An intelligent e-learning scenario for knowledge retrieval

Antonio Martín Departamento de Tecnología Electrónica

Seville University Seville, Spain

toni@us.es

Carlos León Departamento de Tecnología Electrónica

Seville University Seville, Spain cleon@us.es

Abstract—— E-Learning is a critical support mechanism for educational institutions to grow the performance of their students, teachers, as well as useful for organizations to enhance the performance of their employees. The efficient retrieval of learning knowledge is a critical support mechanism for educational institutions to enhance the skills of their students and at the same time useful for learning process. Teachers and students face many difficulties when working with knowledge retrieval in education. Studies show that still it demands more effective approach. Semantic and Artificial Intelligence represent potential technologies for realizing e-Learning requirements. The convergence of e-Learning and digital libraries creates challenge to be solved not only at storage resources level but also at the knowledge retrieval level. Nowadays availability of infrastructure, flexibility of time, learning resources and their means of sharing has increased adaptability of Digital Library to learn and attain knowledge to a great extent. The objective of our study is to present an e-learning management system for Digital Library in Seville University that provides a general platform for learning environment. The idea is investigated from a search perspective possible intelligent infrastructures form constructing decentralized digital libraries where a global semantic schema exists. We suggest a conceptual architecture for a semantic and intelligent search engine. This project is a collaborative effort that proposes a new form of interaction between engineering students and E-learning platform, where the latter is adapted to individuals and their surroundings. We propose a comprehensive approach for discovering information objects in large digital collections based on analysis of recorded semantic metadata and the application of Artificial Intelligent technologies.

E-learning systems, Ontology, Retrieval, Case-based reasoning, Digital Library, Knowledge Management, Intelligent Agents

I. INTRODUCTION

The current Digital Library (DL) is a powerful tool for research and education, but its utility is hindered by the failure of the user can find easily the reputable E-Learning sources. In the traditional search engines the learning knowledge is treated as an ordinary database that manages the contents and positions. An effective retrieval mechanism is needed. The traditional retrieval and management methods for Learning Objects (LOs) include alphabetical indexing and keyword-based searching [1]. Other methods include merely searching on surface features like title, location, publication date, document number [2]. None of these provides a satisfactory

solution because they return a lot of irrelevant cases or missing the relevant cases.

Despite large investments and efforts have been made, there are still a lot of unsolved problems [3]. There are a lot of researches on applying these new technologies into current Digital Libraries information retrieval systems, but no research addresses the semantic and intelligent artificial issues from the whole life cycle and architecture point of view. Although search engines have developed increasingly effective, information overload obstructs precise searches. Our work differs from related projects in that we build an ontology-based contextual profiles and we introduce an approaches used metadata-based in ontology search and expert systems [4]. Case-based retrieval, an application of ontologies, could be an appropriate solution, because of the nature of use of LOs and users opinions.

We study improving the efficiency of search methods to search a distributed data space like a Digital Library. The objective has focused on creating technologically complex environments in Learning and Teaching, Education, in the Digital Library domain. We presented an intelligent approach for contextualize a search engine. It incorporates semantic Web and artificial intelligent technologies to enable not only precise location of digital library resources but also the automatic or semi-automatic learning [5]. By using E-learning classification knowledge and the given class lattice, the search engine responds to user queries with hierarchically structured navigable results, instead of a conventional flat ranked document list, which greatly aids users in locating information from numerous, diversified LOs. For this reason we are improving representation by incorporating more metadata from within the information. We focus our discussion on case indexing and retrieval strategies and provide a perception of the technical aspects of the application. Our approach for realizing content based search and retrieval information implies the application of the Case-Based Reasoning (CBR) technology [6].

This paper is organized as fallows. The first section provides a general overview about our prototype architecture. The second section focuses on the Ontology design process. Then we summarize its main components and describe how can interact Intelligent Artificial and Semantic Web to enhancement a search engine. Next we study the CBR

978-1-4673-1456-5/12/$31.00 ©2012 IEEE Page 165

framework jColibri and its features for implementing the reasoning process over ontologies [7]. Finally we present the results of our ongoing work on the adaptation of the framework and we outline the future works.

