Archive for the ‘recommender systems’ Category

Should we hang out with people we don’t like

Tuesday, February 17th, 2009

Homophily-based algorithms may not be good for us (previous post). Few points from the Guardian Week:

  • The faintly depressing human tendency to seek out and spend time with those most similar to us is known in social science as “homophily”, and it shapes our views, and our lives, in ways we’re barely aware of.
  • Technology, Zuckerman argues, risks making things worse: on the internet, most obviously, it’s possible to exist almost entirely within a feedback loop shaped by your own preferences
  • We long to have our opinions confirmed, not challenged, and thus, as the Harvard media researcher Ethan Zuckerman puts it, “Homophily causes ignorance.” (It also makes us more extreme, studies show: a group of conservatives, given the chance to discuss politics among themselves, will grow more conservative.)
  • The unspoken assumption here is that you know what you like – that satisfying your existing preferences, and maybe expanding them a little around the edges, is the path to fulfilment. But if happiness research has taught us anything, it’s that we’re terrible at predicting what will bring us pleasure. Might we end up happier by exposing ourselves more often to serendipity, or even, specifically, to the people and things we don’t think we’d like?

Someone is already at work: Ethan Zuckerman’s work toward a Serendipity Engine

Sybils in RecSys

Friday, February 6th, 2009

SybilGuard’s authors will present a paper on how to defend recommender systems from the Sybil Attack.

assignmenthelp

DSybil: Optimal Sybil-Resistance for Recommendation Systems

I’m waiting to read the paper to see which real data they’ve used and how it would possibly work on typical social networks of recsys websites, which aren’t that big and may well not be  fast mixing (controversial SybilGuard’s assumptions)

Question/Answers @ RecSys Doctoral Symposium 2008

Friday, November 28th, 2008

I came across an interesting blog post by @HDrachsler, who I started following on twitter after this year’s RecSys conference. The post contains a recording of the question/answer time at the RecSys doctoral symposium (which I unfortunately did not attend). The clearest voice in the recording is Prof. Joseph Konstan, who (obviously, I know) has some very interesting things to say about collaborative filtering, recommender system research, and the state of the field. Here are some notes that I jotted down while I was listening: (more…)

A Pitch on Future Recommender Systems

Thursday, November 27th, 2008

Yesterday I attended a workshop that was aimed at fostering research collaboration between our department and BSkyB. After a short introduction by the head of the department, a number of members of staff gave short (10 minute) pitches about their past and current research, and areas they are interested in for potential collaboration. The range of work being done in the department is huge- perhaps this deserves a post of its own.

(more…)

Papers at RecSys 2008

Monday, October 27th, 2008

A lot of interesting papers have been discussed during RecSys conference. Here are a few list of interesting ones.

(more…)

Tutorials at RecSys 2008

Friday, October 24th, 2008

Yesterday was the first day of RecSys 2008, and was dedicated to three very interesting tutorials:

1. Robust Recommender Systems. Robin Burke introduced the wide range of attacks that typical collaborative filtering algorithms are vulnerable to; scenarios that arise when people attempt to force, rather than express, opinions. An attack was strictly defined as a set of profiles intending to obtain excessive influence on others, which can be aimed at pushing (making recommendation more likely) or nuking (i.e. recommendation less likely) items. His talk was an interesting blend of attack strategies, knowledge that attackers need to have, and a high-level description of approaches aiming at preventing or fixing the system when attacked. Of course, there are strong overlaps between this work and work in other areas (p2p trust, adversarial information retrieval, search engine spam..); I particularly like this area as pushes the point that recommender systems are about people/dynamic datasets, and not just prediction.

2. Recent Progress in Collaborative Filtering. Yehuda Koren (who has recently moved from AT&T to Yahoo! Research) gave a tutorial about the leading approaches in the Netflix prize competition. The techniques he described blend matrix factorisation and neighbourhood models, and include a number of other factors (such as user biases and time) that result in techniques that have multiple-billions of parameters (and the resulting ranking of team BellKor in the competition). His work is remarkable and worthy of the progress prizes he has been awarded thus far. He also explored alternative techniques of evaluating recommender systems, explaining his take on evaluating top-N recommendation lists.

3. Context-Aware Recommendations. Gedas Adomavicius and Alex Tuzhilin introduced their work on incorporating context into recommender systems, including pre-, post-, and hybrid-filtering of recommendation algorithm results based on user context. A running example that was repeated throughout the tutorial was going to the theatre with your girlfriend on the weekend: if you always watch comedy, then your recommendations can be filtered to match what you did in previous instances of the same context (i.e. you can be recommended comedy). They have done a lot of cool stuff on multi-dimensional recommenders, extending the common rating scales into cubes of ratings, and stressed more than once that this is virgin territory. Their work is also impressive, but raised a few questions. For example, should context be described by a well-enumerated taxonomy? Moreover, if you always watch comedy at the theatre with your girlfriend on weekends, then why should you need a recommender system in the first place (especially a collaborative one- what happened to serendipity or diversity)? They have a number of papers that are worth reading before trying to answer these questions!

Social network collaborative filtering

Monday, October 13th, 2008

Interestingly, “This paper demonstrates that “social network collaborative filtering” (SNCF), wherein user-selected like-minded alters are used to make predictions, can rival traditional user-to-user collaborative filtering (CF) in predictive accuracy. “

Machine Learning Applications to Music

Wednesday, October 1st, 2008

Louis gave a very interesting talk about his research on applying machine learning to music. Interestingly, among other things, he discussed two issues of music retrievial:

1) Does the use tags improve retrieval algorithms? In my opinion, the answer is a qualified yes. Case in point: this month at Recsys, Licia and Valentina will present an effective  way of retrieving items (e.g., music files) . Their technique exploits two types of similarity  (item similarity and user similarity), both of which are computed only from (user-specified) tags.  Check section 3 of this paper (short description) and their RecSys paper (complete description).

