Profession and academia

Wall Street Journal » Blogs » Digits
By Jennifer Valentino-DeVries

Can you trust Twitter in a disaster?

Researchers at Yahoo analyzed tweets after the Chilean earthquake earlier this year and found evidence that the Twitter community works like a “collaborative filter,” questioning reports that turn out to be fake and confirming those that are true.

Lately the microblogging service has been a source of news after catastrophes such as earthquakes. It’s quick and easy and can be a direct conduit from eyewitnesses to the outside world. But like anything else coming from unsubstantiated sources, news from Twitter faces big questions of credibility.

The Yahoo researchers didn’t find anything to suggest that information should be considered reliable because it’s tweeted or re-tweeted. But what they found was that when false rumors entered Twitter, about half of the tweets related to the information denied it.

When the researchers studied tweets about rumors that were later confirmed to be true, they found that less than 1% of tweets about that information denied it. Some tweets questioned the true information, but eventually it became supported, and almost all tweets affirmed it.

In one example, news of a tsunami after the Chilean earthquake on Feb. 27 spread quickly through Twitter “while government authorities ignored its existence,” the authors — Marcelo Mendoza, Barbara Poblete and Carlos Castillo — wrote. That rumor turned out to be true.

News of baseless rumor about a tsunami warning elsewhere in Chile also spread quickly through Twitter, but the vast majority of tweets mentioning it denied or questioned it.

The New Scientist magazine featured our upcoming SIGIR paper; it will be presented in Geneva next week:

Demographic data can help, say Ingmar Weber and Carlos Castillo at Yahoo Research Barcelona, Spain. For example, they say that when US women type in the search term "wagner", they are most likely to be thinking of the 19th-century German composer. US men, on the other hand, may well be thinking about the makers of spray painters.

By giving a search engine some basic demographic information, such as age, gender and educational background, it is possible to boost the engine's chances of identifying user intent correctly, say Weber and Castillo. That personal information can be gleaned when people sign up to the other services, such as email, that search engines provide.

To check their theory, the researchers analysed data collected from Yahoo account holders through its search engine over a 12-month period. They then identified ambiguous search terms by looking at searches for which the top few results concerned wildly divergent concepts, and recorded which result the user chose to click on. By running those ambiguous search terms through their demographically modified search engine, they managed to get the chosen link to appear as the top-ranked result 7 per cent more often than in the standard Yahoo search.

Ingmar Weber and Carlos Castillo: The Demographics of Web Search. SIGIR 2010.

Update 2010-07-13: the paper was also posted in Slashdot.

Update 2010-07-14: the paper was also recommended by The Economist (free exchange blog).

The Symposium on Discrete Algorithms (SODA) is one of the top venues in computer science. In this year's edition SODA'10, there is a paper by Flavio Chierichetti, Silvio Lattanzi and Alessandro Panconesi: "Rumour Spreading and Graph Conductance".

In the paper, there is a very interesting footnote in the front page:

For the role of the Italian Ministry of University and Research please see the Unacknowledgements.

This is what you find at the end of the paper (emphasis added):

Unacknowledgements: This work is ostensibly supported by the the Italian Ministry of University and Research under the FIRB program, project RBIN047MH9-000. The Ministry however has not paid its dues and it is not known whether it will ever do.

I hope the Ministry will pay now!

Correlator is a new way of searching information. It is currently built on top of the English Wikipedia, but can use any source of information as its data collection.

Given a query, it first assembles an "overview" page synthesized from sentences in Wikipedia articles. It also analyzes matching phrases to find names of persons, locations, and dates related to the query. Then, these results are presented in a way that is dependent on the type of entity: persons are shown in a sociograph, locations on a map, and dates on a timeline.

Correlator was launched a few days ago. TRY IT!: Dinosaurs in Argentina, transistor or Picasso and Peace.

The Adversarial Information Retrieval on the Web workshop is now in its fifth edition! The deadline for research articles is February 13, 2009.

Adversarial Information Retrieval addresses tasks such as gathering, indexing, filtering, retrieving and ranking information from collections wherein a subset has been manipulated maliciously. On the Web, the predominant form of such manipulation is search engine spamming (or spamdexing), i.e., malicious attempts to influence the outcome of ranking algorithms, aimed at getting an undeserved high ranking for some
items in the collection. Topics include:

• Link spam
• Content spam
• Cloaking
• Blog/forum/wiki spam
• Tag spam
• Review and rating spam
• Click fraud detection
• Reverse engineering of ranking algorithms
• Web content filtering
• Online advertisement blocking
• Stealth crawling

For more information, see http://airweb.cse.lehigh.edu/2009/cfp.html