Recommender System Important Questions (CCS360)

[0:00]Hi friends, welcome back to because. In this video, we'll see the important questions of the 6th sem elective paper CCS360 Recommender System, friends. For this subject, we don't have previous regulation, so there are no previous year question papers. The important questions I'm telling are all based on my experience and what I know about the topic, just a guess, friends. I haven't gone in-depth into the subject, looked at the books, or line by line. Based on a general overview, there's 70% theory and 30% numerical/programs. If we look at passing, like, you won't get proper notes for it, so you can only study from the book. There are no previous year question papers, so it's a medium difficulty paper to pass. But for understanding, it's a very easy paper, friends. Compared to other electives, Recommender System is very easy to understand, and we can connect it with our real-life. There are many things to score marks, so it's a medium difficulty paper. If you study well, you can definitely score good marks, no problem with that. So, the main points, as usual, are that there are no previous year question papers here either, and everything is just based on my intuition. Also, I don't have notes for all subjects, so please don't comment for notes. So, for the textbook, which book should we refer to? Charu C. Aggarwal's Recommender Systems: The Textbook, Springer, 2016. This covers the majority of everything, friends. The few things not covered here will be covered in the books below. So, if you look at local authors, they would have combined content from here and content from the internet. But if we look at your evaluation answer key, we will have the key points and diagrams from this textbook. So, Unit 1. In Unit 1, if we look at it broadly, the most important things are similarity measures, what different techniques we use, then dimensionality reduction, and then the very, very crucial part, Singular Value Decomposition (SVD). You must definitely study this Singular Value Decomposition. Besides that, study dimensionality reduction and similarity measures. Then, if you have time after studying those, you can look at the highlighted parts in blue: traditional and non-personalized recommender systems, how they worked, and then data mining: how we mine data for recommender systems. Study those techniques. Then, lastly, you can look at the basic taxonomy of recommender systems. But SVD is something you must definitely study. Then, moving on to Unit 2. In Unit 2, friends, the recommender system itself has two types: content filtering and collaborative filtering. So, Unit 2 is completely based on content filtering. So, I can't say this is important or that is important. Initially, you will study high-level architecture. After studying high-level architecture, then you will study how to create profiles for the items we have, and then how to create profiles for users. Once you have item profiles and user profiles, how do we compute similarity between the two? So, we will study all this. So, from the high-level architecture to the user profile, everything must be definitely studied. You can omit the rest if you want, but studying them is a very good thing, friends. Only then will content filtering be fully complete. Otherwise, it won't be fully complete. Then, in Unit 3, we will study about collaborative filtering. The main things to study here are what collaborative filtering is. How does user-based collaborative filtering work, and how does item-based collaborative filtering work? These two are very, very important questions. After finishing those, then the rest of the things. Why do we normalize our rating, how do we normalize it? How do I get similarity weight? Then, how do I select the neighborhood? You can look at all these things after finishing the first two questions, that's the second part. Then, regarding Unit 4, it's mainly about how attacks happen on recommender systems, what types of attacks there are, and how we detect those attacks. There are two types: individual attack and group attack. So, the first half, we study about attack types, what attacks there are, and then how we detect them. That's my first half. What is the second half? It's about what I can do to prevent those attacks. So, how can I design a good, strong recommender system so that none of the attacks impact my recommender system? So, how can I block my recommender system from being impacted? So, at least if you study the first half fully, or the second half fully, there's a high probability of questions coming, friends. Especially from the first half, there's a definite, sure shot probability of a question coming if you study up to group attack. Then, coming to the last unit, Unit 5. In Unit 5, we have constructed the recommender system, and we are going to evaluate that recommender system. So, how is its performance? How can we evaluate it online? How can we evaluate it offline? Then, when evaluating, what is my objective? What is my goal? What final decision am I going to make? That is, how am I going to evaluate it, we study in Unit 5. Then, in the second part, we study what type of issues arise when designing a recommender system, like scalability. Then, there will be different different issues. Then, how do I measure accuracy? Then, what are the limitations in the measures I use? This is the second half. In the second half, what do we study? We study about design issues, what problems there are in accuracy metrics, and what metrics. Then, lastly, what are the limitations in the evaluation measure I use? This is the second half. Either study the first half or the second half completely. Both is well and good. But you must definitely study one half, friends. So, that's all for this subject, friends. If you have any doubts, comment. We'll see in the next subject. If you like the video, share it with your friends. Thanks. Bye.