Categories (tags):

Recommender Systems

recommender_systems

Users reviews are THE market

Users reviews of products (like "I bought an Ipod and it was not working" or "I went yesterday to XYZ Restaurant and it was fabolous" or "i saw 'paradise now' and it was great") are the basic building blocks of Recommender Systems. And of course they are able to determine the success or failure of a product. Many people nowadays before buying a product check "what Internet is saying about this product?", usually the level of information awareness is precisely this one.
So, it should not be surprising that:
- There are authors on Amazon who write reviews of their own books under pseudonyms
at least one U.S. author was mistakenly outed on Amazon.com's Canadian website as having written a review of his own work. The real names of thousands of people who had posted anonymous customer reviews under pseudonyms like "a reader from St. Louis" were revealed online for several days - a mistake that finally was corrected after reviewers, some of them authors themselves, complained.
- a restaurant is suing zSurvey.com, a company that collects restaurant reviews from common consumers and posts them online and in a book, for damaging its reputation. (...) seeking a public apology and 50,000 yuan (US$6,173) each in compensation.They are also demanding the Website delete all of the negative comments it has posted online and stop publishing a guide book with negative comments".
- and mainly that Amazon Gets Patents on Consumer Reviews
Review your local dry cleaner, pay $10 million?
User reviews are a hot new content area, being used by Google (Quote, Chart), Yahoo (Quote, Chart) and MSN to sweeten their local search results. But as of Thursday, such consumer reviews could put search providers, as well as thousands of e-commerce sites, video rental or review sites and online booksellers, in the sights of Amazon.com's (Quote, Chart) lawyers.
The patents are simply absurd (you can read them in the article) and I'm not going to comment them and I'm very happy that at least for now Europe voted against Software Patents).
About reviews, I think that creators should be free to publish their opinions (in term of reviews in this case), they should own their reviews (hreview seems a great format for this task), reviews should be released under very liberal licences and everyone should be allowed to aggregate the reviews and do whatever she prefers with this information: offer a Recommender System service, use them for her own decisions, .... Reviews are one of the cornerstones of the Information society and they should be usable by anyone who has an idea.

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Semantic web	semantic_web
Social Software	social_software
Trust and Reputation	trust_and_reputation

Presentation in standard format, S5

Some days ago I had to give a presentation for the 2K* symposium, a joint initiative of research groups from different IT institutions, based in Trento and in Genova. The 40 mins presentation was titled "Trust in Recommender Systems: an historical overview and recent developments" (check the source code!). It is heavily based on an old presentation, I just added some slides about microformats, a concept I wanted to convey to the audience.
Anyway, I took the occasion to try to create the presentation in HTML using S5: A Simple Standards-Based Slide Show System developed by Eric Meyer. I think I will create all my future presentation in S5 from now on. The advantages: it "forces" you to keep the slides simple (no unnatural flow of information) and short (however you can have animations, check this slide); it is easy to publish the presentation on the Web, anyone can link to a specific slide, search engines find the information and index them, it is highly standard, evolutionary and small-pieces-loosely-connected-philosophy-like (for exaple it would be possible to create a small piece of javascript code that collect slides from different presentations in some meaningful automatic way to create a new presentation, but the possibilities are endless of course, especially if using the S5 format based on XOXO microformat), I can create the presentation with whatever text editor (perfect if you are in text mode), it does not require the viewer to have some fancy program (openoffice for the freedom lovers, powerpoint for the others) but a browser suffices.
You can find many presentations in S5 format in the microformats wiki; I also liked this presentation of Firefox, with style vulpes-flagrans or with style greenery. Yes, I know the stile I used for my presentation is not that great, if someone with graphical skills would like to create a style for me, it will be very appreciated of course.
For starting playing with S5, I suggest you S5 primer (you need to download HTML code and edit it) or S5present, an open-source web-based slideshow application (you just create an identity there and then use the site for creating the presentation). Guess what? S5 Presents was written in under 10 hours and 500 lines of code using the fantastic Ruby on Rails framework.
[question about English: "take the occasion to"? "take the chance to"? I wanted to say that I used this fact as an opportunity to try the technique. How do you say it in English?]
Tag: s5

Categories (tags):

Folksonomy	folksonomy
Free software	free_software
PhD	phd
Recommender Systems	recommender_systems
Semantic web	semantic_web
Social Software	social_software
Trust and Reputation	trust_and_reputation

