PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM An Educational Ontology-based M-Learning System https://doi.org/10.3991/ijim.v10i4.6011 AbdelGhani Karkar and Jihad Mohamad Al Ja’am Qatar University, Doha, Qatar Abstract—Smart devices applications can assist young chil- dren in improving their learning capabilities and compre- hension skills. However most of learning applications are built without taking into consideration the effective needs and background of Arab users. They are somehow incom- patible with their local environment and culture. We pro- pose in this paper a mobile-based educational system that displays illustrations automatically to characterize the con- tent of children’ stories on animal domain, and we use some intelligent technique to answer users’ questions using an educational ontology. The generation of illustrations passes through different phases which include text processing, extraction of word-to-word relationships, building and accessing an ontology, and using Internet search engines for retrieval of complimentary information. The system can be used also by instructors to teach the children in a non- conventional manner. They can customize semantically the structure of the questions and determine how the answers will be displayed. In order to customize questions and an- swers, we have defined a semantic infrastructure to define the logic of terms, and the workflow of answers. The aim of our system is to improve the children educational skills to grasp vocabulary and grammar using multimedia with a portable smart device which includes observation, compre- hension, realization, and deduction. Children will then be able to continue learning outside the limited time of their schools and from any location. Index Terms—Mobile Learning; Question Answering; Edu- cational Ontology; Multimedia; Education. I. INTRODUCTION Nowadays children are surrounded with various types of digital smart devices, educational software and edu- tainment games. Mobile devices opened a new horizon in learning and communication. They can be handled in pockets, by hand, or around the necks, which help anyone to learn and communicate from any location and at any time. With a smart device, like an iPhone or an iPad, we can capture pictures, record video clips, and share these resources spontaneously with friends and families. We can also use these devices to learn in an unconventional man- ner. In fact, using a portable device for learning is advan- tageous and more attractive as the learners become freely able to access educational resources through the Internet or specific systems. It is noticed that children enjoy learn- ing using an iPad or a tablet while playing and moving. They can interact with these items and become more en- gaged and motivated than just sitting for a while using a computer device on desk or when being in a classroom. By establishing more relationships with their friends and teachers, engendered ideas, and outstanding observations of the world, children are seeing and recognizing the world contrastively. This upgrade of educational technol- ogy shows a considerable movement in the path smart devices can be utilized to stretch the minds of children and break the centralized learning paradigm. Edutainment games offer also a new way of learning while playing. Thus children can enhance their learning skills while en- tertaining, moving or when being at home or even in bed. Unfortunately, the learning resources for Arab children are very limited. In fact, the developed games and apps are not suitable for the culture, language and background of the Arab children. In fact, less than 1% of available apps are Arabic-based. This is a real challenge of Arab learners. A new initiative is launched recently by the Arab League for cultural and education [1] to develop Arab- based learning Apps to promote the usage of the Arabic language in education and edutainment through mobile devices. However, it takes a very long time to build a repository with rich Arabic apps and edutainment games. In this paper, we propose a mobile-based educational system that can map Arabic stories for children to illustra- tion and answer questions through an educational ontolo- gy. It allows children to grasp Arabic vocabulary and grammar in a simple and natural way using multimedia. In fact, multimedia technology can keep the children en- gaged for a longer time [2] and also has a great impact on their way of learning. The proposed system is composed of a main server as a back end and mobile devices as a front end. We use Arabic natural language processing (NLP) techniques to generate illustrative pictures to mine the users’ questions and represent the content of the an- swer in different models. We develop an educational on- tology where its concepts are linked semantically (i.e., a lion is linked with the forest, a Camel is linked with the desert). The system allows the teachers to address logical queries and get in return multimedia-based intelligent responses which can be used to enhance the learning skills of children. The instructor can add new knowledge to the educational ontology and associate preferred illustrations. The system can analyze the questions of users using NLP tools (i.e., parser, tagger), extract key concepts, and re- trieve then the information. The rest of the paper is organized as follows: in section 2 we discuss the existing multimedia systems with the question answer technique. In section 3, we give details about our proposed system and its features. In section 4, we discuss the technical specifications and the usage of the mobile device. In section 5, we show assessment re- sults. In section 6, we discuss the main findings, and finally in section 7 we conclude the paper. 