Course talk:CPSC522/Improve recommendation system by integration

From UBC Wiki

Contents

Thread titleRepliesLast modified
Critique206:58, 22 April 2016
My comments and critique:106:50, 22 April 2016
Critique106:49, 22 April 2016
Suggestions106:48, 22 April 2016

Hi Arthur,
Good work so far! You did a good job presenting your work today. It made understanding your page a lot easier after seeing you present.
Here are few comments/suggestions:

  • In the 'Result' section of your page, you mention that "Content features do help to improve the nearest neighbour algorithm accuracy becasue of its ability to improve the cosine similiarty calculation." I know you have a link to the wikipedia page for cosine similarity but it would be better if you briefly explained what cosine similarity is and how content features help improve the nearest neighbour algorithm’s accuracy.
  • From your result, the RMSE value for integrated data is higher than the RMSE value for original data in general. Isn't RMSE supposed to be minimized? Also, can you give me an insight into why RMSE increases with the number of neighbours? Please correct me if I'm wrong.
  • You should also have your references in your wiki page and cite relevant sections in the text.
  • And finally, I’ve noticed a lot of typos and grammatical errors in the page. Please try to fix them.


Thanks for your page. I enjoyed your work.

Best regards,
Adnan

AdnanReza (talk)06:00, 22 April 2016

Hi Adnan

Thanks for your review and suggestion. for the RMSE, I previously had the wrong result because during the coding process, I miscoded the column and row for their sparse matrix calculation so that the respective row are multiplied with respective row and resulted the wrong answer. However, I have already debugged it and rerun the experiement. The RMSE for the integrated one is better as you can see on the page as well as during today's presentation.

For the reason why RMSE increase along with the neighbour, I think it is because the more neighbour you have, the more content you have. However, the matrix will become increasingly sparse. And when the postive accuracy of more content is counter-balanced by the negative content effect: so called the sparse matrix effect, then you will have worse RMSE.

Regarding the typos, can you take an example? I have already used grammarly.com to help check the errors. maybe it just doesn't work...


Arthur

BaoSun (talk)06:39, 22 April 2016

From your results section:
"Content features do help to improve the nearest neighbour algorithm accuracy becasue of its ability to improve the cosine similiarty calculation.
There are few similar typos, but nothing to worry about. Your content is good. Keep up the good work.

AdnanReza (talk)06:58, 22 April 2016
 
 

My comments and critique:

Hi Arthur,

Firstly, let me thank for your project and your contribution in CPSC 522 Wiki pages. I like your page and find your topic of project interesting. I like to share my opinion and give you my comments, maybe they might help you.

1) First of you can treat this assignment's wiki page as a research paper and include the usual and common sections such as: abstract, motivation, introduction, method, evaluation, results, conclusion, future work and acknowledgment. 2) Maybe snapshots are not the most suitable way of presenting data a figure in my opinion. 3) Maybe put all the codes and pseudo-codes in file or link and just give a pointer link to that. 4) Maybe add some areas for future work or possible areas of improvement. 5) Finally, I really want to see the reference section. it really adds value.

Thanks for your page it was a pleasure to read it and I hope these comments can help you improve your contribution.

Mehrdad Ghomi

MehrdadGhomi (talk)02:28, 21 April 2016

Hi Mehrada,

I have already reviewed the format and changed according to your advisefor the 1,2,3,4,5 points. Thank you for your advise to make my paper more professional!

Regards Arthur

BaoSun (talk)06:50, 22 April 2016
 

Hi Arthur,

Nice work indeed. But I think it will be better if you include what does the results suggests/it's impact. You should also label the axes of the plots.

TanujKrAasawat (talk)02:38, 21 April 2016

Hi Tanuj,

Done already.

Thanks

BaoSun (talk)06:49, 22 April 2016
 

Suggestions

Hi Arthur,

Very nice page. Here are some suggestions:

Background For "Collaborative Filtering" section, a figure of the relationships between users/items would help understanding the model. For “Content-based Filtering” section, it would be better if you can give a figure of the user profile (the two-dimensional table). Reference for “Inverse Document Frequency” could be added.

Experiment If I understand correctly, RMSE is the value to be minimized. From your result, the RMSE value for integrated data is higher than the RMSE value for original data in general. Could you tell me if there’s something I missed? What do you think could be the reasons of the fluctuation shown in the original dataset? I.e., why it is not increasing as the number of neighbours increases? Also why would the RMSE steadily increases as the number of neighbours increases?

Other Could you give a more explicit explanation on the figure “Architecture overview of Integration Recommendation System”? More specifically, what do the arrows mean? What are the input/output of each part? Thanks again for your page.

Bests,

Yu Yan

YuYan1 (talk)08:42, 21 April 2016

Hi Yu,

Thank you for your time for viewing my paper.

For the background part, I have already inserted corresponding pictures to illustrate the meaning.

For the RMSE, I previously had the wrong result because during the coding process, I miscoded the column and row for their sparse matrix calculation so that the respective row are multiplied with respective row and resulted the wrong answer. However, I have already debugged it and rerun the experiement. The RMSE for the integrated one is better as you can see on the page as well as during today's presentation. For the reason why RMSE increase along with the neighbour, I think it is because the more neighbour you have, the more content you have. However, the matrix will become increasingly sparse. And when the postive accuracy of more content is counter-balanced by the negative content effect: so called the sparse matrix effect, then you will have worse RMSE.

Regarding the "Architecture overview of Integration Recommendation System", it is a just flow chart and you can refer to the system overview to get a better understanding for the whole system. I should update this workflow because I have already changed the overview already. I combined both training and testing together to feed the final collaborative filtering in a unified matrix.

Regards Arthur

BaoSun (talk)06:48, 22 April 2016