AAAI05: terrific talk by Marty Tenenbaum

AI Meets Web 2.0: Building The Web of Tomorrow Today by Dr. Jay M. Tenenbaum.
Terrific terrific talk, fascinating. I should have podcasted it because you really missed something (except I have nothing to record audio on, would you consider sending me your old mp3 recorder pen?). I was so excited during the talk that I happened to take a photo of almost any slide. Actually the slides were 94 and I photoed 59 of them! Incredible to me as well.
Anyway, you might want to read the slides (pdf) or maybe you want to have a look at my pictures (possibly as a slideshow).
He introduced all the stuff I enjoy, such as Blogs, RSS, wiki (wikipedia), folksonomies, tags, flickr, Del.icio.us, microformats (aka Lower case semantic web), technorati, pubsub, greasemonkey (bookburro, greasemap) and much more; all tied together in a fascinating, convincing, making-sense manner!
After his presentation, we spoke about my research and he seemed interested. He invited me to visit commerce.net for one month or so and I have to say that I really like the idea. I spoke also with Rohit Khare that is actually working with Tenenbaum and he has a whole bunch of very clever, fascinating, realizable ideas that would really make an impact. They also underline more than once that this kind of architecture/language-of-web2.0 projects should be open source and I totally agree with them and like it.
Actually after the presentation, while I was speaking with Marty and Rohit, there was also Jesse Andrews, the creator of the mind-blowing book burro (actually he got most of the attention, totally deserved by the way). I guess it should be too cool having someone presenting your hack on a conference and then go to meet that person and say "You know the Book Burro extension you presented? Well, I'm the creator of it!". Cool! If you want to see how Jesse looks like, here is a picture of him and wait some more great hacks from him in few days.

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Presentation on "Trust in Recommender Systems: an historical overview and recent developments"

I was invited by Stefano Mizzaro to give a lecture in his course in "Web Information Retrieval". I spoke about "Trust in Recommender Systems: an historical overview and recent developments". It was a lot of fun (at least for me). And I thought I could share the slides with you. They are in OpenOffice .sxi format (it is an open format, so if you program does not read a commonly used open format, you probably better change it). They are released under a Attribution-ShareAlike Creative Commons licence. This means that if you want to use them you just have to give credit to me and re-share your slides under the same licence. If you don't want to re-share your derivative work under the same Creative Commons licence, you are still free, free of not using them. Enjoy.

Categories (tags):

Free software	free_software
Metadata	metadata
Recommender Systems	recommender_systems
Semantic web	semantic_web
Social Software	social_software
Trust and Reputation	trust_and_reputation

My suggestion to Google: getoutfoxed('s author)

Google, do hire Stan before Yahoo! does it. Stan is the author of "Outfoxed - Personalize your internet." I didn't play with the code yet (seems a Linux version is not yet ready at the moment, but on the way). Yes, the code is open source (Mozilla Public Licence), sweet! Anyway, the detailed description is fantastic! It is a bit like what I want to do for my PhD thesis. The difference? Stan did it! Check the site: it has a lot of interesting pages such as The Outfoxed Idea (A collection of thoughts on the theoretical aspects of Outfoxed, and the whole idea of using social networks for metadata distribution). Or at least the page A Third Phase of Internet Search in which Stan pictiorally shows the 3 phases: Naive trust --> PageRank and inferred quality --> Social networks to determine subjective quality

Every search query is a question: "What pages are most related to X?" Current search engines assume there is a single correct answer to each query. But consider a query like "Britney Spears." (The most popular Google query for 2004.) If you're a fan, you probably want to see her official site and maybe lyric pages. If you're a musician, you probably want to see reviews and music tabs. Of course, current search engines can't do this because they only consider "objective" measures like the number of links to a page. (See The good, the bad, and the subjective) What is needed is subjective, trusted ratings of the pages.

Categories (tags):

Books	books
Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Read the books people you dislike dislike

