Nguyen Ngoc Chan

A recommender system based on historical usage data for web service discovery

The tremendous growth in the amount of available web services impulses many researchers on proposing recommender systems to help users discover services. Most of the proposed solutions analyzed query strings and web service descriptions to generate recommendations. However, these text-based recommendation approaches depend mainly on user’s perspective, languages, and notations, which easily decrease recommendation’s efficiency. In this paper, we present an approach in which we take into account historical usage data instead of the text-based analysis. We apply collaborative filtering technique on user’s interactions. We propose and implement four algorithms to validate our approach. We also provide evaluation methods based on the precision and recall in order to assert the efficiency of our algorithms.

Assisting business process design by activity neighborhood context matching

Speeding up the business process design phase is a crucial challenge in recent years. Some solutions, such as defining and using reference process models or searching similar processes to a working one, can facilitate the designers work. However, recommending the whole process can make the designer confused, especially in case of large-size business processes. In this paper, we introduce the concept of activity neighborhood context in order to propose an approach that fasten the design phase regardless the size of business process. Concretely, we recommend the designer the activities that are close to the designing process from existing business processes. We evaluate our approach on a large collection of public business processes. Experimental results show that our approach is feasible and efficient.

A Web Service Recommender System Using Vector Space Model and Latent Semantic Indexing

The tremendous growth in the amount of available web services (WS) impulses many researchers on proposing recommender systems to help users discover services. Most of the proposed solutions analyzed query strings and web service descriptions to generate recommendations. However, text based recommendations approaches depend mainly on users perspective, languages and notations which easily decrease recommendations efficiency. In this paper, we propose to take into account users behaviors instead of text based analysis. We apply collaborative filtering technique on users interactions. We propose and implement two algorithms based on Vector Space Model and Latent Semantic Indexing to validate our approach. We also provide evaluation methods with different datasets in order to compare and assert the efficiency of our two algorithms.

Context-Based Service Recommendation for Assisting Business Process Design

The WS-BPEL provides a standard for business processes abstraction and execution, in which, the business processes abstraction is the key step for the completeness and success of business processes. The business processes abstraction includes the behavior and interactions between services which are sketched out by business processes designers. The current business process design is labor-intensive and time consuming, especially when it is required to be detailed to ensure the success of the business execution. In this paper, we propose an approach that helps the business process designers facilitate the design step by providing them a list of related services to the current designed model. We propose to capture the requested service’s composition context specified through the process fragment surrounding it and recommend the services whose composition context in existing designed service compositions best match the given fragment context. Provided experimental evaluations in this paper show that our approach is efficient in realistic situations.

Collaborative filtering technique for web service recommendation based on user-operation combination

The tremendous growth in the amount of available web services impulses many researchers on proposing recommender systems to help users discover services. Most of the proposed solutions analyzed query strings and web service descriptions to generate recommendations. However, these text-based recommendation approaches depend mainly on users perspective, languages and notations which easily decrease recommendations efficiency. In this paper, we present our approach in which we take into account historical usage data instead of the text based analysis. We apply collaborative filtering technique on users interactions. We propose and implement three algorithms based on Vector Space Model to validate our approach. We also provide evaluation methods based on the precision, recall and root mean square error in order to compare and assert the efficiency of our algorithms.

An Automated Approach for Assisting the Design of Configurable Process Models

With the intention of design by reuse, configurable process models provide a way to model variability in reference models that need to be configured according to specific needs. The design of configurable process models is a well known complex and error-prone task. Thus, many approaches have been proposed to automate their design by merging existing process models into configurable reference models. However, the complexity introduced by such approaches remains an open issue. The designer ends up with one model that integrates a family of process variants making the process design and update a complex task. In this work, we propose to assist the design of configurable process models with configurable process fragments. Concretely, we present an algorithm for extracting, clustering and merging process fragments around a particular activity to construct a configurable fragment. The approach has been implemented as an extension of the Signavio Process Editor and evaluated against a large collection of process models. Experimental results show that our approach is efficient and produces comprehensible configurable fragments.

Assisting Business Process Design with Configurable Process Fragments

In recent years, many approaches have been proposed to facilitate business process design. They attempted to measure the similarity between business processes, merge business process models, mine event logs or recommend activities. In this paper, we present a merging approach that also aims at facilitating business process design. However, instead of merging business process models, we merge process fragments around a particular activity to construct a consolidated fragment for each activity. This consolidated fragment is presented as a configurable sub-process which allows process designers to overview the interactions of an activity and configure them to create business process variants according to particular requirements. The approach has been implemented as an application and tested against a large collection of business process models taken from different domains. Experimental results show that our approach produces concise and efficient configurable fragments.

Mining Event Logs to Assist the Development of Executable Process Variants

Developing process variants has been proven as a principle task to flexibly adapt a business process model to different markets. Contemporary research on variant development has focused on conceptual process models. However, process models do not always exist, even when process logs are available in information systems. Moreover, process logs are often more detailed than process models and reflect more closely to the behavior of the process. In this paper, we propose an activity recommendation approach that takes into account process logs for assisting the development of executable process variants. To this end, we define a notion of neighborhood context for each activity based on logs, which captures order constraints between activities with their occurrence frequency. The similarity of the neighborhood context between activities provides us then with a basis to recommend activities during the process of creating a new process model. The approach has been implemented as a plug-in for ProM. Furthermore, we conducted experiments on a large collection of process logs. The results indicate that our approach is feasible and applicable in real use cases.

Web Services Recommendation Based on User's Behavior

The tremendous growth in the amount of available web services (WS) impulses many researchers on proposing recommender systems to help users discover services. Most of the proposed solutions analyzed query strings and web service descriptions to generate recommendations. However, text based recommendations approaches depend mainly on users perspective, languages and notations which easily decrease recommendations efficiency. Moreover, new published web services often have lower priorities to be selected for recommendations.\\In this paper, we propose to take into account users behaviors instead of text based analysis. We apply collaborative filtering technique on users interactions. We propose and implement three algorithms (user-based, operation-based and priority-based) to validate our approach. We also provide evaluation methods which indicate that our approach produces high quality recommendations in case users have stable behavior.

Learning Resource Recommendation: An Orchestration of Content-Based Filtering, Word Semantic Similarity and Page Ranking

Technologies supporting online education have been abundantly developed recent years. Many repositories of digital learning resources have been set up and many recommendation approaches have been proposed to facilitate the consummation of learning resources. In this paper, we present an approach that combines three recommendation technologies: content-based filtering, word semantic similarity and page ranking to make resource recommendations. Content-based filtering is applied to filter syntactically learning resources that are similar to user profile. Word semantic similarity is applied to consolidate the content-based filtering with word semantic meanings. Page ranking is applied to identify the importance of each resource according to its relations to others. Finally, a hybrid approach that orchestrates these techniques has been proposed. We performed several experiments on a public learning resource dataset. Results on similarity values, coverage of recommendations and computation time show that our approach is feasible.