From CSULA CS Wiki
Jump to: navigation, search

Contents

Jan 5 (week 1)

This week was the introductory class for CS 590 where Dr. Abbott gave an idea about the course.

Jan 12 (week 2)

This week I was trying to understand what was expected out from the students who are taking non Thesis Option. For the course, I would be working on a Platform and find more information on the same. I was thinking on platforms like FLEX, AJAX or ADOBE AIR to work on. Finally, I ll give a presentation next week on AJAX and how JavaScript manipulates DOM.

Jan 19 (week 3)

This week I was looking into how AJAX works by manipulating the DOM. I will be presenting the same in class.

PPT demo

Jan 26 (week 4)

This week I met with Dr. Kang and Dr. Wagstaff on the possibility on building a recommender system for my Masters Thesis Project. The professors seemed interested in the topic but want me to be more specific about the project. To understand the problem better, this week I was gathering information on Recommendation Systems and Collaborative Filtering from the web.

To start off with, I read the paper titled "Clustering Items for Collaborative Filtering" by Mark O’Connor & Jon Herlocker from Dept. of Computer Science and Engineering, University of Minnesota. The paper talks about data partitioning and clustering data for collaborative filtering. This technique reduces one big chunk of data space into a set of smaller spaces and hence leads to fewer data points in one cluster. Initially the author suggests that this technique will decrease the time to compute a prediction since there is less data to consider. The paper compares four well-known clustering/partitioning algorithms namely,

    Average link hierarchical agglomerative, 
    ROCK (A Robust Clustering Algorithm for Categorical Attributes), 
    kMetis and hMetis (Multilevel k-way Graph Partitioning). 

Even though it makes logical sense that clustering and data partitioning will increase prediction accuracy, the paper did not find this to be the case consistently in their experiments.

Feb 2 (week 5)

This week I was doing some more research on the Recommender Systems and met Dr. Wagstaff to discuss on the same. For the presentation part, I realized that currently my knowledge is not enough to give a presentation on RS. So, I decide to present a small animation technique built in javascript and css. In the presentation I ll give a demo on what the script does and also, how does it work.

Just to note, I was planning to show how the following libraries work but for last 1 week, their website is down. http://script.aculo.us/

PPt demo

Feb 9 (week 6)

This week I had a long conversation with Dr. Kang on recommender systems and my view on the Master's Thesis Project. She had very similar question to what Dr. Abbott had mentioned on the first day of class that needs be answered in the proposal. I mentioned to Dr. Kang that I am not looking to develop or create something new with this project, at least not as of now, but develop multiple solutions that exist on the problem of recommender system and compare their performance in computation time as well as recommendation accuracy. I also spoke to some of the e-commerce companies around LA area to see if they would be interested in sharing their sale data for my project which would be a very important aspect for me to proceed with this. I am hoping I ll have a response from them by the end of next week. Also, I continued researching on the project on Google and compiling a list of algorithms that people have used to solve this problem. Also, I found some good information about the annual recommender system conferences started in 2006.

Feb 16 (week 7)

This week I discussed Recommender Systems with Dr. Wagstaff for being my advisor for the thesis project. Also, I studied the paper Recommender Systems for Large-scale E-Commerce: Scalable Neighborhood Formation Using Clustering which deals with clustering in the field of recommender systems and proposes a solution for a system which has to deal with a large data set like any e-commerce website. The paper talks about the effectiveness of the collaborative approach for recommender system and also the drawbacks like Sparsity and Scalability. The paper then talks about the Clustering approach which reduces the sparsity of the data set and also is faster in prediction generation. The paper finally compares collaborative filtering with clustering on the basis of Mean Absolute Error (MAE) and in general the paper finds out clustering out performing the collaborative filtering.

Feb 23 (week 8)

This week I read a paper on Automated Collaborative Filtering (ACF) called Understanding and Improving Automated Collaborative Filtering Systems. As obvious from the name, the paper talks about ACF in general, the success for the technique on ACF in Recommender systems and the future work to improve it. The paper discusses the neighborhood approach to predict the recommendations and evaluates it performance. For evaluating performance, the author discusses the techniques like Mean absolute error, Precision and Recall,ROC area, Spearman rank correlation, Kendall’s tau-b correlation, Breese’s utility metric and ndp. The author discusess which evaluation method would be more useful in which case.

Mar 1 (week 9)

This week I started working on my Prospectus for Recommender Systems. I also created a small presentation for the class on recommender systems and go over this prospectus in the class too. ppt

Mar 8 (week 10)

This week I continued working on my Prospectus for Recommender Systems. I added all the comments I received from Dr. Abott. Also, I added some of the algos I would be using. The exact papers for these algos would be determined after I meet with Dr Wagstaff.

Mar 15 Prospectus

This week I got an approval from Dr. Pamula for this thesis. The prospectus can be found here.