I know the title is hard to parse. Let use some parenthesis: Read [the books [people [you dislike] dislike]]. That is, there are people you dislike, they dislike some books, you possibly will like these books. Pietro Speroni reports that A right winged newspaper: Human Events online, asked a panel of 15 conservative scholars and public policy leaders to help us compile a list of the Ten Most Harmful Books of the 19th and 20th centuries. (here the list) and how "The list have it all, it’s the most complete list of texts I found that were really important to understand the world we are living in". The rationale behind is: if neocons believe these books are harmful and since I think neocons are harmful, I should read these books. While this is ok on real world, this reasoning does not work in Trust-aware Recommender Systems, topic in which I'm phding. In online communities (in which it is easy to create fake identities) this is subject to a simple attack and anyone could easily game the system. The idea: since I get recommended the items disliked by people I dislike, the user I dislike could pretend to "dislike" the item she wants I get recommended. Ex: a neocon identity could pretend to dislike the book "why bush is right" (hopefully this does not exist and it is just an example) and I get recommended it. For this reason, in algorithms I designed, I decided that the opinions of people you dislike should not influence your recommendations at all, they are simply discarded because otherwise they are able to influence your recommendations and hence game the system. Well, not sure, I'm good in explaining it (English is hard...). Maybe you want to check some papers of mine in which hopefully I was helped in writing in a clearer way. Since we are speaking of books, maybe you want to check the list of books I've read (actually it is not at all complete or updated, I was trying to keep it with allconsuming.net and to decentralized publish it also in semantic web formats (RSS | XML) but in fact I created it once and never updated ... maybe in a short future there will be a tool that will allow me to keep a list of read books, with comments and to automatically publish it on my blog, in that case I'll probably try again to keep it updated. Or such a tool is already there? If so, please let me know).
The list of books that neocons think are harmful is

# The Communist Manifesto by Karl Marx and Freidrich Engels
# Mein Kampf by Adolf Hitler
# Quotations from Chairman Mao by Mao Zedong
# The Kinsey Report by Alfred Kinsey
# Democracy and Education by John Dewey
# Das Kapital by Karl Marx
# The Feminine Mystique by Betty Friedan
# The Course of Positive Philosophy by Auguste Comte
# Beyond Good and Evil by Freidrich Nietzsche
# General Theory of Employment, Interest and Money by John Maynard Keynes
# The Population Bomb by Paul Ehrlich
# What Is To Be Done by V.I. Lenin
# Authoritarian Personality by Theodor Adorno
# On Liberty by John Stuart Mill
# Beyond Freedom and Dignity by B.F. Skinner
# Reflections on Violence by Georges Sorel
# The Promise of American Life by Herbert Croly
# Origin of the Species by Charles Darwin
# Madness and Civilization by Michel Foucault
# Soviet Communism: A New Civilization by Sidney and Beatrice Webb
# Coming of Age in Samoa by Margaret Mead
# Unsafe at Any Speed by Ralph Nader
# Second Sex by Simone de Beauvoir
# Prison Notebooks by Antonio Gramsci
# Silent Spring by Rachel Carson
# Wretched of the Earth by Frantz Fanon
# Introduction to Psychoanalysis by Sigmund Freud
# The Greening of America by Charles Reich
# The Limits to Growth by Club of Rome
# Descent of Man by Charles Darwin
(I copied the list from Pietro Speroni's reading list)

Categories (tags):

Blogging	blogging
Metadata	metadata
PhD	phd
Recommender Systems	recommender_systems
Reviews	reviews
Semantic web	semantic_web
Social Software	social_software

hReview: a semantic microformat for reviews

Some weeks ago, Tantek was introducing a new microformat hReview.
We are pleased to announce the first public draft (v0.1) of hReview, jointly co-authored by representatives from America Online, CommerceNet Labs, Microsoft, Six Apart, Technorati, and Yahoo!. hReview is an open microformat standard for publishing and indexing distributed reviews on the Web. This standard enables users to contribute, identify, and aggregate review content on their own web sites and blogs as well as on community sites.
I didn't have time yet to dig into it but it is good that they analyzed previous attempts (I was trying to use RVW by Alf Eaton and to keep my list on Allconsuming but I didn't put too much effort into this) and that they ask for Feedback; almost all the links are to Wikipages so you can edit them directly there.
In general I really appreciate the work of Technorati (I also wrote a paper backing their proposal of VoteLinks, submitted to Web Intelligence 2005: "Page-reRank: using trusted links to re-rank authority" (pdf)).
Some other link I'll try to digest later on: jluster on hreview, hreview on technorati, hreview on del.icio.us, organizedshopping on hreview, adriancuthbert suggested to use this_is_an_hreview as common tag (tagspace?).
It would be great to have this format widely adopted so that the amount of decentralized published reviews will become soon huge and I will have a large amount a data for what I'm working on in my PhD: Trust-aware decentralized Recommender Systems. If interested, check my (a bit outdated) PhD proposal at my papers page.