48 http://www.i-jim.org PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM II. BACKGROUNDS Natural language processing consists of mining texts to extract useful information. These information can be used to determine the main scope of the texts in order to classi- fy them, generate summaries, extract specific vocabulary (i.e., educational system) or spot events which can be used as references for taking decisions (i.e., alerting systems). Several NLP-based systems have been proposed recently with applications on different domains (e.g., finance, transportation, question answering, education, etc.) [3-5]. These systems start generally by splitting the text into segments, stem the words of each sentence using a dic- tionary, find synonyms using a thesaurus, remove stop words, and annotate or add tags to the words stems (i.e., verb, noun, adverb). They use intelligent techniques to understand the meaning of each sentence in the text (e.g., machine learning). These techniques include, usage of logical rules, development of ontology on specific do- mains, usage of statistical approaches (i.e., frequency of words in the text, recall and precision), diacratisation, and develop a corpus. Building a learning system that can understand Arabic texts automatically is a challenging task. Such a system can have a great impact on educating Arab children and non-native Arabic speakers in an effi- cient manner. In fact, it allows them to learn Arabic vo- cabulary and grammars almost autonomously and natural- ly especially whenever they use their mobile devices. A good number of educational systems have been proposed in the last decade. Cheng et al. [2] have proposed a com- puter reinforced online multimedia education framework called Crome to enhance the teaching and learning capa- bilities of children and adults. It includes multiple compo- nents for information modelling, learning, teaching, and testing. Wastam et al. [6] have developed a system to help children enhance their knowledge in a specific topic. The system is composed of two components: the first one allows the instructor to select the topic of a story that will be illustrated by flashcards, while the second one asks the children to arrange the scenes in a logical manner accord- ing to the topic of the story. Rosmani and Wahab [7] have developed a multimedia-based system called i-IQRA' to teach children the Holy book of ‘Al-Quran’. It consists of introducing the Arabic alphabet through simple words, reading them automatically, and making quizzes. Tabot and Hamada [6] have proposed a web-enabled multimedia learning system to provide physics instruction on selected topics. Erradi et al. [8] have proposed a multimedia learn- ing platform called ‘ArabicTutor’ which provides Arabic spelling, gives meanings of words accompanied by simple multimedia. Ping et al. [9] have developed an assistive learning system to assist children with hearing impairment learn the Malay language. Dong and Li [3] have proposed a multimedia learning system that can be used by several users simultaneously. Wuang et al. [10] have built a multimedia courseware system that is based on learning theories. All these sys- tems are based on static contents with different learning objectives. They are generally addressed to adult learners through a computer machine. In the other hand, a question/answering (QA) system consists of mining questions from particular resources that can be in form of natural text, interlinked data, or datasets. Zhu et al. [11] proposed a QA system to answer natural language questions from DBpedia [12] using graph tra- versal approach. The system determines entities from the natural question and builds their corresponding subgraph of knowledge base. It uses the instantiated subgraph to solve semantic mapping problem and the disambiguation problem. Therefore, a graph traversal method in DBpedia is used to select the best answer according to the rank of the path. Sander et al. [13] proposed an approach to trans- late natural language sentences into SPARQL queries to provide applicable and intuitive semantic search in the industry. The translation is based on background lexicons and ontology. The result shows some improvement when applied. Hence, the training period and knowledge re- quired for users decreased, while the overall usage of the natural input is increased. Berant et al. [14] proposes a paraphrase system to get higher number of raw text that is not related to any knowledge base. It uses a method to generate a collection of logical forms. Then, it uses a paraphrase model to select the best paraphrase input and output that correspond to the original form. The system improved the accuracy on the recently proposed QA da- tasets. Unger et al. [15] proposed an approach to parse a natu- ral question to generate a SPARQL template that can determine promptly the internal semantic structure of the question. The template is employed based on statistical identification of the entity and predicate detection. There- fore, the slots of the template are filled based on the enti- ties determined using string similarity and natural lan- guage patterns. Joris et al. [16] proposed a scalable query- based search system called “Scalewelis”. It is a query- based faceted search uses semantic web and connects SPARQL end points to enable more expressive search. It uses its own data structure and not capable to scale with large linked-data datasets (i.e., DBpedia). Dima [17] proposed a QA system prototype to extract information stored in RDF through English natural lan- guage. The system translates the natural language into a SPARQL query to extract the required information. The proposed system is capable to process simple questions and cannot solve complex questions. Shekapour et al. [18] proposed a QA