Learning Content Classification and Mapping Content to Synonymous Learners based on 2022 Augmented Verb List of Marzano and Kendall Taxonomy

—Finding suitable learning content for learners with different learning styles is a challenging task in the learning process. Hence it is essential to follow some learning taxonomies to deliver learner-centric learner content. Learning taxonomies are used to express various learning practices and learning habits to be followed by the learner for a better learning process. The investigator has already classified the learners based on the 2022 augmented verb list of Marzano and Kendall taxonomy. The main objective of this paper is to minutely classify the tutor-defined learning contents according to the domains as well as the subdomains of the considered taxonomy which is in text format. Providing personalized learning content could help the learners for a better understanding of learning content and their interrelationship which in turn produce better learning outcomes. Mapping the six levels of learning contents into the corresponding learner is a challenging task. Hence the investigator has chosen seven algorithms including Bagging, XG Boost, Support Vector Machine from Machine Learning and four algorithms including Convolutional Neural Network, and Deep Neural Network in Deep Learning algorithm to classify the learning contents. The experimental results indicate that Support Vector Machine performed well in machine learning and Deep Neural Network yields good performance in deep learning in the learning content classification process. These micro contents were organized using a property graph. Further, the micro contents were retrieved from the property graph using SPARQL for mapping the classified contents to the corresponding learners to achieve personalization in the learning process.


I. INTRODUCTION
Learning is a process of adapting changes in personal and professional to ameliorate the quality of life. According to Stephen Hawking, Intelligence is the ability to adapt the change. Acquiring intelligence, absorbing, adapting and storing new information in memory is uneven among the learner. Hence it is the need of the hour to identify different learning characteristics of the learner to achieve a better learning outcome. The resource used to provide knowledge to the learner is known as learning content. According to the learner's preference and learning styles, learning content has to be provided to the learners. This process is called personalization in the learning process [1]. Personalized learning must pass some control over the learners, providing some input into how they progress through their learning activities. This can be achieved by adapting learning taxonomies in the learning process. Various taxonomies were developed by researchers in the field of Education and Learning since from the year 1956 [2]. This research work adapted Marzano and Kendall (MK) taxonomy to determine the learning style of the learners. MK taxonomy model provides better knowledge about certain fundamental processes in learning, such as emotion, memory, motivation and metacognition. This model also provides greater precision while creating learning objectives, having a more specific map of the types of knowledge that can be acquired and how they are acquired. Due to this greater precision, it is also possible to evaluate more easily [3]. MK taxonomy has six domains from lower order of thinking skills to higher order of thinking skills.
The investigator prepared the questionnaire based on the 2022 augmented verb list of MK taxonomy to find the learning style of the learner and classified the learners into six domains and 22 sub-domains of the considered taxonomy To classify the learning contents into micro contents the same taxonomy has to be utilized.
Text-based learning content was pre-processed to provide good interpretation and usage. It can also reduce the redundancy in the text content. After the content were preprocessed, it has to be classified based on the considered taxonomy. To accomplish the classification of learning contents into the micro-content process, the investigator has chosen seven algorithms from Machine Learning models such as *Corresponding Author. www.ijacsa.thesai.org The classified micro contents were stored in file format. And these micro contents can be represented using Property Graph also called a labelled property graph since it contains nodes(entities), edges(relationships) and properties(attributes). This research work creates ontology for MK Taxonomy to provide learning contents based on the weightage. In ontology, individuals are created for each micro-content with the annotation properties of learning content, keywords and file size. Once all the terms are arranged, the data can be retrieved using the SPARQL query. The representation in the property graph is visualized using the OWLGrEd Visualization tool.
Further, the researcher evaluated the performance of each model and compared them according to precision, recall, F1 score and accuracy. Classified micro contents obtained from the classifiers were mapped to the synonymous learners based on the maximum score on the accuracy of the model. The rest of the paper is systematized as follows. Section II provides an overview of the related works. Section III provides the design and methodology of the proposed method. Section IV expresses evaluation and results and discussion. Section V illustrates the way to represent the learning content organization using a property graph and the method to extract the contents using SPARQL. Section VI discusses mapping the learning micro-content into the corresponding learner according to six domains and 22 sub-domains of the considered taxonomy. Section VII presents the conclusion and Section VIII illustrates the case study of the proposed method.