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Trust Metrics Book

I'm thinking about writing a book on Trust Metrics, or maybe about Trust Metrics and Recommender Systems. (I need to write my PhD thesis anyway so if I can get it published, this is a plus). Well, a search inside-books on Amazon for "trust metric" reveals this is not a too covered topic. Good. Do you have any suggestion? Publisher, topics, whatever. Anyway being able to search inside (almost) every book in one second is astonishing, sometimes I forget about how astonishing the Web is...

Categories (tags):

Recommender Systems

recommender_systems

The economist on Collaborative filtering

Article over at The Economist United we find on Collaborative Filtering. It is interesting to note that it speculates also on attacks to Recommender Systems. An interesting (simple as it should be) idea is the following:
Nolan Miller, of Harvard University's Kennedy School of Government, and his colleagues (...) probabilistic techniques to determine whether a score is likely to be “honest”, by spotting unusual-looking patterns in scoring. Dozens of accounts created on the same day, all of which give high scores both to a bestseller and a new book, for example, might be an orchestrated attempt by a publisher to get fans of the former to buy the latter.

Fiddling the filters
A second concern about collaborative filtering is that as it grows in importance, people may increasingly try to manipulate it: publishers, for example, might start recommending their own books. Last November, Michael O'Mahony of University College, Dublin, published a paper demonstrating that even today's most advanced collaborative filtering systems are not all that robust when subjected to malicious users seeking to subvert their ranking systems. None of the existing systems is explicitly designed to combat malicious use. Can such “recommendation spam” be prevented?

But even if the problems of privacy and dishonesty can be overcome, there may be a limit to how accurate the recommendations made by collaborative-filtering systems can be. This arises from the fact that people's opinions change. You may enjoy a new album at first, and give it a good score, but change your mind after a few weeks once the novelty has worn off. But your old score still stands.

Categories (tags):

Copyright	copyright
PhD	phd
Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Controversial books: patenting the obvious?

Interesting NYTimes's article (if you don't want to register, use BugMeNot where you can find shared login and password pairs). Mikhail Gronas discovers that "reviewers gave more five-star reviews than two-star reviews, creating an upward sloping curve". (...) "But the most telling variable is the one star rating. Professor Gronas found that books high on what he called the "controversiality index" are given almost as many one-star as five-star ratings, creating a horseshoe-shaped curve. As it turns out, these books also tend to have high sales."
I've found these patterns analyzing Epinions.com ratings and trust statements (chech the graphs' on the paper (pdf)) but actually I don't think they are that surprising: they seem pretty obvious and I just reported them passing by.
What is really depressing is that Dartmouth is now in the process of patenting software that will be used to determine the "controversiality index".
I'm happy that in Europe we are still fighting against a so-stupid-policy of being able to patent everything, no matter how trivial it is. In this case the controversiality level of a book is something like "if a book received as much 5 ratings as 1 and if the 5 and 1 ratings together are the vast majority of ratings and if the number of received ratings is over a threshold (probably depending on release time), then the book is controversial" (putting it in formula that produces a controversiality value would require 10 minutes at most).
By the way, I'm currently working on the concept of controversiality of users and hopefully a paper is on the way. Controversial users are users who are trusted by many and distrusted by many. (Bush is a good example, but this can happen to highly visible persons in general). The idea is that Local Trust Metrics make sense expecially for highly controversial users (for example, users who are trusted by more than 200 users and DIStrusted by more than 200 users in the community). For those users, it does not make sense to predict a trust value of 0.5 saying that you should trust this user as 0.5 but, instead, to predict you should trust this controversial user as 1 if, for example, all your friends trust her and 0 if all your friends distrust her.

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Social Software	social_software
Trust and Reputation	trust_and_reputation

Using social software for good: car pooling

Paul Resnick is researching on "ride sharing services that dynamically match riders with rides". Read the very interesting and clear SocioTechnical Support for Ride Sharing scenario document. The idea is to make car pooling easier using ICT. If your interests contain trust, recommender systems and making the earth a better place, you should definitely read the paper. Maybe I'll try to put up a project and submit to the local government, there was a car pooling project in Trento but it seems dead. Contact me if you are interested! [My impression is that often research does not produce useful and real benefit for society, this is a case in which we can put our brain activity for creating something useful and that can make a difference].

