: the range of NGD in step 2.2 is NGD (x, y) e [0, ∞).
CN201811394933.8A 2018-11-21 2018-11-21 Web service recommendation method based on QoS (quality of service) perception of user and service clustering Active CN109471982B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811394933.8A CN109471982B (en) 2018-11-21 2018-11-21 Web service recommendation method based on QoS (quality of service) perception of user and service clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811394933.8A CN109471982B (en) 2018-11-21 2018-11-21 Web service recommendation method based on QoS (quality of service) perception of user and service clustering

Publications (2)

Publication Number Publication Date
CN109471982A CN109471982A (en) 2019-03-15
CN109471982B true CN109471982B (en) 2022-06-17

Family

ID=65673117

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811394933.8A Active CN109471982B (en) 2018-11-21 2018-11-21 Web service recommendation method based on QoS (quality of service) perception of user and service clustering

Country Status (1)

Country Link
CN (1) CN109471982B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112417310B (en) * 2019-08-21 2023-10-03 上海掌门科技有限公司 Method for establishing intelligent service index and recommending intelligent service
CN112967102A (en) * 2021-02-04 2021-06-15 江苏警官学院 Method for establishing customer portrait by logistics data
CN112989194B (en) * 2021-03-18 2022-11-29 南京邮电大学 Recommendation method and system integrating user request and service preference of Internet of vehicles
CN113342963B (en) * 2021-04-29 2022-03-04 山东大学 Service recommendation method and system based on transfer learning
CN113497831B (en) * 2021-06-30 2022-10-25 西安交通大学 Content placement method and system based on feedback popularity under mobile edge network
CN117997959B (en) * 2024-04-07 2024-06-04 厦门两万里文化传媒有限公司 Resource intelligent matching method and system based on meta universe

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103793505A (en) * 2014-01-27 2014-05-14 西安理工大学 Network service collaborative filtering method based on user-service characteristics

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8386485B2 (en) * 2008-07-31 2013-02-26 George Mason Intellectual Properties, Inc. Case-based framework for collaborative semantic search

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103793505A (en) * 2014-01-27 2014-05-14 西安理工大学 Network service collaborative filtering method based on user-service characteristics

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Hybrid Collaborative Filtering algorithm for bidirectional Web service recommendation;Jie Cao 等;《Knowledge and Information Systems》;20121010;全文 *
基于协同过滤和QoS预测的Web服务推荐方法研究;黎懋靓;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160615(第 06 期);全文 *

Also Published As

Publication number Publication date
CN109471982A (en) 2019-03-15

Similar Documents

Publication Publication Date Title
CN109471982B (en) Web service recommendation method based on QoS (quality of service) perception of user and service clustering
CN111797321B (en) Personalized knowledge recommendation method and system for different scenes
CN110162706B (en) Personalized recommendation method and system based on interactive data clustering
CN108460101B (en) Point of interest recommended method of the facing position social networks based on geographical location regularization
CN108573041B (en) Probability matrix decomposition recommendation method based on weighted trust relationship
CN107943897B (en) User recommendation method
TW201504829A (en) Method and system for searching images
CN109062962A (en) A kind of gating cycle neural network point of interest recommended method merging Weather information
CN105678590B (en) Cloud model-based topN recommendation method for social network
CN116244513B (en) Random group POI recommendation method, system, equipment and storage medium
CN115712780A (en) Information pushing method and device based on cloud computing and big data
CN111324807A (en) Collaborative filtering recommendation method based on trust degree
CN111475744A (en) Personalized position recommendation method based on ensemble learning
CN109241442A (en) Item recommendation method, readable storage medium storing program for executing and terminal based on predicted value filling
CN110059257B (en) Project recommendation method based on score correction
CN114637920A (en) Object recommendation method and device
CN114564594A (en) Knowledge graph user preference entity recall method based on double-tower model
CN108268877A (en) A kind of method and apparatus for identifying target terminal
CN117371511A (en) Training method, device, equipment and storage medium for image classification model
CN114417166B (en) Continuous interest point recommendation method based on behavior sequence and dynamic social influence
CN110309424A (en) A kind of socialization recommended method based on Rough clustering
CN110825965A (en) Improved collaborative filtering recommendation method based on trust mechanism and time weighting
CN114757391B (en) Network data space design and application method oriented to service quality prediction
Zou et al. An effective collaborative filtering via enhanced similarity and probability interval prediction
CN111460318B (en) Collaborative filtering recommendation method based on explicit and implicit trusts

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant