CN108629010B - Web service recommendation method based on theme and service combination information - Google Patents
Web service recommendation method based on theme and service combination information Download PDFInfo
- Publication number
- CN108629010B CN108629010B CN201810424947.3A CN201810424947A CN108629010B CN 108629010 B CN108629010 B CN 108629010B CN 201810424947 A CN201810424947 A CN 201810424947A CN 108629010 B CN108629010 B CN 108629010B
- Authority
- CN
- China
- Prior art keywords
- service
- web
- similarity
- services
- web services
- 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
Links
Images
Abstract
The invention discloses a web service recommendation method based on subject and service combination information, which comprises the steps of 1) obtaining historical scoring records of user web services and description information and service combination information of the web services; 2) acquiring a service-theme matrix by using an LDA (latent Dirichlet Allocation) model, and calculating theme similarity between web services according to the service-theme matrix; 3) calculating a service combination similarity matrix between web services; 4) selecting a neighbor set of each web service according to the topic similarity and the service combination similarity matrix; 5) fusing the neighbor set and the similarity matrix with the SVD matrix decomposition model to predict scores; 6) selecting a plurality of services with the highest scores for recommendation according to the prediction scores; the method introduces the consideration of the service subject information and the service combination information, introduces a new dimension for the web service recommendation, improves the recommendation precision, can solve the problem of cold start of part of service recommendation, and has wide market prospect.
Description
Technical Field
The invention relates to the technical field of computer network service, in particular to a web service recommendation method based on theme and service combination information, which is a collaborative filtering method for web service recommendation.
Background
With the rapid development of Service Oriented Architecture (SOA), the internet has begun to emerge with a large number of web services. A Web Service (Web Service) is a Service based on XML and HTTPS, whose communication protocol is mainly based on SOAP, and the description of the Service discovers and obtains metadata of the Service through WSDL and UDDI. A Web service is an application that exposes to the outside an API that can be called via the Web, which can be called via the Web in a programmed way. With the development of Web 2.0 technology, a novel Web service mashup application is gradually popular on the Internet, the mashup application integrates a plurality of external data sources and services, the existing Web services or data sources are utilized to combine the resources to establish a new Web service, and the value of the new service is greater than the simple superposition of the used service combinations. According to the ProrammableWeb statistics of the Web service website, by 9 months of 2017, the number of APIs reaches 14653, and the number of mashups reaches 6259.
In the face of a large number of web services, most users lack sufficient experience or ability to select appropriate services, so that it is urgent to recommend some web services suitable for their own needs to users, a recommendation system is one of effective means for solving the problem of information overload, and collaborative filtering is the most common method in the recommendation system. However, since the historical information of the user on the web service is very sparse, the calculation similarity is not very accurate, and in order to improve the recommendation precision, many current methods consider adding some user characteristics or article characteristics, such as time information and functional topic information, in the traditional recommendation model.
However, at present, due to the rectification and modification of web service websites such as programammablet web and the like in the aspect of privacy protection, time information of a user is more and more difficult to obtain, and for this reason, the recommendation effect is effectively improved by considering the utilization of service combination information and function theme information. The service combination information is the calling relation of mashup application to the atomic web service, most users of the web service are technical developers, when the users like a certain mashup, the users are likely to be interested in the web service API called by the mashup application, meanwhile, the introduction of the service combination information can introduce new dimensionality for recommendation, and web services similar in structure can be recommended for the users.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the defects in the prior art, the invention provides a web service recommendation method based on topic and service combination information.
The technical scheme is as follows: a web service recommendation method based on topic and service combination information comprises the following specific steps:
step 1, acquiring historical preference records of a user for web services and labels, descriptions and service combination information of the web services;
step 2, training by using an LDA model to obtain a service-theme matrix, and calculating theme similarity between web services according to the service-theme matrix;
step 3, calculating the service combination similarity of the web services according to the service combination information among the web services;
step 4, combining the theme similarity and the service combination similarity, and calculating a set of neighbor services of each web service;
step 5, predicting the score of each web service of the user by utilizing an SVD matrix decomposition model in combination with the similarity of the theme and the service combination and the neighbor service set in the step 4;
and 6, selecting a plurality of web services with the highest scores according to the scores of the web services by the users, and recommending the web services to the users.