A. Action Verbs
To express the noticeable behaviour of the learner the learning objective must start with action verbs. Action verbs were used to monitor the learner and the throughput of the learning objectives. Choosing the right verb for different types of the learner is an art [4]. The verb list of Marzano and Kendall Taxonomy was first published in 2007 and it needed an up-to-date update to include the later verbs. This is because the recent education system utilizes new vocabularies as per the current technology. The existing action verb list in the taxonomy may not be fulfilling to achieve the throughput of learning objectives. Hence it is the need of the hour to augment the verb list of Marzano and Kendall Taxonomy.
Augmentation is achieved by gathering suitable verbs from sixteen existing taxonomies and open domains. Hence the researcher has made an exhaustive search to update the verb list from 95 to 360 verbs as shown in Table I.

B. Adaptive Learning Path and Contents
A learning path is a progression of activities and concepts to be chosen by the learner to construct their knowledge or skills in a specific area. Traditional learning system provides the same content and learning path to all learners. But the learner's knowledge, circumstance, and preference are different, and their performance and satisfaction may decrease if they have been given the same content and learning path [5]. Presenting learner-centric learning content can ameliorate the effectiveness and performance of the learning process. To achieve this goal the researcher presented a new model as shown in Fig. 1. www.ijacsa.thesai.org These micro contents can be represented using Property Graph. Then the micro-content will be retrieved using SPARQL from the property Graph. Further, these micro contents are mapped into the corresponding learners according to six domains and sub-domains of MK taxonomy based on the descending order of the size of the micro contents.

C. Existing Content Classification Approaches
Learning Contents can be in the form of images, text, audio, animations, and video. Text contents can be classified as rule-based, supervised learning-based, and combined classifier-based approaches.
A set of handcrafted rules are utilized in the rule-based approach. In supervised learning text classification approach classification made based on learning past observations. The combined classifier utilized both a machine learning trained base classifier and a rule-based classifier for showing improvement in the throughput [6]. Table II illustrate   Learning style is the strategy to accommodate receiving and processing the received information which are two phases of learning [17]. The process of recognizing the behaviour of the learner then spontaneously generates a natural learning path, and tailoring the learning contents to an individual learner is known as adaptation in learning which is the prime need for personalized learning [18]. Learning taxonomy can be employed to understand the learning levels of the learners scientifically. Hence this research focused on classifying the learning contents based on a learning taxonomy for betterpersonalized learning.
This research focused only on text learning contents. These learning contents were preprocessed and classified based on the augmented verb list of MK Taxonomy into micro contents. Then the suitable micro contents were assigned to the corresponding learner to accomplish the personalized teaching-learning process. Fig. 2 depicts the design architecture for text-based content classification. The design contains two main modules a pre-processing module and a Classification module.

Micro Contents based on MK Taxonomy Verbs Text Content
Text Pre-Processing Content Classification www.ijacsa.thesai.org The pre-processing module is the first module of content classification. Text pre-processing is essential for eliminating all the irrelevant objects from the data and making it ready for further processing. This is because raw text data might have insignificant text which makes it difficult to understand and investigate. Hence proper pre-processing must be implemented on raw text data [19]. This research work utilizes three pre-processing techniques as Case Conversion, Stop word removal, and Tokenization.
Case Conversion: Converting all the text content into the lower case is utilized to discard unproductive words [20].
Stop Word Removal: Articles, prepositions, pronouns, and conjunctions in any language are called stop words. "The", "a", "an", "so", and "what‖ are examples of stop words in English. Removal of such words would help in the size reduction of a dataset and further the training time can also be reduced due to the lesser number of tokens involved in the training [21].
Tokenization: Splitting text contents into smaller units is known as tokenization. The individual units are called tokens. Tokens can be words, phonemes, or maybe full sentences [22]. This research work utilizes sentence tokenization. The learning contents were divided into sentences and considered tokens.
In the e-learning environment, a large volume of learning materials was available in various formats. But it is necessary to provide appropriate learning content to the respective learners according to the six domains and sub-domains of MK taxonomy. This study utilizes NetBeans IDE open-source integrated development environment using Java and libraries such as NLKT, pandas, TensorFlow, NumPy, sklearn, text blob, and seaborn for the classification process.

IV. RESULTS AND DISCUSSIONS
For the evaluation process, this study has selected a dataset from the responses received from hundred students for three different learning contents. In a convenient sampling technique analysis on data-set can be carried out either by taking multiple sampling or by repeating the survey. The researcher adapted multiple sampling techniques to produce a reliable result.

A. Evaluation Metrics
Accuracy, precision, recall and f1-score are the measures for evaluation utilized by this study for understanding, measuring relevance and correctness of classification of learning content into micro-contents. Accuracy is used to check the correctness of the model. The exactness of the results is expressed by precision. The completeness of the quality of the results was measured by a recall. F1-score is the weighted average of precision and recall. F1-score is used to evaluate the binary classification system [23].
The evaluation of this study was performed based on the number of keywords classified per domain of MK Taxonomy. A maximum of ten keywords were considered for the classification of learning content into micro-content in each domain of the considered taxonomy.