II. ONTOFAMA ARCHITECTURE

In order to support semantic retrieval knowledge in a digital library, we develop an E-Learning model for innovative intelligent retrieval LOs, which provides effective knowledge management. We develop a prototype named OntoFAMA based on ontologies and expert systems. The architecture of our system is shown in figure 1, which mainly includes three parts: ontology knowledge base, the search engine, and the intelligent user interface.

Figure 1. OntoFAMA Architecture.

The OntoFAMA platform is an environment, in which software agent can be executed to retrieve LOs, and which is wrapped by a Web service. Agents are intended to assist learners with the focused search for LOs, according to the specifications they made. The search parameters of an agent, the start of a search, or the access to the list of retrieved LOs, for example, can be controlled by invoking appropriate Web service operations which extract metadata from Los

OntoFAMA is a system which consists in a database of learning experiences, different situations in general, that are thought at a higher abstract level, the purpose being to manage diverse knowledge, in a controlled and organized manner. It allows persisting knowledge by categorizing it and by filtering it, based on its relevance. The stored learning knowledge can be modified, adapted to new situations; it can be used as a set of past experiences that can be used for providing solutions to new encountered situations, by relating the new experiences to past ones, which exist in the database. Ontology will be considered as knowledge structure that will identify the concepts, property of concept, resources, and relationships among them to enable share and reuse of LOs that are needed to acquire learning knowledge in a specific search domain. The metadata descriptions of the resources and LOs (cases) are abstracted from the details of their physical representation and are stored in the case base [8].

OntoFAMA uses Case-Based Reasoning (CBR) engine. CBR can provide OLs for new situations by comparing them with previous similar encountered situations [9]. Case Base has a memory organization interface that assumes that whole case-base can be read into memory for the CBR to work with it. The knowledge acquisition process automatically learns semantic classification information from different LOs. The CBR

classifier applies learned taxonomy to classify newly collected knowledge from the information sources in the DL.

The acceptability of a system depends to a great extent on the quality of this user interface component [10]. We have implemented a new interface who allows retrieving cases enough to satisfy a query. In our system the user interacts with the system to fill in the gaps to retrieve the right cases. The interfaces provides for browsing, searching and facilitating Web contents and services. It consists of one user profile, consumer search agent components and bring together a variety of necessary information from different user’s resources. The objective of profile intelligence has focused on creating of user profiles: Staff, Alumni, Administrator, and Visitor. The user interface helps to user to build a particular profile that contains his interest search areas in the digital library domain.

III. CASE-BASED REASONING

CBR is widely discussed in the literature as a technology for building information systems to support knowledge management, where metadata descriptions for characterizing knowledge items are used. In our CBR application, LOs are described by metadata concerning desired characteristics of a Library resource, and the result to the search is a pointer to a learning resource described by metadata [11]. These characterizations are called cases and are stored in a case base. CBR case data could be considered as a portion of the knowledge (metadata) about an OntoFama object. Every case contains both a LOs pointer and reusable learning resource description used for similarity assessment:

- A description of a framework learning resource. The possible LOs described by means of framework instantiation actions. These goals will be formally described in terms of framework domain taxonomy and they will be used for indexing cases.

- Solution. Additional information justifies these steps. Our experience developing has shown that execution graphs are a good technique to represent the list of actions that user should do to reach a LO, so they will be used to represent the solutions in our simple cases.

The development of a quite simple Case-Based Reasoning application already involves a number of steps, such as collecting case and background knowledge, modeling a suitable case representation, defining an accurate similarity measure, implementing retrieval functionality, and implementing user interfaces. Compared with other AI approaches, CBR allows to reduce the effort required for knowledge acquisition and representation significantly, which is certainly one of the major reasons for the commercial success of CBR applications. Nevertheless, implementing a CBR application from scratch remains a time consuming software engineering process and requires a lot of specific experience beyond pure programming skills.

All the cases that have the same structure form a case type. Each case type has one or more similarity functions associated. They are used in order to describe the way a new case is compared with existing cases and to establish which the most similar cases are, figure 2.