Excerpt from SocioTechnical Support for Ride Sharing by Paul Resnick, Associate Professor University of Michigan, School of Information

I. Intro/Overview
In America, there is tremendous unused transportation capacity in the form of unoccupied seats in private vehicles. Not only would filling some of those seats reduce smog, congestion, and fuel consumption, but it also could create opportunities for increasing local social capital. The major barriers to ride sharing include coordination of routes and schedules, safety risks, social discomfort with sharing what are currently private spaces, and an imbalance of costs and benefits among the affected parties. Despite these barriers, ride sharing does occur, both in the form of recurring carpools and van pools. According to one estimate, more than twice as many people in America share a ride to work in a private vehicle as use public transportation to get there [ref.] In a few cities, there is even instant ride sharing among strangers. Emerging changes in the technology infrastructure of our society may soon make it possible to reduce some of the barriers that have limited the appeal of instant ride sharing. The first change is the widespread deployment of cell phones and other mobile communication devices, with the prospect that they soon be integrated with a position-sensing infrastructure. The second is advances in computational power that may allow for dynamic route matching of drivers and riders. The third is the development of reputation systems on the Internet for maintaining trust among strangers. Research is need on how to leverage these developments to create a SocioTechnical infrastructure for instant ride sharing.

II. Scenario
Janine is new to instant ride-sharing. She is twenty-five and single. She s trying to save money and besides, it s such a hassle to park at the hospital where she works as a research assistant administering clinical trials. She sometimes stays late at work, so she never joined a carpool, but she s decided to try the new Ann Arbor/Ypsilanti instant carpool system. She was a little worried about taking rides with strange men, so she set her profile to only accept rides from women, or from men who had a history of giving at least 10 previous rides without any complaints from riders. She logs onto the website and enters her address and her destination address. She finds that if she walks only to the corner of her current block, she ll have to wait an average of 15 minutes to get a ride, and sometimes much longer, but if she walks two blocks further, to a main street, she can usually get a ride within 3 minutes. She decides to walk the two blocks. This first morning, she s kind of curious about what kind of person picks up riders, so she checks off the box that indicates she s willing to converse the driver.
She s still a little nervous, so she doesn t allow any of her personal information (name, address, or interests) to be revealed to the driver. She s talked to other people who found people to play music with or got a ride all the way home by revealing some information, but she s decided to wait and see how the whole system works first. As she walks out the door, she calls the number she had pre-programmed into her cell phone. The system tracks her progress as she walks to the main street and tells her that a blue Toyota Matrix is just three blocks away and that she should hold up her instant ride-share sign. It gives her a code that she s supposed to say to the driver, and a code that the driver is supposed to say to her. Sure enough, the car pulls up. The driver is a forty-something woman, smartly dressed with a white lab coat on the passenger seat. They exchange codes and Janine jumps in the back. The driver asks Janine what she does at the hospital and soon they discover that the driver and Janine s boss are good friends from way back, and tells a humorous story about her boss when he was first getting started in medical research. As they pull into a choice parking space at the hospital parking lot, reserved for multiple occupant vehicles, the driver smiles and says, You saved me 5 minutes driving around and around in this lot. Thanks. Maybe I ll take you again some time, but my schedule s very irregular so I m not sure when. Thank you! says Janine as they walk off in different directions. As she walks away, she calls the ride sharing system again from her cell phone and presses a button to indicate that she arrived safely, that she would be happy to ride with that driver again, and that she recommends her to other passengers.

[/end of Excerpt]

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Semantic web	semantic_web
Social Software	social_software
Trust and Reputation	trust_and_reputation

CiteULike: A free online service to organize your academic papers

[I'll write something about my trip in Israel later on, as time permits]
I just found on HubLog an online service I was really waiting for: CiteULike (a prototype service to manage your personal library of academic papers). When you are logged in and visiting a page related to a paper, you can post that paper to your online library using a bookmarklet. In doing so, you can also specify tags, a list of keywords you'd like to associate with this article (a la del.icio.us and flickr) and optional notes. The service is very similar to del.icio.us (simple, tag-powered and social), but precisely tailored for academic papers. You can also see all the papers tagged under a certain tag (for example networks). Cool!

You can see your library (see mine), and see which other users are reading the papers you find interesting. The about page tells you what is coming soon. I think that "exporting those data in semantic web formats" and "opening the API" can be interesting additions to the list. This would be great for creating Trust-aware Recommender System tailored for researchers.
The big problem I see is that only papers in (PubMed, HubMed, JSTOR, arxiv, IngentaConnect) can be added for now. Most of the papers I'm interested in are not stored on those online repositories.
I wish it would be possible to add Citeseer (I'm involved in a project whose goal is to relaunch citeseer), eprints archives and Springler (see my last paper page on Springler for a typical paper page).
I'd like also to be able to keep some blog posts (not published) in my online library and papers that researchers keep in their homepages: using the URL as key for the "paper" could do the work but this will make the site just as del.icio.us is now and I think this is not the goal of the online service. Maybe it would make sense to introduce two levels of papers: certified (by some recognized authority such as PubMed) and uncertified (such as my papers I keep on my blog) but I'm not sure this is a good idea.