system that can extract knowledge from interlinked data. It translates natural language questions into corresponding SPARQL queries and extract infor- mation from interlinked datasets. He et al. [19] proposed a novel QA method. It uses first-order logic to resolve the ambiguities problem while resolving the answer. The proposed method can cover more textual expressions and answer more questions that manually setting the pattern of the text. The experiments showed effective results of using the proposed method. As many multimedia-based educational systems and question-answering ones have been proposed. Providing a complete multimedia educational system that can auto- matically illustrate the content of stories, and can provide explanatory answers that are based on reasonable predi- cates for educational purpose is not well covered in the literature. In this paper, we propose a mobile educational system that can provide illustrations for Arabic stories and answer questions through an educational ontology. Hence, the system can be used by instructors and individuals to promote existing educational systems. III. THE PROPOSED SYSTEM Our proposed system can generate multimedia tutorials with dynamic contents and answer questions by using an educational ontology to depict answers with illustrative iJIM ‒ Volume 10, Issue 4, 2016 49 PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM pictures by employing online search engines. The system can be used to teach young children the Arabic vocabulary and grammar in an effective manner using multimedia by mining children stories’ texts. Non-Arabic native learners can also use the system to learn the Arabic language effi- ciently. We have collected thirty educational stories for children and added manually their words to a corpus. The main concepts or keywords of Arabic texts are represented by multimedia. Thus the children can see the graphical representation of words related to named-entities and actions. For instance, for the following sentence ‘a lion is running’, the children can see the images of a lion as it is a named entity. They can see also images and clips of a running lion. In fact the stem of the word ‘running’ is ‘run’ and it is a verb. Thus it corresponds to an action. We use a modern standard Arabic parser to extract infor- mation (i.e., places, named entities, events), an education- al ontology to get related illustrations, and an online search engine (i.e., Google) to get additional multimedia elements not available in the ontology. The instructor can feed the system with an Arabic text and get in return a collection of illustrations corresponding to the text main concepts. He/she can adapt these illustrations based on the specific need of each child. The dynamic generation of illustrations would become further time efficient accord- ing to the process of machine learning. The system is composed of several components which are the following: 1) Arabic text parsing: first, the Arabic text is segmented into paragraphs, sentences and words. Each word is annotated automatically to identify its part- of-speech (e.g., noun, verb, adjective); 2) Educational Ontology: we have created an Arabic ontology based on Arabic WordNet [20] to extract knowledge and entity relationships. We have used also the DAML+OIL [21] to annotate the Arabic terms. The ontology is used also to get further details about the Arabic terms and their relation- ships. For instance, it is possible to identify the required content that should be given to children according to their education levels, to access educational semantic models created upon children stories. For instance, we can identi- fy that the ‘lion’ animal is a carnivore and lives in the forest while the ‘camel’ animal lives in the desert and it is a herbivore. All instances inside the ontology are linked semantically with different logical rules (e.g., is_a, has_a, eat_a, etc.). A semantic query can be addressed to the ontology to get a particular instance and it can return all associated relations recursively. Illustrations are linked with most entities through assigned links to avoid storing large binary objects inside the ontology. 3) Corpus: we have created a corpus to store information about the common used terms in stories of children. And 4) Multi- media: this component is used to provide illustrations that are requested by the ontology. Subsequently, further search queries can be sent to online search engine (e.g., Bing, Google) to fetch more images which should be validated by the instructors. Each Arabic term can have a large set of illustrations acquired from the search engine. Instructors should select the most appropriate ones to present to the children. A. Applied Methodology We applied the IEEE 1074-2006 standard to implement our educational ontology. It is a standard for evolving software project life cycle procedure [22]. The methodol- ogy used is based on the concept of Category Theory, which is initially a mathematic foundation theory. Many researchers such as Husserl and Hartmann declared that ontology depends on the category theory. Methontology is the used methodology to implement the Educational on- tology [23]. It follows a circular approach which enables evaluation and improvement of the ontology to make it more accurate. A. Arabic text processing and relations extraction The Arabic text to be processed can be either selected from a list of predefined stories or entered manually to the system. The processing of the text passes through several phases: 1) Segmentation: the text will be segmented into paragraphs. Each paragraph is split into sentences of sev- eral words. 2) Tagging: we use a customized version of the Stanford Parser [24] and MADAMIRA tool [25] to determine the part-of-speech (POS) of each Arabic word. For instance, we use the tags NN for a Noun, VB for a Verb, and ADJ for an Adjective. 