978-1-4673-1456-5/12/$31.00 ©2012 IEEE Page 166

Figure 2. The case-based reasoning-cycle

The features that OntoFAMA CBR offers are the

following:

Retrieving the past education experience which is the most similar with a new experience.

Manipulate already stored learning experiences in order to allow knowledge to adapt.

Specify the way by which similar teaching experiences are compared.

Apply a CBR cycle that allows to process information based on Case-Based Reasoning principles.

For this purpose we can used a Case-Based Reasoning (CBR) shell, software that can be utilized to develop several applications that require cased-based reasoning methodology. In this study we used the CBR object-oriented framework development environments JColibri. This framework work as open software development environment and facilitate the reuse of their design as well as implementations. Our motivation for choosing this framework is based on a comparative analysis between it and other frameworks, designed to facilitate the development of CBR applications.

IV. ONTOLOGY DESIGN AND DEVELOMENT INTERFACE

We need a vocabulary of concepts, resources and services for our information system described in the scenario requires definitions about the relationships between objects of discourse and their attributes [12]. Developing a knowledge representation model for the intelligent retrieval of learning objects can greatly facilitate the users’ ability to correctly access tacit legal knowledge, which is an important goal of knowledge learning management. To contribute to very good representation of knowledge, OntoFAMA uses ontology as a data model. Ontology is used to specify knowledge domain, representing a well defined vocabulary which describes E-Learning objects and relations between them and user profiles. A grammar uses the terms from the vocabulary, being capable to express meanings, for the specified domain. Using the ontology concept, OntoFAMA has the ability to perform reasoning over the learning knowledge, being able to infer new knowledge, based on the one that it already has.

By using ontologies in the learning knowledge classification and the given class lattice, the search engine responds to user queries with hierarchically structured navigable results, instead of a conventional flat ranked document list, which greatly aids users in locating information from numerous, diversified LOs. We integrated three essential sources to the system: learning electronic resources, library catalogue and personal Data Base. The W3C defines standards that can be used to design an ontology [13]. We wrote the description of these classes and the properties in RDF semantic markup language. We choose Protégé as our ontology editor, which supports knowledge acquisition and knowledge base development [14]. It is a powerful development and knowledge-modeling tool with an open architecture. Protégé uses OWL and RDF as ontology language to establish semantic relations [15].

Ontology classes and individual attributes are taken and translated in a case structure. Cases are grouped by using a so called case structure. The case structure, like cases as well, has an XML format by which it specifies how information is structured [16]. By the aid of the OWL (Web Ontology Language) and using the powerful semantic web framework called JENA, information described using ontology can contribute to create case structures. OntoFAMA project contains a collection of codes, visualization tools, computing resources, and data sets distributed across the grids, for which we have developed a well-defined ontology using RDF language. RDF is used to define the structure of the metadata describing digital library resources. Our ontology can be regarded as quaternion OntoSearch:={profile, collection, source) where profiles represent the user kinds, resources contains all the services and resources of the digital library and matter cover the different information sources: catalogue, history fond, intranet, Web, etc.

In order to realize ontology-based intelligent retrieval, we need to build case base of knowledge with inheritance structure. The ontology and its sub-classes are established according to the taxonomies profile, as shown in figure 3. This shows the high level classification of classes to group together OntoFAMA resources as well as things that are related with these resources.

Figure 3. Class hierarchy for the OntoFama ontology

As shown the figure, profile ontology includes several

attributes like E-learning_Resources, Digital_Collections, Catalogue, Science_Resources, etc. After ontology is established, we need to add enough initial instances and item

978-1-4673-1456-5/12/$31.00 ©2012 IEEE Page 167

instances to knowledge base. For this purpose we have followed next steps. First we choose a certain item, and create a blank instance for item. Then the domain expert, in this case the librarian fills blank units of instance according the domain knowledge [17].