In step 2 of the invention, text information such as labels and descriptions of web services needs to be preprocessed before LDA model training, wherein the preprocessing process comprises the steps of word segmentation, spelling check, content cleaning, special symbol and stop word removal, word drying and the like, and after preprocessing, a word set S of the web services S is obtained<word1,word2,...>Taking the LDA theme model as an input, outputting the theme distribution of each web service, specifically Si＝{T1,T2,T3…, and the generation formula of the LDA model:
p(w|s)＝p(w|t)*p(t|s)
p (w | s) represents the probability of each word w in the service s description document, p (w | t) represents the probability of each word w under the topic t, and p (t | s) represents the probability of each topic t in the service s description document.
Calculating topic similarity Sim of web services using cosine similarityT(WSi,WSj) The calculation formula is as follows:
wherein WSi,WSjDenotes the ith and jth web services, SiIs the topic distribution vector, S, of the ith servicejIs the topic distribution vector for the jth service,<Si,Sj>represents the inner product of the vector, | SiAnd | is the modulus of the vector.
In step 3 of the present invention, the specific process of calculating the service combination similarity is as follows: extracting service combination information of the web service, specifically, M ═ API1, API2 …, where M represents mashup service M, API1, API2 … is an atomic service called by the mashup service, establishing a service-API call relation matrix, and judging whether the web service is the atomic service or the mashup service, if the web service is between two atomic web services, according to the call relation matrix, calculating a formula as follows:
wherein M isiMashup service set, M, representing the invocation of the ith web servicejRepresenting the mashup service set that calls the jth web service.
If the two web services are two mashups or one mashup and one web service, according to the calling relationship matrix, the calculation formula is as follows:
wherein A isiRepresenting the set of atomic services called by the ith web service, AjRepresenting the atomic service set called by the jth service;
in step 4 of the invention, when the neighbors are selected, k nearest neighbors on the theme and the service combination are respectively selected, the selection principle is that the similarity is larger and closer according to the similarity between the web services, and k is recommended to be set to be 5.
WS-selecting web services using the following constraint objectivesiNeighbor set N (ws)i):
N(wsi)＝{l|l∈Top-k(i),sim(i,l)＞0,i≠l}
Wherein Top-k (i) represents a set of k services with the greatest similarity to the ith web service;
in step 5 of the invention, an SVD prediction model combining service combination information and subject information is solved by adopting a random gradient descent method, the termination condition of the random gradient descent is that the predicted value of a training set is converged, a fixed step number is adopted, the step number is 150, hyper-parameters alpha and beta are used for controlling the proportion of the subject information and the service combination information, and the numerical value is obtained by cross validation.
In step 6 of the method, according to the model learned in step 5, for each user, the score value of the user on each web service is calculated, the score values are sorted from high to low, a plurality of web services with the highest score values are selected and recommended to the user.
Compared with the existing recommendation method, the method mainly has the following advantages: (1) the theme information and the service combination information are combined, so that the recommendation precision is deeper, and the recommendation dimensionality is wider; (2) the problems of sparse data of the cooperative matrix and inaccurate similarity calculation are solved; (3) the partial cold web service can be recommended through the neighbor relation on the service combination information or the subject information; (4) the model is trained by adopting a matrix decomposition and random descent method, so that the training time of the model is shorter.
Drawings
FIG. 1 is a body frame diagram of the method of the present invention;
FIG. 2 is a flow chart of calculating topic similarity;
FIG. 3 is a flow chart of computing service portfolio similarities;
FIG. 4 is a graph comparing the RMSE index of the present invention with other methods;
FIG. 5 is a graph comparing the MAE index of the present invention with other methods.