3) Linking-Grammar: tagged words will be linked with each other according to their POS. For instance, the word that is tagged as noun can specify the entity name. It can therefore be linked directly with a verb tagged word to specify an action. 4) Semantic-Extraction: One of the intelligent tasks that are performed by our system is the semantic relation extrac- tion (SRE). It can identify entity acquaintance, relation- ship with other entities, and its types, (e.g., a lion is car- nivorous and eats rabbit; if we know that a tiger is carniv- orous, we can infer that the tiger can also eat rabbit). The SRE is used also for knowledge extraction (i.e., actor, action, behaviours, etc.) that are required during the phase of text processing. Eventually, all linked words will be mapped together and stored in mapping-tables (i.e., hash tables) which allow quick retrieval of information, a se- mantic query will be dynamically generated for each row to extract the semantic knowledge from the ontology which refers to as OntologyItem. Each OntologyItem contains all semantic information, details, associated illus- trations, and further illustrations downloaded using online search engines. Fig. 1 shows the different text processing phases to generate the illustrations. Figure 1. Arabic text processing phases. B. Question processing Processing the user’ question passes by several phases: 1) Segmentation: the question will be divided into differ- ent sentences according to some interrogation marks. 2) Entity-relationships: the relation between words is based on the formal concept analysis (FCA) to assign relations (triplet). The assignment is based on the composition of the two-matrices. The entity-sentence matrix is built de- pending on the appearance of entities (e.g., tiger, cat, etc.) with their sentences. The property-sentence matrix is built according to the occurrence of property-values (e.g., yel- low will be linked with color, jump will be linked with action, etc.) with their sentences. The connection is result- ed by the multiplication of the two matrices. For instance, the question ‘What does the cat eat? What is its cover?’ will provide the relation: {cat: [eat: ?v1, cover: ?v2]}. The 50 http://www.i-jim.org PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM entity-sentence matrix is used also to determine the link- age of the obtained result from external ontology for new information (i.e., processing information of the abstract property obtained by DBpedia ontology). 3) Semantic Relation Extraction (SER): It is one of the intelligent missions that are achieved by our system. It can determine entity acquaintance and feasible relationships with other entities based on semantic rules. For instance, for the question ‘How the crocodile can hunt gazelle?’ we need to find possible relations between the crocodile and the gazelle, since the crocodile cannot run but the gazelle can. Therefore, after fetching possible relations, we can deter- mine a possible answer that a crocodile lives in river and the gazelle drinks from river. The SER is used also for knowledge extraction (i.e., entity, property, behaviour, etc.) that is needed during the phase of entity- relationships. If the question contains information about entities that are not previously defined in the ontology. A request will be sent to DBpedia ontology to get additional information. The obtained result will be mapped with our defined ontology to benefit from inference and reasoning. 4) Processing Data: we have defined several question patterns at first stage to cover most important questions that can be asked. Each pattern has different processing operation and its own answer. Hence, answers can differ to provide different kind of explanations. C. Ontology Architecture The ontology is created using Protégé [26], an open source tool for editing ontologies and knowledge pro- curement system. It supports the development of ontolo- gies and reserves an application platform with knowledge based systems. Declaring an instance of a specific class type requires: 1) selection of the class; 2) creation of an instance, and 3) filling values of the instance. Fig. 2 shows a section of the created educational ontology with its rela- tions and concepts. D. Reasoning Reasoning is one characteristic that strengthens the on- tology to derive unprecedented facts that are not expressed explicitly in the knowledge base. Reasoning expression rules can be formed to derive or fetch deductive conse- quences of knowledge. Up to date, different reasoners have been developed [27-29] to support diverse reasoning services (e.g., classification, consistency, realization, etc.) using different interfaces (e.g., Command line, Protégé, OWLlink [30]) and several syntaxes (e.g., RDF/XML, OWL/XML, OWL API [31], etc.). While ontologies focus on the classifications of in- stances to emphasis facts by enforcing logic rules (e.g., multiplicity, bounds, domain, etc.). Reasoners use seman- tic rules to define how new facts will be determined. Be- sides, reasoning can be used to analyze the validity of knowledge (i.e., verifying if inferred instances are valid) or managing knowledge in general. For example, the on- tology declaration “every whale is a mammal” with the relationship “mammals are warm blooded” can be repre- sented by: !x : { Whale( x ) " Mammal( x ) } " WarmBlood- ed( x ) Using the precedent relationship, if x denotes a whale shark, then it is possible to discover the new relationship a whale shark is warm-blooded. IV. SYSTEM AND MOBILE APPLICATION IMPLEMENTATION A. Proposed Modules Fig. 3 shows the different modules that are complemen- tary to fulfil the processing and illustration