ICACT20220190 Question.1
Questioner: jypark9740@g.skku.edu 2022-02-14 오후 9:40:36 |
ICACT20220190 Answer.1
Answer by Auhor quandh@vnu.edu.vn 2022-02-14 오후 9:40:36
Chrome  Click!! |
| The exponential growth of digital information in the recent decades often referred to Big Data, poses new challenges for the current RS. What relevant research is going on to solve this problem? |
Big data have inspired many researchers to improve the performance of recommender systems for handling large-scale datasets. Some relevant research in this area can be mentioned as follows. (Hammou, et al., 2019) used matrix factorization and random forest on big data recommendation systems that focuses on dividing the big data problem into smaller subproblems to speed up the process. (Verma, et al., 2015) explained how to use a recommendation system with big data. The paper proposed using CF, CBF, and hybrid filtering on Hadoop framework. Their result showed that with properly applied algorithms, the program's running time only increases slightly while the amount of the data is grown significantly. (Maximilian, et al., 2018) evaluated the effect of Big Data on recommendation quality by experimenting with the performance of a regression model on the example of Internet Search. |
ICACT20220190 Question.2
Questioner: xiaxuewena@gmail.com 2022-02-16 오후 5:06:01 |
ICACT20220190 Answer.2
Answer by Auhor quandh@vnu.edu.vn 2022-02-16 오후 5:06:01
Chrome Click!! |
| Thank you for your analysis and discussion of the Fast Clustering-based Recommender System for Big Data. Do you have any suggestions for future applications? |
With the exponential growth of the digital information in the recent decades, development of RS in Big data environment is an inevitable trend. Besides finding effectively ways to store and manage large-scale datasets, one concern is the development of parallel recommendation methods. |
ICACT20220190 Question.3
Questioner: tomayoon@ieee.org 2022-02-16 오후 5:10:31 |
ICACT20220190 Answer.3
Answer by Auhor quandh@vnu.edu.vn 2022-02-16 오후 5:10:31
Chrome Click!! |
| First of all, thank you for the good research paper and presentation contents. I would like to know more about the research environment used in this study, and also in detail about the equipment & operating systems, development languages and software libraries you used. |
In the software configures, we use Spark 3.1.2, Hadoop 2.7.7, and Python 3. The system was built in virtual machines with the hardware configures as follows.
Master RAM-8(GB) Disk-60(GB) Ubuntu 18.0.4
Worker-1 4(GB) 30(GB) Ubuntu 18.0.4
Worker-2 4(GB) 30(GB) Ubuntu 18.0.4
Hope it helps.
|
ICACT20220190 Question.4
Questioner: tomayoon@ieee.org 2022-02-16 오후 9:42:34 |
ICACT20220190 Answer.4
Answer by Auhor quandh@vnu.edu.vn 2022-02-16 오후 9:42:34
Chrome Click!! |
| Will you tell us what is the major achievement of this paper, and also what is further study issue from this? Thanks in advance!! |
Thanks for your question. The main achievement of our work is development of a fast recommendation method (i.e. the training time is 3 times faster and the average inference time is almost 133 times faster than traditional USER-Collaborative Filtering). The method, besides, can address the item cold-start problem well. However, as a trade-off, the MAE and RMSE of the method is slightly increased. This will be our future direction to design new clustering-based methods to not only achieve a fast execution time, but also lower the error is. |
ICACT20220190 Question.5
Questioner: rna0415@g.skku.edu 2022-02-16 오후 9:48:17 |
ICACT20220190 Answer.5
Answer by Auhor quandh@vnu.edu.vn 2022-02-16 오후 9:48:17
Chrome Click!! |
| Thank you for your work! I could see your motivation, and experiment with evaluation. I just want to ask you a little more detail about the cold-start. For example, terms of background, or scenarios. Thank you in advance |
The item cold-start problem appears when new items arrive with no or very few interactions. In our experimental dataset, there are 141 movies received only 1 rating. Among them, we select items appeared in the test set, but not in the training set to perform experimental comparisons. By doing so, we can test with those completely new items since they are not observed in the training set. Hope this answer helps. |
ICACT20220190 Question.6
Questioner: duongld11@viettel.com.vn 2022-02-16 오후 9:52:04 |
ICACT20220190 Answer.6
Answer by Auhor quandh@vnu.edu.vn 2022-02-16 오후 9:52:04
Chrome Click!! |
| Thank you for the nice video. I would ask you how is the performance of this method when working with other datasets? |
Thank you for your question. One limitation of our work is the experimental comparisons on only one dataset. In our future work, we will expand with different datasets, and experiments on larger datasets will be considered. |
ICACT20220190 Question.7
Questioner: namacabale@gmail.com 2022-02-16 오후 10:03:29 |
ICACT20220190 Answer.7
Answer by Auhor quandh@vnu.edu.vn 2022-02-16 오후 10:03:29
Chrome Click!! |
| Dear Author, you have a very good research. Is the technology today is capable of implementing your proposed scheme? Is there an existing technology almost similar to your proposal that has been deployed in real world environment? Thank you and stay safe. |
Thanks a lot for your interesting question. First, Big data is now an inevitable trend. Many have been storing large amounts of data that arise in their business processes. Not a few of them utilize different Big data frameworks such as Hadoop, Spark to store and query the information. Recommendation in Big data environment is the current step in some global famous companies like Amazon, eBay, Facebook, or the next step in some E-Commerce companies in our country like Tiki.
Secondly, for the development of recommender systems in Amazon, you can find more details in this following paper: https://assets.amazon.science/76/9e/7eac89c14a838746e91dde0a5e9f/two-decades-of-recommender-systems-at-amazon.pdf |
ICACT20220190 Question.8
Questioner: vuvietvu@vnu.edu.vn 2022-02-16 오후 10:51:29 |
ICACT20220190 Answer.8
Answer by Auhor quandh@vnu.edu.vn 2022-02-16 오후 10:51:29
Chrome Click!! |
| Thanks for your work. I have a question that what factors your clustering method based on. |
We performed clustering on users. To do so, each user will need to be represented as a feature vector - called a user profile. Particularly in our experimental dataset - MovieLens, we based on the average rating that the user has given for each movie genre to form the user's feature vector. |