Detailed Description
The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will occur to those skilled in the art upon reading the present disclosure and fall within the scope of the appended claims.
As shown in fig. 1, the method for recommending web services based on topic and service combination information mainly includes six steps:
step 1, acquiring historical preference records of a user for web services and labels, descriptions and service combination information of the web services;
step 2, training by using an LDA model to obtain a service-theme matrix, and calculating theme similarity between web services according to the service-theme matrix;
step 3, calculating the service combination similarity of the web services according to the service combination information among the web services;
step 4, combining the theme similarity and the service combination similarity, and calculating a set of neighbor services of each web service;
step 5, predicting the score of each web service of the user by utilizing an SVD matrix decomposition model in combination with the similarity of the theme and the service combination and the neighbor service set in the step 4;
and 6, selecting a plurality of web services with the highest scores according to the scores of the web services by the users, and recommending the web services to the users.
In step 1, the user may crawl (http:// www.programmableweb.com) the historical preference record of the web service, the function description information of the web service, and the service combination information from the programable web site using the API provided by the website. The data we used in this example are shown in table 1:
TABLE 1
| Number of users | 510 |
| Atomic web services number | 2713 |
| Mashup number | 2100 |
| Average mashup contains API number | 2.2 |
| Data sparsity ratio | 99.13％ |
In step 2, an LDA model is used for calculating topic similarity of the web service, the processing flow is as shown in figure 2, firstly, text preprocessing is required, the preprocessing process comprises the steps of word segmentation, spell checking, content cleaning, special symbol and stop word removal, word drying and the like, and after preprocessing, a word set S of the web service S is obtained<word1,word2,...>Taking the LDA theme model as an input, outputting the theme distribution of each web service, specifically Si＝{T1,T2,T3…, taking the functional description of the two web services Google Buzz and Desktop music player of table 2 as an example:
TABLE 2
After text pre-processing, the result is shown in table 3, where each web service consists of stems of several words:
TABLE 3
Training the preprocessed text content by using an LDA model, wherein a generation formula of the LDA model is as follows:
p(w|s)＝p(w|t)*p(t|s)
p (w | s) represents the probability of each word w in the service s description document, p (w | t) represents the probability of each word w under the topic t, and p (t | s) represents the probability of each topic t in the service s description document. Here, the number of topics is set to 10, and the service-topic distribution of each web service is obtained, and the service-topic distributions of the above Google Buzz and Desktop music player are as follows:
S1＝<0.00240964，0.17108434，0.60481928，…,0.21927711，0.00240964>
S2＝<0.008,0.008,0.728,…,0.248,0.008>
for the above service-topic distribution, the topic similarity of S1 and S2 is calculated using cosine similarity, and the calculation formula is as follows:
wherein<S1,S2>Represents a vector S1And S2Inner product of, | S1L is a vector S1The similarity of Google Buzz and Desktop music player on the subject is 0.2891145.
In step 3, the service combination similarity between web services is calculated through the service combination relationship of the crawled web services, and the processing flow is shown in fig. 3, first, the combination relationship of the web services needs to be obtained, and the following web services are obtained as an example:
TABLE 4
We build a service-api call relationship matrix as follows:
| Google Maps | Youtube | MTV | Google Search | |
| S1 | 1 | 1 | 0 | 0 |
| S2 | 0 | 1 | 1 | 0 |
| S3 | 0 | 0 | 1 | 1 |
when we need to require the service combination similarity of S1 and S2, we use the following calculation formula according to the calling relationship matrix:
wherein A is1Atomic service set, A, representing S1 calls2Representing the atomic service set called by S2.
|A1∩A2|＝1,|A1∪A2I | ═ 3, so the similarity Sim of S1 and S2 was calculatedC(S1,S2) Is 1/3.