generation operation in the proposed system. The Text Tools module is responsible of performing various text processing, which include, segmentation, tagging, term weighting using a statistical corpus. The Corpus Module supplies services of getting the most used terms with their illustra- tions. The Semantic Models module is used for reasoning. The Online Search Engine module provides prospection services for illustrations on the Internet. B. System Architecture The system is composed of two components: a server and a mobile application as shown in Fig. 4. The system is based on the transactions of messages, the combination with previous software applications is viable by com- municating with the server and getting the desired infor- mation according to the open API of the system. The Client mobile application is a hybrid application based on HTML5 technology that can work on smart devices (e.g., iPhone or Android). The client application establishes a full-duplex TCP connection with the server to ensure transaction of messages which is done from both sides, to obtain reliable information and latest updates. Eventually, the administrator can recognize connected users, track server performance, and check if the server noted any failure as depicted in Fig. 4. Figure 2. A section of the educational ontology. Figure 3. The system modules. !"#$%&'() *'+,$ -#.&(# /0%+12% 344 5+'6# 714&,1%# 8#41%# 944"$%+1%# :+16#+$# ;<'=)><1&( 344 ?''@4#/(@&(# A&(B!"#+C 9(. 5+'D#+%CE1F# GH'+/12<1&(I) G,#DI >41$$ G,#DI G,#DI 7 G,#DI G<1$I G,'F1&(K144I > G,'F1&(K144I G<1$I) G#(@&(#I) G,'F1&(K144I) G<1$I iJIM ‒ Volume 10, Issue 4, 2016 51 PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM C. System Operation Whenever the server is launched, it loads its packages and components as shown in Fig. 5. It loads the defined ontology, text parser components, and finally it opens a TCP connection to wait for the users’ requests. Hence, all phases are processed on the server side to decrease power consumption on the client side (i.e., mobile device). After starting the Client mobile application, it assays to bind the server, which validate if the provided authentication is veracious and set up the connection accordingly. Upon an effective connection trial, the user will be eli- gible to enter or open existing Arabic stories and process them. When the user write a question, a client request composed from, 1) the header “ProcessStory”, and 2) the content of the story, placed together in a JSON format. When the server gets the request from the client, it reads the header of the message and executes the intended function accordingly. During the phase of processing the story content, the server passes through various phases: 1) segmentation; 2) tagging and stemming; 3) semantic knowledge extraction, and 4) finally linking Arabic terms with their illustrations. The processing result will be stored in a structure composed from the annotation, se- mantic information, and the stem of each Arabic term. Fig. 6 shows a client/server sequence diagram to show the various phases of communication. During the phase of answering questions, fetching pos- sible relations is required according to the context of the question. Fig. 7 shows an example to answer the ques- tion: Prove a crocodile can hunt gazelle. During the processing phase, the server logs all per- formed steps to allow the administrator to be able to track the required information, perform certain analysis, and detect any occurred failure as shown in Fig. 8. Moreover, the administrator can configure the server to report 1) critical failures by email; 2) maximum connected users is reached, or 3) the memory became full. Figure 4. The proposed architecture. Figure 5. The NLP server startup. Figure 6. A client-server sequence diagram. Figure 7. Demonstration of how Crocodile can hunt Gazelle. Figure 8. Processing the user mobile request. >+'2',&4# 8&6#+ ?1L#44# 4&6#$)&( ,+&(B$) 3M"1%&2) 8#D%&4#$ 3(%#4'D# N#+O&6'+'"$ >1+(&6'+'"$ 21()#1% 21()<"(% *1%#+2'"+$# E1%"+14 &$)1 &$)1 &$)1 &$)1 &$)1 &$)1 Crocodile is kind of Aquatic Reptiles and Carnivorous Carnivorous can eat Herbivorous Gazelle is kind of Antelope and Herbivorous Crocodile can hunt Gazelle in River Answer: 52 http://www.i-jim.org PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM On the client side, the processing request and response of the story is done in different threads, to keep the user able to continue her/his work without any interruption. Finally, Fig. 9 shows the different results that will be displayed for the user on her/his mobile device. V. EXPERIMENTS AND EVALUATION In order to take initial valuable evaluation of the differ- ent components, we have evaluated them separately. Then we identified the necessary items. We have used “RStu- dio1” tool to analyze data using different techniques (e.g., ANOVA, ANCOVA, Model Selection, etc.), and to repre- sent different statistical diagrams. We have developed different functionalities to collect data, analyze the re- sponse time, get experts feedback through heuristic based questionnaires, assess the performance of users, and so. Assessments results are preliminary and still not final in this study. Ratings are simulated by 7-point rating system filled by two instructors and 20 children. We calculated the processing time and extraction speed using forecasting algorithm [32] based on knowledge extraction from 5 stories and 20 instances from the ontology. The P-value and F-value had been used in the assess- ment. The P-value describes the probability statement that is involved in attaining the probability of realizing test statistics at least as extreme as the observed one, assuming that the null hypothesis is true. In General, a p-value of 0.05 (i.e. at the level of 5%) or less rejects the null hy- pothesis. In the other hand, to confirm the significance of the test, we can use the F-statistics, see Fig. 10. The F- value is the ratio of mean squares, and it is equal to: !!!"