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Travelling to Cyprus and Israel

I'm at Coopis 2004 right now (in Agia Napa, Cyprus) and next week I'll move to Jerusalem in order to meet Zvi and other people of the Multiagent Systems Research Group of the Hebrew University of Jerusalem. I'll be back in my office on November 9.
I was hoping to do a lot of work during the coopis conference but the wireless network is not working very well and so expect few or no blogging at all.
I almost forgot to say that I'm presenting "Trust-aware Collaborative Filtering for Recommender Systems" (find it under papers section). Check it out if you are interested in Recommender Systems and Trust.

Categories (tags):

FOAF	foaf
PhD	phd
Recommender Systems	recommender_systems
Social Software	social_software
Trust and Reputation	trust_and_reputation

Paper accepted for Coopis --> looking for cheap place in Cyprus (through 2 degrees of separation)

Good news: my paper "Trust-aware Collaborative Filtering for Recommender Systems" got accepted for Coopis2004.
Bad news: the conference is hyper-expensive.
So I'm looking for hyper-cheap (possibly free) hospitality in Larnaca, Cyprus, from 25 Oct to 29 Oct 2004. I checked on couchsurfing (a site where people offers ospitality in their houses and a super-cool YASN [yes, you can express your friends list]) but I found none in Cyprus.
If we take for true the six degree of separation theorem, I should be connected to everyone in Cyprus by only six degrees of separation. So I guess there should be at least some cypriots in my friends of friends set, now i only need to find one of the connecting friends. So if you know someone in Cyprus, please become my friend and close the circuit (and don't forget to write down the path from me to the cypriot host in the comments below). Thanks.

Categories (tags):

Recommender Systems	recommender_systems
Social Software	social_software
Trust and Reputation	trust_and_reputation

KnoBot

KnoBot [UPDATE: the link is often broken. the knobot page on Sourceforge is always up (thanks Zbigniew)]- An agent for decentralised knowledge exchange :KnoBot combines semantic web technology with a P2P design to build a trust based decentralised system for information selection and discovery.
I should check it better but looks a lot like what I want to do for my PhD.
On KnoBot news I found a similar and interesting project: the Matrix Public Network project.
Both ot the project have running code, so we can try them out.

Categories (tags):

Emergent Democracy	emergent_democracy
Recommender Systems	recommender_systems

Boycott the Daily Me!

From Boycott the Daily Me! by Sunstein:
"For democracy to work, people must be exposed to ideas they would not have chosen in advance. Democracy depends on unanticipated encounters. It is also important for diverse citizens to have common experiences, which provide a kind of social glue and help them to see they are engaged in a common endeavor. A world where people only read news they preselect creates a risk of social fragmentation."
This is my greatest fear about Trust-aware Recommender Systems (or in general systems that personalize user experience): that people will be exposed only to what they already approuve and like.

This already happens to me in real life: most of the people I know are against the Iraq Invasion and I often (and wrongly) think that all the Italians share this opinion. Then, when I meet someone that agrees with this war, I am initially totally shocked. Using personalized services, it is even easier to never "meet" people who are not perfectly like-minded.
As Sunstein puts: "People who think the world economy is in trouble are likely, after discussion, to fear economic catastrophe". This is a risk I often feel after discussions with activist groups I'm involved in.
Sunstein ends with: "Democracy is undermined when people choose to live in echo chambers of their own design". About echo chambers, see Joi
Anyway Josh , the project's Product Manager of Newsbot (the new personalized news-provider by Micro$oft), has some interesting points about how "personalized news - but when done well - (...) leads us even further away from Sunstein's dystopic future".
(Found via Erik)

Categories (tags):

Recommender Systems

recommender_systems

New personalized services

Erik comments on personalized services presented in these last couple weeks:
- a9.com: (by Amazon) a personalized experience, though not quite personalized search).
- Newsbot (also in Italian): (by Micro$oft) personalized news (with or without login).
- Findory: personalized news (with or without login), but not new.
- Google Personalized (weak attempt).
Interesting competition.

