Mustapha Aznag

Leveraging Formal Concept Analysis with Topic Correlation for Service Clustering and Discovery

With a growing number of web services, discovering services that can match with a users query becomes a challenging task. Its very tedious for a service consumer to select the appropriate one according to her/his needs. In this paper, we propose a non-logic-based matchmaking approach that uses the Correlated Topic Model (CTM) to extract topic from semantic service descriptions and model the correlation between the extracted topics. Based on the topic correlation, service descriptions can be grouped into hierarchical clusters. In our approach, we use the Formal Concept Analysis (FCA) formalism to organize the constructed hierarchical clusters into concept lattices according to their topics. Thus, service discovery may be achieved more easily using the concept lattice. In our approach, topic models are used as efficient dimension reduction techniques, which are able to capture semantic relationships between word-topic and topic-service interpreted in terms of probability distributions. In our experiment, we compared the accuracy of the our hierarchical clustering algorithm with that of a classical hierarchical agglomerative clustering. The comparisons of Precision@n and Normalised Discounted Cumulative Gain (NDCGn) values for our approach, Apache lucene and SAWSDL-MX2 Matchmaker indicate that the method based on CTM presented in this paper outperform all the others matchmakers in terms of ranking of the most relevant services.

Probabilistic Topic Models for Web Services Clustering and Discovery

In Information Retrieval the Probabilistic Topic Models were originally developed and utilized for topic extraction and document modeling. In this paper, we explore several probabilistic topic models: Probabilistic Latent Semantic Analysis (PLSA), Latent Dirichlet Allocation (LDA) and Correlated Topic Model (CTM) to extract latent factors from web service descriptions. These extracted latent factors are then used to group the services into clusters. In our approach, topic models are used as efficient dimension reduction techniques, which are able to capture semantic relationships between word-topic and topic-service interpreted in terms of probability distributions. To address the limitation of keywords-based queries, we represent web service description as a vector space and we introduce a new approach for discovering web services using latent factors. In our experiment, we compared the accuracy of the three probabilistic clustering algorithms (PLSA, LDA and CTM) with that of a classical clustering algorithm. We evaluated also our service discovery approach by calculating the precision (P@n) and normalized discounted cumulative gain (NDCGn). The results show that both approaches based on CTM and LDA perform better than other search methods.

Correlated Topic Model for Web Services Ranking

With the increasing number of published Web services providing similar functionalities, it’s very tedious for a service consumer to make decision to select the appropriate one according to her/his needs. In this paper, we explore several probabilistic topic models: Probabilistic Latent Semantic Analysis (PLSA), Latent Dirichlet Allocation (LDA) and Correlated Topic Model (CTM) to extract latent factors from web service descriptions. In our approach, topic models are used as efficient dimension reduction techniques, which are able to capture semantic relationships between word-topic and topic-service interpreted in terms of probability distributions. To address the limitation of keywords-based queries, we represent web service description as a vector space and we introduce a new approach for discovering and ranking web services using latent factors. In our experiment, we evaluated our Service Discovery and Ranking approach by calculating the precision (P@n) and normalized discounted cumulative gain (NDCGn).

Web Services Discovery and Recommendation Based on Information Extraction and Symbolic Reputation

This paper shows that the problem of web services representation is crucial and analyzes the various factors that influence on it. It presents the traditional representation of web services considering traditional textual descriptions based on the information contained in WSDL files. Unfortunately, textual web services descriptions are dirty and need significant cleaning to keep only useful information. To deal with this problem, we introduce rules based text tagging method, which allows filtering web service description to keep only significant information. A new representation based on such filtered data is then introduced. Many web services have empty descriptions. Also, we consider web services representations based on the WSDL file structure (types, attributes, etc.). Alternatively, we introduce a new representation called symbolic reputation, which is computed from relationships between web services. The impact of the use of these representations on web service discovery and recommendation is studied and discussed in the experimentation using real world web services.