#$% ! !"##$%$&'$(!!"#$""%!!"#$%& !"##$%$&'$(!!"#$"%!!"#$%& If the F-value overreaches the critical value, then the null hypothesis will be rejected and we can conclude that there is a significant impact in the assessment. There is a difference between P-value and F-value; P-value describes the probability, whilst F-value is a value of the test. The assessment set is based on the constructed educa- tional ontology which consists from six domains as shown in Table 1. The evaluation of user performance is based on thirty educational stories categorized in five different domains. This section discusses the evaluation of the following different parts which include: (1) Text processing; (2) Semantic Extraction; and (5) User Performance. A. Evaluation of Text processing and Semantic Queries In this section, the evaluation of the third-party compo- nent is partially maintained. In fact, the parser component serves in generating sentence tree hierarchy and words linkage between each other. This linkage is used to gener- ate dynamic semantic queries. The evaluation include: number of words in the text, number of paragraphs, pro- cessing time in milliseconds (ms), accuracy of generated semantic queries, and experts rating, as shown in Table II. The accuracy is computed by comparing the different parts of the generated semantic-queries with proposed ones (e.g., domain, term, color, etc.). 1 RStudio: http://www.rstudio.com/ Figure 9. Launching the mobile application on iPhone devices. TABLE I. NUMBER OF INSTANCES IN ONTOLOGIES PER DOMAIN Domain Ontology Classes Instances Carnivorous 13 55 Herbivorous 13 42 Birds 9 26 Fishes 8 14 Insects 8 16 TABLE II. AVERAGE TEXT PROCESSING FOR FIVE DOMAINS Domain Number of Words Paragraphs Processing Time (ms) Experts Rating Carnivorous 131.80 3.00 650 88% Herbivorous 132.00 1.60 651 87% Birds 158.40 2.20 701 78% Fishes 117.20 2.40 629 92% Insects 132.20 2.60 652 86% According to the result obtained we have considered two regression models to compare the two models by F value and using ANOVA: 1) Words = !0 + !1 * Processing Time and 2) Words = !0 + !1 * Processing Time + !2 * Experts Rating where !0 denotes the intercept, !1 and !2 denote the slop term in the average change of values. Analysis of Variance Table Model 1: Words ~ ProcessingTime Model 2: Words ~ ProcessingTime + ExpertsRating Res.Df RSS Df Sum of Sq F Pr(>F) 1 23 1202.5 2 22 150.9 1 1051.6 153.31 2.174e-11 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Figure 10. Model Comparision using F-Distribution and ANOVA. Discussion: based on the above results obtained in Fig. 10, we can conclude from the value of resulted F- distribution that the probability that there is no difference between the two models is 0%. This means that the two covariance variables, Processing Time and Accuracy, are required in the analysis. iJIM ‒ Volume 10, Issue 4, 2016 53 PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM B. Evaluation of Semantic Extraction Semantic extraction is based on tagged words resulted from the Text-Parser component and the generated seman- tic SPARQL query. The evaluation of the extracted se- mantic extraction include: number of structured instanced loaded, processing speed in milliseconds (ms), accuracy, and experts rating, as shown in Table III and Fig. 11. The comparison depends on matching the different properties of the semantic-instances (e.g., type, child-instances, do- main, etc.) Discussion: based on the above results, the confidence interval between the two modes real image and grayscale provided the highest confidence interval, where 99% of hypothetically supervised interval will obtain the true value at significance level of 0.05. However, the confi- dence interval between 3D image and toy model will pro- vide a hypothesis value of 23%. It is clear that the confi- dence interval between 3D image and other modes remain lower than others. We can conclude that the retrieved image cannot be easily classified as a 3D image or if it fits in other domains. C. Evaluation of user performance In this section, different inputs are used to evaluate the performance of the users. A total of 20 children between 9 and 10 years old have been selected and divided into four different levels according to their knowledge. The evalua- tion comprises: reading time in seconds, number of cor- rect answers, scores, and so. Table 5 illustrates the aver- age results among the different measurement units. Fig. 12 illustrates the distribution of the average cor- rect answers among the four different groups. In fact, users are allowed to make different tests and know their mistakes. The grade of the last test is used if the same test is taken. The average of correct answers is calculated among the complete disparate sort of tests. Fig. 13 shows the average scores regression lines of the four different groups according to the average reading time while accessing more images. The figure is generated according to the result of the above ANCOVA test. Each line represents its group. Regression lines denote the esti- mated meetings points between both average scores and average reading time. TABLE III. AVERAGE SEMANTIC KNOWLEDGE EXTRACTION PERFORMANCE No. Mode Number of Instances Speed (ms) Accuracy 1 Real Image 26.20 350 83% 2 Cartoon 26.60 351 79% 3 3D Image 31.60 401 75% 4 Grayscale 23.60 329 72% 5 Toy Model 26.40 352 78% TABLE IV. AVERAGE PERFORMANCE FOR FOUR GROUPS OF USERS Group Reading Time (sec) Correct Answers Mistakes Scores Quiz Com- pletion Time (sec) 1 396.76 87.92% 13.16% 88.24% 283.20 2 435.33 88.92% 10.64% 88.36% 273.36 3 449.48 92.56% 8.44% 90.92% 263.04 4 423.14 88.72% 12.00% 87.48% 268.80 Discussion: based on the results above, it is obvious to observe the relation between the average reading time while accessing more images and the average scores. The direction of the regression lines is tending upward. Even- tually, we can conclude that the probability of achieving higher score by users who spend more reading time and accessed more images. Figure 11. Differences in Accuracy mean levels of as.factor(Domain). Figure 12. The average distribution of correct answers. Figure 13. Average scores and reading time while accessing more images. 