V. CONVERSATION INTERFACE

A more tangible goal of the work is the development of a specialized intelligent search engine, configurable to be adapted to different profile users. The engine must be instantiated to run in the online DL context, and must offer advanced search capabilities [18]. For the distributed retrieval of learning resources, we use profile users, which are used for personalized searches according to user specifications. The profile agent is an environment, in which software agents can be executed to retrieve E-learning resources, and which is wrapped by a Web service. Profile agents assist to learners with the search, according to the specifications they made.

We present a way for personalized retrieval in the e-learning search engine, that takes advantage of semantic techniques to represent the learning knowledge and the user/learner profiles as ontologies, and ranks search queries based on how the contained terms map to the ontology system. As we have seen in previous sections our system has a graphical user interface for determining initial user requirements in search. Rather than building static user profiles, contextual systems try to adapt to the user’s current search. OntoFAMA monitors user's tasks, anticipates search-based information needs, and proactively provide users with relevant information. This configuration contains the user requirements most typically described the relative needs, tasks, and goals of the user for an individual search, figure 4.

Figure 4. Semantic process to obtain information

The process of using precedents for making arguments and giving learning opinions employs the tacit knowledge contained in the precedents. Past reusable LOs searches can be seen as sources of new object searches because they contain the tacit knowledge about how query terms were once applied in a specific user profile. The purpose of the study was the implementation of e-learning management system from different user perspectives.

In an intelligence profile setting, people are surrounded by intelligent interfaces merged, thus creating a computing-capable environment with intelligent communication and processing available to the user by means of a simple, natural, and effortless human-system interaction. The engineering student enters query commands and the system asks questions

during the inference process. Besides, the user will be able to solve new searches for which he has not been instructed, because the user profiles what he has learnt during the previous searchers [19]. We have developed a graphical selection interface as illustrated in figure 5. The easiest to implement interfaces communicate with the user through a scrolling dialog.

Figure 5.User Profiles, Graphical User interface

The user inputs the keywords in the user profile interface.

Suppose the user is looking for some resource about “Computer Science electronic resource” in the library digital domain of Seville. Required learning resources should contain some knowledge about “Computer Science” and related issues. By using E-learnig classification knowledge and the given class lattice, the search engine responds to user queries with hierarchically structured navigable results, instead of a conventional flat ranked document list, which greatly aids users in locating information from numerous, diversified LOs.

The perform search examines the subclasses of the designate class in the hierarchical tree, calculate the category importance and sort the classes based on the ranking in descending order. The search engine retrieve objects in the designate class associated with the query terms, calculate each object’s relevance significance, and sort these according to the ranking in descending order. After searching, some resources are returned as results, which are shown in figure 6.

Figure 6. Search engine results page

978-1-4673-1456-5/12/$31.00 ©2012 IEEE Page 168

The results include a list of LOs with titles, a link to the

repository, and a short description showing where the keywords have matched content within the LO.

A. Retrieval of similar cases process

CBR systems typically apply retrieval and matching algorithms to a case base of past search-result pairs. CBR is based on the intuition that new searches are often similar to previously encountered searches, and therefore, that past results may be reused directly or through adaptation in the current situation. The main purpose of establishing intelligent retrieval ontology is to provide consistent and explicit metadata in the process of knowledge retrieval. Our system provides multilayer retrieval methods:

1. Intelligent profiles interface: Low-level selection of query profile options, which mainly include the four kinds of user: Teachers, staff, students and visitors. The users can specify certain initial items, i.e. the characteristics and conditions for a search.

2. Ontology semantic search can query on classes, subclasses or attributes of knowledge base, and matched cases are called back.

3. The retrieval process identifies the features of the case with the most similar query. Our inference engine contains the CBR component that automatically searches for similar queries-answer pairs based on the knowledge that the system extracted from the questions text [20]. The system uses similarity metrics to find the best matching case. Similarity retrieval expands the original query conditions, and generates extended query conditions, which can be directly used in knowledge retrieval. Similarity measures used in CBR are of critical importance during the retrieval of knowledge items for a new query. Unlike in early CBR approaches, the recent view is that similarity is usually not just an arbitrary distance measure, but function that approximately measures utility.

