 Hello, sir. My MTP topic is adaptive recommendation system for MOOC. In this video, I am going to give you a brief overview of my work. Now, this will be my outline, introduction to the MOOC system. Then I will be explaining about the components of adaptive hyper media. Then there are two important parts in my work, which is learning, learner modeling and knowledge representation. Then the last thing is the proposed system that I am proposing for the new way of adaptive recommendations. Then I will be talking about the implementation status and a plan for stage 2. And then I will give my references. Now, this one is the first slide that is giving you a brief overview of how Coursera's and EDX are working, how they are offering their courses. I think I have explained you many times about this slide. So, now the thing is we need some system which is called personalized feedback here after this, before the deadline, before the deadline of any course, of any week. Now, how we are going to achieve this is through adaptive hyper media. Adaptive hyper media is a collection of all those technologies that enables the adaptation in any sort of applications. Now, in this I am giving you the classification of adaptive technologies. Now, here comes the components of a typical adaptive system. Now, these systems are common in any adaptive application on the web. There are two main entities that are interacting with our system that is designer and the learner, who learns from the system and the designer who designs the courses as well as the rules for the adaptive systems. Now, these are the most important conferences, workshops and journals across the world in this field and among the top UMAP is the first UMAP is the first and the most active conference which are basically concerned on user modeling, adaptation and personalization. Now, these are the most active people across the world who are working in this field. Now, the first part of literature survey is the learner modeling. Learner modeling is basically the understanding of a system about the user. Now, there are there is a difference between learner profile and a learner modeling. Learner profile is basically the learner preferences and learner modeling is the systems understanding. There are two types of information that is stored in the learner modeling which is the domain specific information and domain independent information. Now, this is basically the classification of learner modeling. The classification is based on two types of two types of things which is called the granularity and the task. Now, there are there are various approaches in the past for learner modeling which is Bayesian belief network, machine learning technologies, neural network technologies, fuzzy clustering technique, neurofuzzy technique. The reason why I have studied these techniques is to answer one question that why why we need to make make something new something new. Actually, the main the common thing in all these techniques is the learning and the need of a the need of a large amount of data set on which the system learns about the user. But in our application in our case we do not have that data set. Now, the first thing is the Bayesian belief network. Bayesian belief network there are three three things four things in this Bayesian belief network. First is the knowledge, the second is the action, evidences and utilities that I am giving this presentation for a brief purpose. So, I am not going into the details of Bayesian belief networks. Now, the second thing is the machine learning a user leaves some some sort of patterns if while interacting with hyper media and by analyzing these patterns we can understand the needs of the learner and we can add up the system accordingly. But there is a there is a flaw in machine learning techniques in case of our system in because we do not have these click streams search logs and any sort of data set available for a particular user or a for a particular participant. The only thing that we have is their preferences. Now, the neural network method the same the same limitation is associated with neural network method also. Now, the next thing is the fuzzy logic method actually in the in the machine learning method we derive rules from the data, but in fuzzy logic method we makes the rule on the data. So, and fuzzy logic method are less complex than than machine learning method in in terms of computational complexity. Now, there are two types of clustering hard clustering and soft clustering I am not going into the details. Now, the now how now we need to compare all these methods to convince ourselves that which method is the best in our case. Now, there are there are many parameters like computational complexity dynamic modeling labeled and labeled sort of data size of training data uncertainty and noisy data. Now, I have compared all these techniques all these techniques and figure out that fuzzy logic technique is the best suited technique in our case because we we need dynamic modeling of data and we need a technique which is less complex. Now, these are four types of adaptive task which is called prediction recommendation classification and filtering. Our case we are concentrating on this adaptation task that that is why I that is why I more concentrated on fuzzy logic techniques. Now, these are the previous systems of this kind in the past the second part of my literature surveys the knowledge representation we can we can represent our domain into various forms like like you can see that there are four there are four main types firstly I will be explaining about the initial thinking about the system then I will be explaining about hierarchical representation then goal oriented connectionist knowledge representation and the fuzzy cognitive maps. Now, initially I thought of why do not we divide everything into concepts and and on the basis of the concepts on the basis of the quiz we can just recommend for which for which concept the user marked wrong but actually I was wrong why I was wrong because concepts are dependent concept we can exploit the this dependency to achieve adaptation first thing is find let us take an example see concepts even is a find the sum in the for loop second thing is find the average in a for loop. So, we can see that there is a clear cut dependency between these two concepts. So, we can we can exploit this dependency. Now, this is the example of hierarchical knowledge representation but there are some disadvantages associated with this. Now, I came on to the fuzzy cognitive maps fuzzy cognitive maps are the best suited techniques in our case because I have chosen learner modeling as a fuzzy logic technique and the knowledge representation technique is must be specific to the learner modeling technique that is why I have chosen fuzzy cognitive maps. Now, in fuzzy cognitive maps whole domain is divided into some concept nodes and there is some associativity associated with these two in every two given nodes and there is some weights also associated on the basis of these weights we can figure out how much the concept is dependent on other concept. Now, I will be explaining about the proposed system what we are proposing for recommendation system in MOOCs. Now, first we need to we need to think about what are the requirements the first requirement it should be adaptive the second thing it should be dynamic updateable means it should be updated at a run time the third thing is the it should have low computational complexity and the last thing it should be domain independent domain independent by means is we should not require any sort of experts to adjust the values between two given concept nodes for dependency factors. Now, this one will be the this one will be the two