Probabilistic Approach for Diversifying Web Services Discovery and Composition

Due to the increasing number of available web services, discovering the best service that matches a user requirement is still a challenge. In most cases the discovery system returns a set of very similar services and sometimes it is unable to find results for some complex queries. Therefore, integrating web service discovery and composition, taking into account the diversity of discovered results, in a unified way is still a big issue for web services. In this paper, we propose a novel service ranking algorithm for diversifying web services discovery results in order to minimize the redundancy in the search results. This algorithm chooses a set of selected web services based on relevancy, service diversity and service density. We also propose a new method to generate service dependency network using the Formal Concept Analysis (FCA) framework. The generated graph is used to select the composition of discovered web services set. Experimental results show that our method performs better than others baseline approaches.

Semantic Pattern Mining Based Web Service Recommendation

This paper deals with the problem of web service recommendation. We propose a new content-based recommendation system. Its originality comes from the combination of probabilistic topic models and pattern mining to capture the maximal common semantic of sets of services. We define the notion of semantic patterns which are maximal frequent itemsets of topics. In the off-line process, the computation of these patterns is performed by using frequent concept lattices in order to find also the sets of services associated to the semantic patterns. These sets of services are then used to recommend services in the on-line process. We compare the results of the proposed system in terms of precision and normalized discounted cumulative gain with Apache Lucene and SAWSDL-MX2 Matchmaker on real-world data. Our proposition outperforms these two systems.

Multilabel Learning for Automatic Web Services Tagging

Recently, some web services portals and search engines as Biocatalogue and Seekda!, have allowed users to manually annotate Web services using tags. User Tags provide meaningful descriptions of services and allow users to index and organize their contents. Tagging technique is widely used to annotate objects in Web 2.0 applications. In this paper we propose a novel probabilistic topic model (which extends the CorrLDA model - Correspondence Latent Dirichlet Allocation-) to automatically tag web services according to existing manual tags. Our probabilistic topic model is a latent variable model that exploits local correlation labels. Indeed, exploiting label correlations is a challenging and crucial problem especially in multi-label learning context. Moreover, several existing systems can recommend tags for web services based on existing manual tags. In most cases, the manual tags have better quality. We also develop three strategies to automatically recommend the best tags for web services. We also propose, in this paper, WS-Portal; An Enriched Web Services Search Engine which contains 7063 providers, 115 sub-classes of category and 22236 web services crawled from the Internet. In WS-Portal, severals technologies are employed to improve the effectiveness of web service discovery (i.e. web services clustering, tags recommendation, services rating and monitoring). Our experiments are performed out based on real-world web services. The comparisons of Precision@n, Normalised Discounted Cumulative Gain (NDCGn) values for our approach indicate that the method presented in this paper outperforms the method based on the CorrLDA in terms of ranking and quality of generated tags.

Multiple Representations of Web Services: Discovery, Clustering and Recommendation

This paper analyses the well known web servicesrepresentations based on web services descriptions and moregenerally on the content of WSDL files to evaluate theirinterest for discovery, clustering and recommendation tasks.Unfortunately, this analysis shows that these representationsare very basic and do not lead to good results. Therefore,we introduce a new representation called symbolic reputationwhich is computed from relationships between web services.Different implementation issues are discussed and the resultsconsidering real world web services are analysed to determinethe usefulness of the introduced representations for web servicesdiscovery, clustering and recommendation.

Semantic Divergence based Evaluation of Web Service Communities

The number of community detection algorithms is growing continuously adopting a topological based approach to discover optimal subgraphs or communities. In this paper, we propose a new method combining both topology and semantic to evaluate and rank community detection algorithms. To achieve this goal we consider a probabilistic topic based approach to define a new measure called semantic divergence of communities. Combining this measure with others related to prior knowledge, we compute a score for each algorithm to evaluate the effectiveness of its communities and propose a ranking method. We have evaluated our approach considering communities of real web services.

WS-Portal an Enriched Web Services Search Engine

With a growing number of web services, discovering services that can match with a user’s query becomes a big challenging task. It’s very tedious for a service consumer to select the appropriate one according to her/his needs. In this paper, we propose WS-Portal; An Enriched Web Services Search Engine which contains 7063 providers, 115 sub-classes of category and 22236 web services crawled from the Internet. In WS-Portal, severals technologies are employed to improve the effectiveness of web services discovery (i.e. web services clustering, tags recommendation, services rating and monitoring).