Categories (tags):

Recommender Systems

recommender_systems

Evaluating collaborative filtering recommender systems

If you are interested in Recommender Systems, you cannot miss Evaluating collaborative filtering recommender systems by Jonathan L. Herlocker, Joseph A. Konstan, Loren G. Terveen and John T. Riedl.
You need an ACM Web Account and some time (it is 53 pages).

Recommender systems have been evaluated in many, often incomparable, ways. In this article, we review the key decisions in evaluating collaborative filtering recommender systems: the user tasks being evaluated, the types of analysis and datasets being used, the ways in which prediction quality is measured, the evaluation of prediction attributes other than quality, and the user-based evaluation of the system as a whole. In addition to reviewing the evaluation strategies used by prior researchers, we present empirical results from the analysis of various accuracy metrics on one content domain where all the tested metrics collapsed roughly into three equivalence classes. Metrics within each equivalency class were strongly correlated, while metrics from different equivalency classes were uncorrelated.

Categories (tags):

Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Cai Ziegler

During the past week I was in Oxford for the 2nd Trust Management Conference. The presentation (pdf) (sxi) of my paper went well.
Most of the participants were concerned with privacy and the problem of setting up a secure environment for virtual organizations (business basically). I am not too much interested in this topic that is basically agreeing with Microsoft, IBM and HP (that were present with some representatives) about standards for the trust management processes, often reduced to simple access control lists.

Instead I was very happy to meet Cai Ziegler. Cai is working on topics very similar to my interests. But it is doing more (his scope on semantic web recommender systems is broader, since he also takes into account taxonomies), better (its English is simply wonderful) and faster (he is still in his first year of PhD). Can I at least say I'm humble?

Check his publications:
Analyzing Correlation Between Trust and User Similarity in Online Communities,
Spreading Activation Models for Trust Propagation,
Semantic Web Recommender Systems,
Trust Models for Expertise Discovery in Social Networks.
I will try to convince him to start blogging. It would be a great injection of inspiring thoughts for all the blogosphere!

Categories (tags):

Blogging	blogging
FOAF	foaf
Peer to peer	peer_to_peer
PhD	phd
Recommender Systems	recommender_systems
Semantic web	semantic_web
Social Software	social_software
Trust and Reputation	trust_and_reputation

PhD Research Proposal: Trust-aware Decentralized Recommender Systems

I realised today I didn't write yet an entry about my PhD Research Proposal "Trust-aware Decentralized Recommender Systems" (TaDRS).
So here it is the PDF file. If you have any comment or criticism, I'll be happy to hear from you.
The PhD research proposal is a little bit outdated (29th May 2003) but I didn't have a blog at that time. Enjoy and let me know what you think.

UPDATE:
Abstract
This PhD thesis addresses the following problem: exploiting of trust information in order to enhance the accuracy and the user acceptance of current Recommender Systems (RS). RSs suggest to users items they will probably like. Up to now, current RSs mainly generate recommendations based on users' opinions on items. Nowadays, with the growth of online communities, e-marketplaces, weblogs and peer-to-peer networks, a new kind of information is available: rating expressed by an user on another user (trust). We analyze current RS weaknesses and show how use of trust can overcome them. We proposed a solution about exploiting of trust into RSs and underline what experiments we will run in order to test our solution.

Categories (tags):

Recommender Systems	recommender_systems
Social Software	social_software
Trust and Reputation	trust_and_reputation

Reviewr

Reviewr "ties into the API exposed by Ludicorp's [...] new social software application, Flickr and hooks it up to the API exposed by Amazon. The point is that using Reviewr allows you to search for reviews of products by people you know and trust." (via Hublog)
Interestingly, as I was proposing in a previous post, Friendr limits the number of contacts an user can have. It was not a totally dumb idea after all...
Check the services already created using the API and the services documentation (1, 2)

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Social Software	social_software
Trust and Reputation	trust_and_reputation

Other papers analyzing Epinions.com web of trust

Since Seb ha cited my paper as epinions empirical analysis paper, I'd like to mention other 2 papers that analyze epinions web of trust:

the already commented Trust Management for the Semantic Web and the new Propagation of Trust and Distrust.
As a side point, note that we collected datasets of different dimensions. I collected only 49.290 Epinions users because I was following only "this user trusts X" links. Richardons et al. collected 75.000 users (but used only 5.000 of them); I think they followed also "this user is trusted by" links. Guha et al. had access to the real dataset of Epinions which consists of 130.000 users. Note also that Guha et al. had access also to the web of distrust (a sort of black list) while this information is not available on Epinions.com and hence not downloadable.