When we need to calculate the service combination similarity of two atomic services of Google Maps and Youtube, we use the following calculation formula according to the call relation matrix:
wherein S4 denotes Google Maps, S5 denotes Youtube, M4 denotes a service set calling S4, and M5 denotes a service set | M calling S54∩M5|＝1，|M4∪M5Therefore, the combined similarity of S4 and S5 is calculated to be 1/2.
In step 4, the similarity of each web service is sorted from large to small, a plurality of most similar services are selected as the neighbor set of the web service, and according to step 2 and step 3, the neighbor set TN of each web service s on the subject level can be obtained respectively(s)And a neighbor set CN on the service composition level(s)。
In step 5, the similarity matrix obtained in steps 2,3 and 4 and the neighbor set are fused into an SVD matrix decomposition model, and the model is as follows:
wherein r isusRepresents the predicted score of user u for service s, μ represents the global mean deviation, buIndicates a user deviation, bs、bsIndicating a service deviation, puImplicit vector at user end, qs、ql、qkIs implicit vector of server, alpha, beta are weight coefficient, omegasl,gskThe topic and service combination similarity of the two services respectively.
The model can be solved using a stochastic gradient descent method by minimizing the loss function L as follows:
wherein M represents that a user history scoring information set (u, s) belongs to M, R represents that the scoring of the service s by the user u belongs to the set MusThe value is the actual score of the user, a regular term is added into the loss function to prevent overfitting, and lambda is a regular term coefficient.
The input is historical scoring record of the user, a theme similarity matrix, a combined similarity matrix and a neighbor set, and the output is bu,bsP, q, etc., so that the user's score for any service can be predicted by the model.
In step 6, for any user u, the scores of the user for all web services are predicted according to the prediction model in step 5, the predicted scores are ranked from high to low, and a plurality of services with the highest scores are selected and recommended to the user.
In this example, the data set was randomly selected to be 90% training set and 10% testing set for experiments using RMSE (root mean square error) and MAE (mean absolute error) as evaluation indexes, which are defined as follows:
the experimental results are shown in fig. 4 and 5, and it can be seen from the graphs that the RMSE of the present invention is improved by 3.6% compared with the conventional SVD matrix decomposition method, and meanwhile, compared with the method introducing the timing characteristic, the present invention has the RMSE accuracy which is not lost in the timing model, while on the MAE, the method is improved by 8.95% compared with the conventional matrix decomposition method and is improved by 7.6% compared with the timing characteristic method.
According to the invention, the theme information and the service combination information are merged into the web service recommendation, so that the web service recommendation method has new dimensionality, and the defect that the collaborative matrix is too sparse in the traditional recommendation method can be overcome.
Claims (5)
1. A web service recommendation method based on topic and service combination information is characterized by comprising the following steps:
step 1, acquiring historical preference records of a user for web services and labels, descriptions and service combination information of the web services;
step 2, training by using an LDA model to obtain a service-theme matrix, and calculating theme similarity between web services according to the service-theme matrix;
step 3, calculating the service combination similarity of the web services according to the service combination information among the web services;
step 4, calculating a neighbor service set of each web service by combining the theme similarity and the service combination similarity;
step 5, combining the theme similarity and the service combination similarity and the neighbor service set in the step 4, and predicting the score of the user for each web service by using an SVD matrix decomposition model;
and 6, selecting a plurality of web services with the highest scores according to the scores of the web services by the users, and recommending the web services to the users.
2. The method as claimed in claim 1, wherein in the step 2, the LDA topic model is inputted as a description document after text preprocessing of each web service, the LDA topic model is outputted as a service-topic distribution and a topic-word frequency distribution of each web service, and the cosine similarity is used to calculate the topic similarity between web services
The calculation formula is as follows:
wherein the content of the first and second substances,
representing the ith web service and the jth web service,
is the topic distribution vector for the ith service,
is the topic distribution vector for the jth service,
the inner product of the vector is represented as,
is the modulus of the vector.