B. Experiments
The experiments were conducted both on machine learning models and deep learning models and categorized into two. The overall accuracy of this classifier is 80%. The percentage of all the measures will be incremented if the number of keywords is increased in each level of MK Taxonomy. Table IV represents the evaluation metric for the Bagging classifier in the machine learning model.      The evaluation metric for Deep Neural Network is illustrated in Table XI. It is observed that the DNN classifier successfully classified the contents because the accuracy of the classifier is 83% which is high score than the remaining classifiers considered in this study.
The evaluation metrics for CNN and RNN were represented in Tables XII and XIII.    The overall score obtained by all the classifiers using both machine learning and deep learning models were illustrated in Table XVI.  Table XVII shows the Accuracy measure values obtained for each classifier and arranged in descending order based on the percentage of Accuracy.
According to the results, the SVM classifier performed well toward the correctness of classification and the accuracy is measured as 86 per cent. Further, the F1-score, the weighted www.ijacsa.thesai.org average of precision and recall is also 86% as shown in Table  XVIII. The analysis based on accuracy is depicted in Fig. 5.   As per the analysis, it is observed that the Naïve Bayes classifier achieved a considerable value in precision. The higher precision indicates that, less false positive measure. It shows the exactness of the classification of learning contents.  The completeness of the quality of the results was measured by a recall. SVM classifier again occupies the top place among other classifiers for the completeness of the classification of learning content according to keywords of MK Taxonomy.  According to the above analysis, this study concludes that the SVM classifier model provides more accuracy. Hence the micro-contents classified by utilizing the SVM classifier are considered for mapping to the synonymous learner based on the verb list of MK Taxonomy.
(IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 13, No. 8, 2022 378 | P a g e www.ijacsa.thesai.org V. LEARNING CONTENT ORGANIZATION AND RETRIEVAL Information can be represented in the form of tables, charts and graphs to support organizing, analyzing and fetching them when required. Graphs are shines well in representing the connections and relationships among diverse data.
A graph in which the connections between nodes are used to represent the relations along with name and some properties is called Property Graphs. Nodes, labels, relationships, and properties are the components of a property graph. Relationships and connecting nodes of data are capable of storing properties. Appropriate and easy-to-recognize labels are utilized by the property graph for modelling data and its connections. This structured form of data can be easily understood by laymen [24].
This research work utilizes the property graph to represent the classified micro contents according to MK taxonomy.
Visualization of six domains and sub-domains and the corresponding micro-content of MK Taxonomy using OWLGrEd is depicted in Fig. 6. OWLGrEd is an editor for OWL to represent graphical notation in an ontology. Ontology becomes common in the fields of artificial intelligence and machine learning, where knowledge plays a vital role. Ontology creates a generic vocabulary which can be shared with researchers and different stack holders. It consists of machine-interoperable definitions of the domain concepts and the relationship between them. It enables the researcher to retrieve data based on knowledge which is known as knowledge-based retrieval. Ontology shares the knowledge to understand the structure of information which can be reused from the domain knowledge. This feature motivates to development of learning content to enhance the learner to gain knowledge of the subject based on their interest and learning ability. Ontology is a collection of classes, properties, instances and axioms. Classes are also known as the concepts of the domain, properties define the relationship between the concepts, instances are the individuals of each class, and axioms denote the restrictions. Ontology can be defined as, a formal explicit specification of a shared conceptualization'.
The key terms of a domain are identified and arranged hierarchically and the relationships between the terms are established before developing ontologies.
This research work creates ontology for MK Taxonomy to provide learning contents based on the weightage. The levels and sublevels of the considered taxonomy are arranged as classes hierarchically to frame ontology using the Protégé ontology development tool.
The level/domain of learning is identified through the keywords used in the learning content. Each level of MK taxonomy contains a different set of keywords to group the learning content. The keywords are listed as individuals and the relationship between the classes and keywords is established.
The learning contents were partitioned into micro contents to improve the learning ability of the learner. In ontology, individuals are created for each micro-content with the annotation properties of learning content, keywords and file size.
Micro-content (MC) can be represented as In this study, MC 11 represents a micro-content in the subdomain Recognizing in the domain Retrieval. Each microcontent is defined with these annotation properties to retrieve the content based on the file size given in Fig. 7.
Each micro-content is related to the type of class and object property it belongs. Variable content is created to hold the value of micro-content. Once all the terms are arranged, the data can be retrieved using the SPARQL query.
The SPARQL query to retrieve the micro-content based on the file size in descending order is given below and the result is shown in Fig. 8.   The classified learning contents which are retrieved from the graph were mapped to the corresponding leaner to achieve personalization in the learning process. Each micro-content with learning content, keywords and file size defined with annotation properties was used to retrieve the content. Further, these retrieved micro-contents were arranged in descending order based on the size of the files. Based on the score obtained by the learner, they were classified by the researcher. The micro-content classified based on the keyword under the sub-domain Recognize in domain Retrieval is MC 11 .