54 http://www.i-jim.org PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM VI. FINDINGS AND DISCUSSION Based on the objective dissection described in the pre- vious section, we can conceive that the system can assist instructors, children and Arabic learners to access rapidly semantic information to answer questions and to provide illustrations to represent sentence. In the other hand, we have compared our proposed on- tology with DBPedia and our opinion is to enrich the ontology with interlinked semantic information about instances. Hence, enrichment can improve the speed of information retrieval, semantic navigation between in- stance, consistency, and reasoning. In the other hand, adding detailed hierarchical classification can enhance the speed to identify how instances are related to each other in the scope of inheritance gradation. First, enriching can increase the speed of information retrieval. Instead consuming processing time every time to extract a named instance that describes particular value, it is possible to store values as physical instances connected to the prime one through named properties. For instance, it is possible to connect the instance mugger crocodile with 5 meters through the property maximum length. Therefore, it is possible to promptly determine the maximum length of the crocodile. Second, enrichment can improve semantic navigation between instances. The enhancement can be done instant- ly by linking instances semantically. For example, it is possible to connect the instance mugger crocodile with gazelle through the property can_eat. In this context, some generic properties can be applied on the class itself (e.g., reptiles are cold-blooded). In the other hand, not all properties can be generic (e.g., not all kind of reptiles can eat meat). Therefore, some properties should be applied on instances themselves. Third, the enrichment can improve consistency of in- stances. Besides the textual information attached to an instance, diverse kind of semantic information will be attached which help in providing all specifications about the instance. Fourth, the enrichment can enable users to benefit from reasoning. When an instance has mainly textual infor- mation that describes it, the usage of reasoning will be limited to basic criterion. In the other hand, when having enriched instances that have their own characteristics and properties, the reasoning will become valuable and robust to infer to new knowledge based on semantic rules. Eventually, enriching the hierarchical classifications in the ontology can improve locating hierarchical connec- tions when using staircase backward chaining. For exam- ple, when applied in the demonstration to proof that a crocodile can hunt gazelle; knowing that a crocodile can- not run whilst the gazelle can; therefore possible relation between the two instances is required (i.e., river) and another relation to proof that the first instance can perform the task (i.e., hunt). VII. CONCLUSION This study presents a complete multimedia mobile- based system that can display illustrations automatically to characterize the content of stories, and can answer ques- tions using an educational ontology. The proposed system belongs to portable learning technology which can be on mobile devices to teach Arab children in an attractive and non-traditional style. It can improve the learning capabili- ties, memorization skills, communication and understand- ing of children while decreasing teaching overhead and administration through management and automating learning process. REFERENCES [1] ALESCO. Arab League Educational, Cultural and Scientific Organization (2013). Available: http://www.projects-alecso.org [2] L. I. Cheng, A. Basu, and R. Goebel, "Interactive Multimedia for Adaptive Online Education," IEEE MultiMedia, vol. 16, pp. 16- 25, 2009. http://dx.doi.org/10.1109/MMUL.2009.11 [3] A. Dong and H. Li, "Multi-ontology based multimedia annotation for domain-specific information retrieval," in IEEE International Conference on Sensor Networks, Ubiquitous, and Trustworthy Computing (SUTC'06), 2006, pp. 158-165. http://dx.doi.org/10.1109/SUTC.2006.94 [4] L. Zemmouchi-Ghomari and A. R. Ghomari, "Translating natural language competency questions into SPARQLQueries: a case study," in The First International Conference on Building and Ex- ploring Web Based Environments, 2013, pp. 81-86. [5] X. G. W. Hou-Feng, "The Development of Topic Models in Natural Language Processing," Chinese Journal of Computers, vol. 8, p. 009, 2011. [6] D. R. A. Rambli and S. Sulaiman, "A guided digital storytelling prototype system using illustrated flashcards," in 2010 Interna- tional Symposium on Information Technology, 2010, pp. 1-6. [7] A. F. Rosmani and N. A. Wahab, "i-IQRA': Designing and con- structing a persuasive multimedia application to learn Arabic characters," in Humanities, Science and Engineering (CHUSER), 2011 IEEE Colloquium on, 2011, pp. 98-101. http://dx.doi.org/10.1109/chuser.2011.6163884 [8] A. Erradi, S. Nahia, H. Almerekhi, and L. Al-kailani, "ArabicTu- tor: A multimedia m-Learning platform for learning Arabic spelling and vocabulary," in Multimedia Computing and Systems (ICMCS), 2012 International Conference on, 2012, pp. 833-838. http://dx.doi.org/10.1109/icmcs.2012.6320220 [9] T. P. Ping, C. P. Chan, H. Sharbini, and A. A. Julaihi, "Integration of cultural dimensions into software localisation testing of assis- tive technology for deaf children," in Software Engineering (MySEC), 2011 5th Malaysian Conference in, 2011, pp. 136-140. http://dx.doi.org/10.1109/mysec.2011.6140658 [10] Y.