We used a computational based retrieval where numerical similarity functions are used to assess and order the cases regarding the query. The retrieval strategy used in our system is nearest-neighbor approach [21]. This approach involves the assessment of similarity between stored cases and the new input case, based on matching a weighted sum of features. A typical algorithm for calculating nearest neighbor matching is next:

n

ii

n

i

ffsimi

RI

w

wCaseCasesimilarity

Ri

Ii

1

1

,

),(

(1)

Where wi is the importance weighting of a feature (or slot), sim is the similarity function of features, and I

if and R

if the

values for feature i in the input and retrieved cases respectively.

An important advantage of similarity-cased retrieval is that if there is no case that exactly matches the user’s requirements, this can shown the cases that are most similar to her query. The use of structured representations of cases requires approaches

for similarity assessment that allow to compares two differently structured objects, in particular, objects belonging to different object classes.

VI. EXPERIMENTAL EVALUATION

Experiments have been carried out in order to test the efficiency of Artificial Intelligent and Ontologies in retrieval reusable learning knowledge in a digital library. These are conducted to evaluate the effectiveness of run-time ontology mapping. The main goal has been to check if the mechanism of query formulation, assisted by an agent, gives a suitable tool for augmenting the number of significant LOs, extracted from the Digital Library, to be stored in the CBR.

For our experiments we considered 50 engineering students with different profiles. So that we could establish a context for the users, they were asked to at least start their essay before issuing any queries to OntoFAMA. They were also asked to look through all the results returned by OntoFAMA before clicking on any result. We compared the top 10 search results of each keyword phrase per search engine. Our application recorded which results on which they clicked, which we used as a form of implicit user relevance in our analysis. We must consider that retrieved LOs relevance is subjective. That is different people can assign distinct values of relevance to a same LO. In our study we have agreed different values to measure the quality of retrieved learning resources, excellent, good, acceptable and poor.

After the data was collected, we had a log of queries averaging 5 queries per user. Of these queries, some of them had to be removed, either because there were multiple results clicked, no results clicked, or there was no information available for that particular query. The remaining queries were analyzed and evaluated, figure 7.

Figure 7. Comparative of valid POs percentage

In each experiment we report the average rank of the user-clicked result for our baseline system, Google and for our search engine OntoFAMA. Then we calculated the rank for each retrieval document by combining the various values and comparing the total number of extracted learning resources and LOs consulted by the user (table 1).

TABLE I. ANALYSIS OF RETRIEVED LOS RELEVANCE FOR SELECT QUERIES

Excellent Good Acceptable Poor OntoFAMA 5,5 % 39,3 % 40,6 % 14,4 % Google 2,7 % 31 % 44,8 % 21,3 %

978-1-4673-1456-5/12/$31.00 ©2012 IEEE Page 169

We can observe the best final ranking was obtained for our prototype OntoFAMA and an interesting improvement over the performance of Google.

VII. CONCLUTIONS AND FUTURE WORK

This paper has presented an intelligent search system, automatic classifier for the E-learning domain. Which use CBR and ontology techniques to organize and retrieve LOs. Our platform consists of a knowledge acquisition process, object semantic classifier, and a CBR search engine. In our approach, the content is encapsulated inside a tuple (object, attribute, valor) and not directly visible on the Data Base. Thus, during the search process, metadata has to be extracted out of DL before it can be matched with the learner’s requirements. This can be managed by an agent in a hidden way. We refer to for a more detailed overview of the application of agents to educational systems. We have used all the profile agents effectively to generate preferred, relevant and recommended personalized profile for the different users. The Model has good characteristics in providing preference to the learner with novel approach of finding nearby meaning of query and learner can also recommend LOs by their opinion. With this characteristic of the model ability of an individual will be increased to learn through collective experience.