SPARQL query to Retrieve the micro-content
This study proposed a novel method to perform the mapping process. The learners' characteristics were obtained by the response received from them through the tool questionnaire according to the 2022 verb list of MK Taxonomy. Questions were rationalized to 50 according to six levels of MK Taxonomy as 8,8,10,8,8,8 which can be considered as weightage (w) for each domain as shown in Fig. 9. Eight questions in D1 in turn sub-dived into 3, 3 and 2.
Dataset has been constructed from the response fetched from the hundred learners. The correct response was represented as 1 and the incorrect response was represented as 0. Further, the total score against each domain was calculated as illustrated in Table XXI. This provides a way to quantify each type of learning style in the learner.
Based on the score (SC) obtained by the learner out of each domain and sub-domains of MK Taxonomy, the number of micro-contents (NMC) retrieved from a graph as per each domain, and file size are the parameters for providing micro contents to the synonymous learner. Equation (2)  The MCs were arranged in descending order based on the file size. Hence as per the above calculation, the MCs were mapped to the synonymous learners to achieve personalization in the learning process according to MK Taxonomy.

VII. CONCLUSION
The main objective of this paper is to specifically classify the learning contents based on the specific characteristics of the learner and according to the domains as well as the subdomains of the considered taxonomy. The learning contents in text format were represented in a property graph and retrieval of the same is achieved to fulfil the personalization process in the learner-centric environment. The learners were classified according to MK Taxonomy. Hence the classified learning contents were assigned to the synonymous learners to achieve personalization in the learning process. www.ijacsa.thesai.org Many researchers classified the learners based on Bloom's Taxonomy's cognitive level. But this research work proposed a novel contribution towards the classification of learning contents into micro contents according to the six domains and 22 sub-domains of MK Taxonomy and represents them using a property graph. Further, these micro contents were retrieved from the graph and mapped to the corresponding learners who were classified according to MK Taxonomy. Hence the learner-centric learning contents were provided to the learners for better learning outcomes.

VIII. CASE STUDY
Learning Contents classification can be carried out by the following steps. Fig. 9 shows the Screenshot of the learning content.
Input: Subject: Operating Systems-Tutor defined Text Contents.
An Operating System is recognized as an intermediate between the user of the computer and computer hardware. Important functions of an operating system are identified and listed below. Step 1: Text content Pre-Processing Step 1.1: Case Conversionconvert into a lower case an operating system is recognized as an intermediate between the user of the computer and computer hardware. important functions of an operating system are identified and listed below.
Step1.2: Stop word Removal the operating system recognized intermediate user computer hardware. Important function operating system identifies list.
Step 1.3: Tokenization (Sentence Tokenization) Token 1: operating system recognizes intermediate user computer hardware Token 2: important function operating system identify the list.
Step 2: Verbs are Extracted from the tokens.
Verb list: recognize, identify, list Step 3: Classification based on the verb list according to MK Taxonomy domains and sub-domains using seven ML models and four DL models as shown in Fig. 2. As per the performance metrics, the SVM model is used to classify this study. Keywords or the action verbs in MK Taxonomy were utilized for the classification of tokens into micro contents.
Action verbs ‗recognize' and ‗identify' the sub-domain Recognizing in domain Retrieval. Hence the corresponding MCs were assigned to that sub-domain.
Output: MC 1: operating system recognizes intermediary user computer hardware. MC 2: Important functions operating System identify list.
Step 4: These MCs were represented in the property graph as illustrated in Fig. 8 and retrieved using SPARQL.
Step 5: Mapping the MCs to the synonymous learners.
The total number of MCs in Sub-domain1 Recognizing in domain Retrieval were 02. These two MCs were to be mapped to the learners who were already classified under the same sub-domain as shown in Fig. 9 and the score obtained by the learners as shown in Table XXI  Hence two MCs were provided to the learner in Sub-domain1 Recognizing in domain Retrieval according to MK Taxonomy in a personalized manner.