3. The method for recommending web services based on topic and service composition information as claimed in claim 1, wherein in step 3, the specific process for calculating the similarity of service composition is: extracting service combination information of the web service, specifically M = { API1, API2 … }, wherein M represents mashup service, API1 and API2 … are atomic services called by the mashup service, establishing a service-API call relation matrix, judging whether the web service is the atomic service or the mashup service, and if the two web services are the two atomic services, calculating the following formula according to the call relation matrix:
wherein, the first and second guide rollers are arranged in a row,
representing the ith and jth web services, MiMashup service set, M, representing the invocation of the ith web servicejMashup service set representing j-th web service invocationCombining;
if the two web services are two mashup services or one mashup service and one atomic service, according to the calling relationship matrix, the calculation formula is as follows:
wherein A isiRepresenting the set of atomic services called by the ith web service, AjRepresenting the set of atomic services invoked by the jth web service.
4. The method as claimed in claim 1, wherein when selecting the neighbors, the nearest k neighbors on the topic and the service combination are selected respectively, and the selection principle is that according to the similarity of the web service, the greater the similarity is, the closer the similarity is:
WS-selecting web services using the following constraint objectivesiNeighbor set
：
Wherein
Representing the set of k services with the greatest similarity to the ith web service.
5. The method for recommending web services based on topic and service combination information according to claim 1, wherein in step 5, score prediction is performed by using SVD matrix decomposition method, wherein the prediction model is as follows:
wherein
Representing the rating of the service s by the user u,
the average deviation over the whole world is indicated,
a deviation is indicated for the user by a representation,
the deviation of the service is indicated and,
for the implicit vector of the user terminal,
in order to be an implicit vector of the server,
in order to be the weight coefficient,
topic similarity and service combination similarity between two services s and l;
a neighbor set on the subject level for the service s;
a neighbor set of the service s on the service combination level;
the prediction model is solved by minimizing a loss function by using a random gradient descent method, wherein the loss function L is as follows:
wherein M represents a user historical scoring information set,
indicating that the user u scores the service s to belong to the set M,
the value is the actual score of the user, a regular term is added into the loss function to prevent overfitting, and lambda is a regular term coefficient.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810424947.3A CN108629010B (en) | 2018-05-07 | 2018-05-07 | Web service recommendation method based on theme and service combination information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810424947.3A CN108629010B (en) | 2018-05-07 | 2018-05-07 | Web service recommendation method based on theme and service combination information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108629010A CN108629010A (en) | 2018-10-09 |
| CN108629010B true CN108629010B (en) | 2022-03-18 |
Family
ID=63695706
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810424947.3A Active CN108629010B (en) | 2018-05-07 | 2018-05-07 | Web service recommendation method based on theme and service combination information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108629010B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109344332A (en) * | 2018-10-31 | 2019-02-15 | 郑州云海信息技术有限公司 | A kind of Web API combined recommendation method, apparatus and electronic equipment |
| CN110209954B (en) * | 2019-06-03 | 2022-08-26 | 南京邮电大学 | Group recommendation method based on LDA topic model and deep learning |
| CN110347401B (en) * | 2019-06-18 | 2021-03-16 | 西安交通大学 | API Framework service discovery method based on semantic similarity |
| CN111539784B (en) * | 2020-04-10 | 2023-05-26 | 上海交通大学 | Service package recommendation method and system based on combined semantic learning |
| CN112214335B (en) * | 2020-10-13 | 2023-12-01 | 重庆工业大数据创新中心有限公司 | Web service discovery method based on knowledge graph and similarity network |