2) Exsting retrievial algorithms learn your music taste and assume  that it does not change over time. What if you change your taste? That’s a question Neal will answer at, again, RecSys this month. Check his paper.

“Making Mobile Raters Stick to their Word ” @ Ubicomp

Monday, September 22nd, 2008

In few hours I will present MobiRate. Fortunately, the slides are ready ! See them next. A short description follows.

P.S. I’ll blog about Ubicomp shortly. For now, look at the great coverage by Albrecht Schmidt  ;-)

MobiRate

View SlideShare presentation or Upload your own. (tags: trust systems)

Q&A Session (at the conference):

Q> You have shown that MobiRate effectively protects against *indepedent* malicious individuals. What if  malicious individuals collude?
A> Colluding malicious peole will not be able to tweak  ratings because they cannot produce fake crypto material. However, if malicious people collude, one may well run into updating problems. Phones update their ratings  while they move and, consequently, there are   time windows in which ratings are not up-to-date. During those time-windows, colluding people may succeed in attacking the communities they are in  (e.g., in flooding the system with spam content).

Q>  Phones that run MobiRate audit each other. Are their users aware of that?
A> We have assumed that, in downloading and running MobiRate, people silently agree with  the possibility of their phones being “auditors”. However, people should be able to step back and refuse to be auditors at times; for example, whenever they are running out of battery. This feature should be definetely
included in the next version of MobiRate.

Q> Your solution is general, in that, it is able to collect and store not only user ratings but also user activities!
True.  Instead of monitoring ratings, one could force people in keeping a record of their activities. Before deploying MobiRate, we should carefully think about its misuses and try to prevent them. A good starting point could be to understand how “historical misuses of technology can be studied to be avoided in the future” (link)

Short Description of MobiRate:

(more…)

Dataset and R code for our paper on genres/artists affinity

Tuesday, August 19th, 2008

Justin Donaldson and I have a paper at ISMIR entitled “Uncovering affinity of artists to multiple genres from social behaviour data”. The paper details a project we worked on for the past year or so involving popular music listening activity from a pool of MusicStrands users.

We provide not only the paper, but also the dataset and the code used in our analysis. All of this is available at the website we have set up for the project: http://labs.strands.com/music/affinity/

The main contribution of the project is an analysis and illustration of genres as “fuzzy sets” rather than boolean labels. Through a co-occurence analysis of hundreds of thousands of user playlists, a frequency based “affinity” metric is formed between artists and genres. This affinity metric is a more detailed expression of the style of a given artist’s music. The idea and awareness of predominant genres are a trivial part of any person’s understanding of the vast corpus of popular music. However, genres typically are used as Boolean categorical labels. I.e. an artist is understood to be associated with only one given genre.

By expressing a connection to multiple genres through our affinity metric, a more detailed picture of the artist emerges. We give a lot more examples in the website, so be sure to check it out. - http://labs.strands.com/music/affinity/

 

Claudio Baccigalupo

Social Systems

Monday, June 30th, 2008

This month’s Data Engineering Bulletin is about Recommendation and Search in Social Systems. It sports thoughts on robustness and user experience.

Context Matters

Friday, June 27th, 2008

The view we have of recommender systems is that of two-dimensional systems (users x items) whose main goal is to `recommend items to users’. However, as well illustrated in this paper, “decision making is contingent upon the context of decision making; the same consumer may [...] prefer different products or brands under different contexts”. For example, I (the user) may want to be recommended different restaurants (the item), depending on when I am going (the context), with whom I am going (the context, again), and for what purpose (the context, yet again).
(more…)

Summary of IFIPTM'08 conference sessions

Monday, June 23rd, 2008

The IFIP trust management conference, this year joined for the second time with the Privacy, Security and Trust (PST) conference, was held in June 18th – 20th in Trondheim, Norway. The conference has also been previously known as iTrust. Next year, PST and IFIPTM split again, as PST returns to its roots as a local event in Canada; next year, IFIPTM is organized in the US, and in Japan after that. We’ve summarized IFIPTM workshops on Monday and Tuesday in earlier posts, and now give a quick run-through of what this year’s conference program held.

(more…)

Get Ready to Rummble!

Friday, June 20th, 2008

The very last session of the IFIPTM 2008 conference was a demo session; there were 3 demos run and the one that I liked the most was Rummble.com. Rummble is a web site that, much like other web2.0 ideas, has as foundations a social network: the interesting addition, though (and what makes it so appropriate for a conference on trust) is that when you add a friend you can say how much you trust their opinions. You then go on to “rummble” different locations (shops/restaurants/clubs), by rating, tagging, and describing them with a comment. The neat thing is that combining trust, rating similarity, and social distance, the site can then predict how much you will like other places that you have not rummbled, and colours them accordingly. The site is also fully mobile! (more…)

ACM SAC 2009

Thursday, June 5th, 2008

CALL FOR PAPERS – SAC 2009

The 24th ACM Symposium on Applied Computing
at the Hilton Hawaiian Village Beach Resort & Spa
Waikiki Beach, Honolulu, Hawaii, USA
http://www.acm.org/conferences/sac/sac2009/

IMPORTANT DUE DATES
Aug. 16, 2008: Full paper submission
Oct. 11, 2008: Author notification
Oct. 25, 2008: Camera-ready copy
Mar. 8-12, 2009: ACM SAC in Hawaii, USA

(more…)