-P. Wuang, C.-S. Chiang, C.-Y. Su, and C.-C. Wang, "Effec- tiveness of virtual reality using Wii gaming technology in children with Down syndrome," Research in developmental disabilities, vol. 32, pp. 312-321, 2011. http://dx.doi.org/10.1016/j.ridd.2010.10.002 [11] C. Zhu, K. Ren, X. Liu, H. Wang, Y. Tian, and Y. Yu, "A Graph Traversal Based Approach to Answer Non-Aggregation Questions Over DBpedia," arXiv preprint arXiv:1510.04780, 2015. [12] P. N. Mendes, M. Jakob, A. García-Silva, and C. Bizer, "DBpedia spotlight: shedding light on the web of documents," in Proceed- ings of the 7th international conference on semantic systems, 2011, pp. 1-8. http://dx.doi.org/10.1145/2063518.2063519 [13] M. Sander, U. Waltinger, M. Roshchin, and T. Runkler, "Ontolo- gy-based translation of natural language queries to SPARQL," in NLABD: AAAI Fall Symposium, 2014. [14] J. Berant and P. Liang, "Semantic Parsing via Paraphrasing," in ACL (1), 2014, pp. 1415-1425. [15] C. Unger, L. Bühmann, J. Lehmann, A.-C. Ngonga Ngomo, D. Gerber, and P. Cimiano, "Template-based question answering over RDF data," in Proceedings of the 21st international confer- ence on World Wide Web, 2012, pp. 639-648. http://dx.doi.org/10.1145/2187836.2187923 [16] G. H. Joris and S. Ferré, "Scalewelis: a scalable query-based faceted search system on top of sparql endpoints," in Work. Multi- lingual Question Answering over Linked Data (QALD-3), 2013. [17] C. Dima, "Intui2: A Prototype System for Question Answering over Linked Data," in CLEF (Working Notes), 2013. [18] S. Shekarpour, E. Marx, A.-C. N. Ngomo, and S. Auer, "Sina: Semantic interpretation of user queries for question answering on interlinked data," Web Semantics: Science, Services and Agents on iJIM ‒ Volume 10, Issue 4, 2016 55 PAPER AN EDUCATIONAL ONTOLOGY-BASED M-LEARNING SYSTEM the World Wide Web, vol. 30, pp. 39-51, 2015. http://dx.doi.org/10.1016/j.websem.2014.06.002 [19] S. He, K. Liu, Y. Zhang, L. Xu, and J. Zhao, "Question Answering over Linked Data Using First-order Logic," in EMNLP, 2014, pp. 1092-1103. [20] H. Al-Barhamtoshy and W. Al-Jideebi, "Designing and Imple- menting Arabic WordNet Semantic-Based," in the 9th Conference on Language Engineering, 2009, pp. 23-24. [21] W. Bin, "The Research on Mapping from DAML+ OIL Ontology to Basic-Element and Complex-Element of Extenics," in Interna- tional Conference on Intelligent Computing, 2012, pp. 570-578. http://dx.doi.org/10.1007/978-3-642-31588-6_73 [22] D. J. Schultz, "IEEE standard for developing software life cycle processes," IEEE Std, pp. 1074-1997, 1997. [23] M. Fernández-López, A. Gómez-Pérez, and N. Juristo, "Methon- tology: from ontological art towards ontological engineering," 1997. [24] C. D. Manning, M. Surdeanu, J. Bauer, J. R. Finkel, S. Bethard, and D. McClosky, "The Stanford CoreNLP Natural Language Processing Toolkit," in ACL (System Demonstrations), 2014, pp. 55-60. [25] A. Pasha, M. Al-Badrashiny, M. T. Diab, A. El Kholy, R. Eskander, N. Habash, et al., "MADAMIRA: A Fast, Comprehen- sive Tool for Morphological Analysis and Disambiguation of Ara- bic," in LREC, 2014, pp. 1094-1101. [26] Protégé Ontology Editor (2003). Available: http://protege.stanford.edu [27] D. Tsarkov and I. Horrocks, "FaCT++ description logic reasoner: System description," in Automated reasoning, ed: Springer, 2006, pp. 292-297. http://dx.doi.org/10.1007/11814771_26 [28] E. Thomas, J. Z. Pan, and Y. Ren, "TrOWL: Tractable OWL 2 reasoning infrastructure," in The Semantic Web: Research and Applications, ed: Springer, 2010, pp. 431-435. http://dx.doi.org/10.1007/978-3-642-13489-0_38 [29] R. Shearer, B. Motik, and I. Horrocks, "HermiT: A Highly- Efficient OWL Reasoner," in OWLED, 2008, p. 91. [30] T. Liebig, M. Luther, O. Noppens, M. Rodriguez, D. Calvanese, M. Wessel, et al., "OWLlink: DIG for OWL 2," in OWLED, 2008. [31] M. Horridge and S. Bechhofer, "The owl api: A java api for owl ontologies," Semantic Web, vol. 2, pp. 11-21, 2011. [32] R. H. Shumway and D. S. Stoffer, "An approach to time series smoothing and forecasting using the EM algorithm," Journal of time series analysis, vol. 3, pp. 253-264, 1982. http://dx.doi.org/10.1111/j.1467-9892.1982.tb00349.x AUTHORS A. Karkar is a PhD Candidate in the Department of Computer Science and Engineering, Qatar University, Doha, Qatar. He has published many research papers. He managed as committee member different conferences and chaired several sessions. His research interests include semantic web, software engineering, artificial intelligence, distributed systems, and multimedia learning. (e-mail: a.karkar@qu.edu.qa). J. M. Al Ja'am is with the Department of Computer Science and Engineering, Qatar University, Doha, Qatar. He has published many research papers. He is leading research projects and has many PhD students. His re- search interests include multimedia, assistive technology, learning systems, human–computer interaction, stochastic algorithms, artificial intelligence, information retrieval, and natural language processing. (e-mail: jaam@qu.edu.qa). This article is an extended and modified version of a paper presented at the EDUCON2015 conference held in Abu Dhabi, UAE, 10-13 April 2016. Submitted, 04 July 2016. Published as resubmitted by the authors on 10 August 2016. 56 http://www.i-jim.org iJIM – Vol. 10, No. 4, 2016 An Educational Ontology-based M-Learning System