This post also appears on channel social software

Categories (tags):

Recommender Systems

recommender_systems

I like it

On hublog I have discovered I like it, collaborative filtering for web pages with a 1-click rate bookmarklet.

Categories (tags):

Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Trust related conferences wiki

I collected some conferences related to Trust and Reputation, Social Networks and Recommender systems.
The wiki page is at http://moloko.itc.it/trustmetricswiki/moin.cgi/TrustRelatedConferences and of course you are invited to edit it adding new conferences or moving old conferences.

Categories (tags):

Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

Paper accepted! Oxford I'm coming...

The paper I wrote got accepted! WOW!
It seems I'll be in St. Anne's College, Oxford City from 29 March to 1 April 2004 for the Second International Conference on Trust Management.
Next time I must submit to a conference to be held in Hawaii or Virgin Islands. I deserve some sea and beach.

Categories (tags):

PhD	phd
Recommender Systems	recommender_systems
Social Software	social_software
Trust and Reputation	trust_and_reputation

Paper submitted to iTrust2004

I submitted my paper Using Trust in Recommender Systems: an Experimental Analysis to the Second International Conference on Trust Management 2004.
You can read the PDF file or the HTML version (by latex2html).

Abstract:
Recommender systems (RS) have been used for suggesting items (movies, books, songs, etc.) that users might like. RSs compute a user similarity between users and use this as a weight for the users' ratings. However they have many weaknesses, such as sparseness, cold start and vulnerability to attacks. We assert that these weaknesses can be alleviated using a Trust-aware system that takes into account the ``web of trust'' provided by every user.
Specifically, we analyze data from the popular Internet web site epinions.com. The dataset consists of 49290 users who expressed reviews (with rating) on items and explicitly specified their web of trust, i.e. users whose reviews they have consistently found to be valuable.
We show that users have usually few items rated in commons. For this reason, the classic RS technique is often ineffective and is not able to compute a user similarity weight for many of the users. Instead exploiting the webs of trust, it is possible to propagate trust and infer an additional weight for other users. We show how this quantity can be computed against a larger number of users.

Categories (tags):

Blogging	blogging
CoCoA	cocoa
Movable Type	movable_type
Recommender Systems	recommender_systems
Semantic web	semantic_web
Trust and Reputation	trust_and_reputation

blam! rocks

I've just used blam! in this review of Revolution OS.
Basically blam! add some semantic information to your blog entry when this is a review. The semantic information can be understood by a computer program so that it will be possible to, for example, aggregating all the reviews about a certain book or movie.
Read about OpenReviews and their possible uses from Accordion Guy.
I'm planning to do something similar for my project CoCoA.

The semantic format for reviews is RVW (Review Module for RSS 2.0), created by Alf Eaton. Read an explanation of RVW from Corante.

The RVW specification is a module extension to the RSS 2.0 syndication format. RVW is intended to allow machine-readable reviews to be integrated into an RSS feed, thus allowing reviews to be automatically compiled from distributed sources. In other words, you can write book, restaurant, movie, product, etc. reviews inside your own website, while allowing them to be used by Amazon or other review aggregators.

There should be more than enough RVW metadata out there floating around at this point. The next step is for someone to build a decent aggregator that collates reviews of a particular topic or two. Because of RVS, creating aggregate rating scores and summarizing opinions should be very straightforward. It's really not in the best interests of Amazon, epinions and the like to lose control of their review content, but RVW makes controlling review content impossible in the long term. Anyone got some pull at the Google skunkworks?

Blogware supports the new format and there is also a RVW plugin for Movable Type but I don't understand how it works.

Seb likes RVW and also point out how this semantic information could be used to generate personalized recommendations.

In the case of item types that describe reviews, overall average ratings on any particular product are easy to look up. However, if you choose to provide a description of your personal web of trust to those interfaces (think of blogrolls as a proto-example), you can efficiently get a sense of what your tribe of like-minded individuals thinks of that product. It's the microblogosphere idea again - look up Recommender systems and the microblogosphere for more.

This is essentially what my PhD Research Proposal: Trust-aware Decentralized Recommender Systems is about.

Categories (tags):

Maryland	maryland
Peer to peer	peer_to_peer
PhD	phd
Recommender Systems	recommender_systems
Trust and Reputation	trust_and_reputation

University of Maryland

As part of my PhD program, I'll spend the next 3 months at the University of Maryland. I'll stay here until January 17, 2003.

I also opened up a photo gallery in which I'll post photos taken here.