inputs to our system first one first thing is exercises and the quizzes that is attempted by the participants. Now, this was the high level architecture of our system it will be a four layered architecture first like first layer will consist of the lectures lectures are basically the set of concepts that is C1, C2, C3 and there are some sort of objectives associated with each concept which are hierarchically arranged and there are two types of relationships among the objectives which I explain in the next slide and all the objectives are related to the actual course material. Now, these this one is this one is the hierarchical representation of two concepts the first concept is this the second concept is this concept modules there are two types of two types of dependencies first thing is the hierarchical dependent relationship and the second thing is the non-hierarchical relationship. Now, this this I have constructed one example of concept modules there are six concept modules these are these are the objectives associated with each concept module. Now, let us take an example where this one this one is my first concept module this one is my second concept module let us say this one is the this one is four loop and this one is the while loop and let us say objective four of four loop is associated with objective three of while loop let us say objective four is the calculation of average in a four loop and this one is the calculation of average in a while loop. So, if I if I know the calculation of an average in a four loop that means system should understand that the user might know the calculation of average in a while loop how let us say user attempted all those four four objectives right then system will figure out that objective three is most probably most probably known to the user it it will not considered it is as achieved, but it will considered as a most probable candidate for adaptation. Now, I am calculating the distance of this objective to the leaf objective that is two let us say the user marked this wrong then the system will present this objective to the user. Let us say if this if this is marked then the system will understand that system user must know this one objective two and objective one also. So, this one is the basic idea behind adaptation. Now, this is the flow this is the state diagram of every objective has to go through three three states first is unattempted state the second thing is ready state and achieved state. So, an objective can go from unattempted to unattempted then it can go from unattempted to achieved then it can go from unattempted to ready also and ready to unattempted also. Now, these this was the idea behind adaptation, but there is one more thing we need to calculate the knowledge of a particular participant for a given concept model. So, this knowledge level I have defined by some fuzzy functions. So, I am giving the four fuzzy classes unknown partially known and this one is the known and this one is the learned and the membership function for though in each four classes is defined here. Now, there are three steps in my system the first thing is initialization the second thing is how to update the system's assumption about a learner the third thing is the prioritization of the concepts the first thing is initialization how to initialize the system we know that we have only two source of initialization the first thing is the quizzes and the exercises and we have some preferences also. So, the knowledge level for a particular concept is equals to the number of objectives achieved for a for that concept and the total number of objectives in that concept. Now, this one is an example let us say first firstly everything is unknown to the user and after the participant let us say if he or she have attempted some quizzes or exercises the system figure out that the weightage of known classes 0.2 and learned classes 0.8. So, similarly here similarly for C4 and similarly for C5. So, this one is the initialization part the next thing is a prioritization of concept how to prioritize the concept for a given participant. Now, there are many approaches how to prioritize, but we need to compare all those all these approaches. The first thing is the we can apply some topological sort topological sort is nothing, but we need to give a given concept whose n degree is 0 for the dependency factors there is an advantage select those concept modules which are affecting other concepts. We always are giving we always giving only those concepts whose who are making others concept as ready. There are some limitations associated with this. Now, the approach 2 is select those concepts which have a minimum distance minimum distance from the last ready state to the leaf node. There is also some limitations the next thing is consider the knowledge level of a particular concept module as well as the state of the objective the make try to make from unknown to unsatisfactory known to known to learnt. And similarly we have approach 4 which which which can do which can do some some different patterns like unsatisfactory known to known to learnt and to unknown. So, now these are the 4 available approaches with which I am proposing and we need to compare all those approaches. Now, I also surveyed out how to compare these approaches. Now, first let us look at what is the uniqueness of the system? The uniqueness is it is domain independent means we know we do not need any experts to adjust the values between between concept nodes. The second thing is we do not need any data set. The last thing is it is very less complex only one script should be running behind the system and the runtime updation is also possible. Now, how to evaluate these approaches? Now, there are some papers which tells about how to evaluate a given adaptive system. Now, the basic the basic rule of rule from ITS community is is the empirical approaches. Empirical approaches means we can we it is the approach of basically measurement after implementation. The criteria of evaluation of the knowledge representation technique is a mean number of times that the learners it is advised to read the domain it sounds pretty logical and there are some statistical approaches also, but I have not went through the details of these approaches because I was thinking about that first implement the system and after that think about the evaluation. Now, these are the interfaces which I have designed in in python taking edx into consideration because edx is written into django and we can easily integrate these these interfaces with the system. So, this one is the login page and the second will be the user interface. User interface I am I am giving the recommendations in this form. Let us say these are three types of traffic lights green light yellow light and red light green light means you are these concepts are ready to be learnt yellow light means just wait some time and the red light is take caution. So, by giving these this sort of sorting I am giving the recommendation feedback to the user. Now, I have not designed the user interface for for course instructor till now, but I am working on that. Now, this one is the preliminary use case diagram of my application. Now, this this one is a plan for my stage 2 and in June to October I did the literature survey part and find out the problems in existing systems. In November I will be doing the requirement requirement elicitation part the SRS document and implementation plan. In January I will be doing actual implementation the coding part January and February and in April I will be making the text test plans and I will do testing and in May I will resolve issues deploy the system and write a paper for review map if possible. So, these are my references and thank you.