Finally the study analyzes the implementation results, and evaluates the viability of our approaches in enabling search in intelligent-based digital libraries. The results demonstrate that by improving representation by incorporating more metadata from within the information and the ontology into the retrieval process, the effectiveness of the LO retrieval is enhanced. Experiment indicates that OntoFAMA performs as well or better than expert users in both learning knowledge acquisition and learning knowledge classification. Future work will concern the exploitation of learning knowledge coming from others libraries and services and further refine the suggested queries, to extend the system to provide another type of support, as well as to refine and evaluate the system through user testing. It is also necessary the development of an authoring tool for user authentication, efficient ontology parsing and real-life applications

REFERENCES [1] Z. Jianfeng, “Study on E-learning Education Model Based on Web

Intelligence”, e-Education, e-Business, e-Management, and e-Learning, 2010. IC4E '10. International Conference on, vol., no., pp.554-557, 22-24 Jan. 2010

[2] A.P. Ali, H. Dehghan, J. Gholampour, “An agent based multilayered architecture for E-learning system”, E-Learning and E-Teaching (ICELET), 2010 Second International Conference on , vol., no., pp.22-26, 1-2 Dec. 2010

[3] D. Govedarova, S.Stoyanov, I. Popchev. “An Ontology Based CBR Architecture for Knowledge Management in BULCHINO Catalogue”. International Conference on Computer Systems and Technologies. 2008

[4] P. Warren. “Applying semantic technologies to a digital library: a case study” Library Management Journal, Emerald. 2005.

[5] H. Stuckenschmidt, and F. Harmelen, “Ontology-based metadata generation from semi-structured information”. K-CAP, pp. 163–170, ACM. 2001.

[6] Y. Sure, and R. Studer, “Semantic web technologies for digital libraries”. Library Management Journal, Emerald, vol. 26. 2005.

[7] GAIA - Group for Artificial Intelligence Applications. “jCOLIBRI project - Distribution of the development environment with LGPL”, http://gaia.fdi.ucm.es/grupo/projects/. Complutense University of Madrid. 2009.

[8] H. Ding, “Towards the metadata integration issues in peer-to-peer based digital libraries”. GCC (H. Jin, Y. Pan, N. Xiao, and J. Sun, eds.), vol. 3251 of Lecture Notes in Computer Science, Springer. 2004.

[9] J. Toussaint, K. Cheng. “Web-based CBR (case-based reasoning) as a tool with the application to tooling selection”. International Journal of Advanced Manufacturing Technology. 2006.

[10] R. Guha, R. McCool, and E. Miller, 2003. “Semantic search. In Proceedings of WWW2003”. 2003.

[11] G.F. Luger, “Artificial Intelligence, Structures and Strategies for Complex Problem Solving”. 4ª edition. Ed. Pearson Education Limited. 2002.

[12] D. Taniar, J. Wenny Rahayu, “Web semantics and ontology”. Hershey, PA: Idea Group Pub, 2006.

[13] W3C. “RDF Vocabulary Description Language 1.0: RDF Schema”. http://www.w3.org/TR/rdf-schema/.

[14] PROTÉGÉ,. “The Protégé Ontology Editor and Knowledge Acquisition System”. <http://protege.stanford.edu/>. 2009.

[15] J. Heflin, “OWL Web Ontology Language Use Cases and Requirements”. W3C Recommendation. 2004

[16] A. Gomez-Perez, A.O. Corcho, M. Fernandez-Lopez, M., “Ontological Engineering. Advanced information and knowledge processing”, Berlin: Springer. 2003.

[17] M. Horridge, H., Knublauch. “A Practical Guide To Building OWL Ontologies Using The Protégé-OWL Plugin and CO-ODE Tools”. The University Of Manchester. Reino Unido.2004.

[18] I.H. Witten, and D. Bainbridge, “How to Build a Digital Library”. Morgan Kaufmann. 2003.

[19] M. Bridge, H. G¨oker, L. McGinty, “Case-based recommender systems”. Knowledge Engineering Review.2006.

[20] S. Staab,R. Studer, “Handbook on Ontologies”. International Handbooks on Information Systems, Springer, Berlin. 2005.

[21] B. Díaz-Agudo, P.A. González-Calero, J. Recio-García, A. Sánchez-Ruiz, “Building CBR systems with jColibri”. Journal of Science of Computer Programming, 2007.

978-1-4673-1456-5/12/$31.00 ©2012 IEEE Page 170