| CN113342963B (en) * | 2021-04-29 | 2022-03-04 | 山东大学 | Service recommendation method and system based on transfer learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102231757A (en) * | 2011-06-29 | 2011-11-02 | 浙江大学 | On-line service combination recommendation system and recommendation method thereof |
| US20120259859A1 (en) * | 2009-12-31 | 2012-10-11 | Taggy, Inc. | Method for recommending best information in real time by appropriately obtaining gist of web page and user's preference |
| CN103761254A (en) * | 2013-12-26 | 2014-04-30 | 清华大学 | Method for matching and recommending service themes in various fields |
| US20140280946A1 (en) * | 2013-03-12 | 2014-09-18 | Xerox Corporation | System and process to recommend cloud service cloud configuration based on service similarity |
-
2018
- 2018-05-07 CN CN201810424947.3A patent/CN108629010B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120259859A1 (en) * | 2009-12-31 | 2012-10-11 | Taggy, Inc. | Method for recommending best information in real time by appropriately obtaining gist of web page and user's preference |
| CN102231757A (en) * | 2011-06-29 | 2011-11-02 | 浙江大学 | On-line service combination recommendation system and recommendation method thereof |
| US20140280946A1 (en) * | 2013-03-12 | 2014-09-18 | Xerox Corporation | System and process to recommend cloud service cloud configuration based on service similarity |
| CN103761254A (en) * | 2013-12-26 | 2014-04-30 | 清华大学 | Method for matching and recommending service themes in various fields |
Non-Patent Citations (2)
| Title |
|---|
| Web Service Recommendation with Ontology-based Similarity Measure;Haoxiang Xia等;《Second International Conference on Innovative Computing, Informatio and Control (ICICIC 2007)》;20080114;第1-4页 * |
| 基于服务网络的服务组合推荐方法;潘伟丰等;《系统工程理论与实践》;20140625;第34卷;第131-142页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108629010A (en) | 2018-10-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108629010B (en) | Web service recommendation method based on theme and service combination information | |
| US11593894B2 (en) | Interest recommendation method, computer device, and storage medium | |
| US11347752B2 (en) | Personalized user feed based on monitored activities | |
| KR101888919B1 (en) | Method and Apparatus for Marketing of Advertisement Based on User Influence | |
| Chen et al. | Location-aware personalized news recommendation with deep semantic analysis | |
| US20190197416A1 (en) | Information recommendation method, apparatus, and server based on user data in an online forum | |
| US20170236073A1 (en) | Machine learned candidate selection on inverted indices | |
| Cleger-Tamayo et al. | Top-N news recommendations in digital newspapers | |
| US20150287091A1 (en) | User similarity groups for on-line marketing | |
| CN112434151A (en) | Patent recommendation method and device, computer equipment and storage medium | |
| CN106682686A (en) | User gender prediction method based on mobile phone Internet-surfing behavior | |
| CN107291755B (en) | Terminal pushing method and device | |
| Piao et al. | Analyzing aggregated semantics-enabled user modeling on Google+ and Twitter for personalized link recommendations | |
| CN110334356A (en) | Article matter method for determination of amount, article screening technique and corresponding device | |
| Natarajan et al. | Recommending news based on hybrid user profile, popularity, trends, and location | |
| Hsieh et al. | A keyword-aware recommender system using implicit feedback on Hadoop | |
| CN113869931A (en) | Advertisement putting strategy determining method and device, computer equipment and storage medium | |
| Sajeev et al. | Effective web personalization system based on time and semantic relatedness | |
| Chen et al. | Context-aware ensemble of multifaceted factorization models for recommendation prediction in social networks | |
| Kim et al. | Topic-Driven SocialRank: Personalized search result ranking by identifying similar, credible users in a social network | |
| Cui et al. | Personalized microblog recommendation using sentimental features | |
| Chen et al. | User intent-oriented video QoE with emotion detection networking | |
| Wei et al. | Using network flows to identify users sharing extremist content on social media | |
| Aliannejadi et al. | Venue suggestion using social-centric scores | |
| Xiao et al. | Predicting audience gender in online content‐sharing social networks |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 210008 No. 22, Hankou Road, Gulou District, Jiangsu, Nanjing Applicant after: NANJING University Address before: 210046 Xianlin Avenue 163, Qixia District, Nanjing City, Jiangsu Province Applicant before: NANJING University |