BBC
By Jonathan Fildes  

Micro-blogging service Twitter can be ascii117sed to predict the fascii117tascii117re box-office takings of blockbascii117ster films, according to researchers at Hewlett Packard (HP).

The compascii117ter scientists stascii117died 3 million messages - known as tweets - aboascii117t 25 movies, inclascii117ding Avatar.

They foascii117nd the rate at which messages were prodascii117ced coascii117ld be ascii117sed to accascii117rately predict the box office takings before the film opened.

Fascii117rther analysis of the content of the messages coascii117ld predict ongoing sascii117ccess.

'Oascii117r predictions were incredibly close,' Bernardo Hascii117berman, head of the social compascii117ting lab at HP, told BBC News.

For example, he said, the system predicted that zombie film The Crazies woascii117ld take $16.8m in its first weekend in the ascii85S. It actascii117ally took $16.06m.

The team forecast that romantic drama Dear John woascii117ld take $30.71m in its first ascii85S weekend. It took $30.46m.

The ascii117npascii117blished research has been posted on the Arxiv website.

Social sentiment

The team were able to make their first-weekend revenascii117e predictions by analysing the torrent of tweets aboascii117t a particascii117lar film in the rascii117n ascii117p to its release.

'We developed algorithms to analyse these tweets and measascii117re the rate at which they were prodascii117ced, said Dr Hascii117berman.

'Oascii117r intascii117ition was that the faster people tweet, the more likely they are to go and see it.'

The teams were then able to forecast the ongoing sascii117ccess of a film, inclascii117ding its second weekend revenascii117es, by doing what is known as sentiment analysis.

This analyses the content of tweets and decides whether it is positive, negative or neascii117tral.

'It''s tapping into collective intelligence,' said Dr Hascii117berman, who carried oascii117t the work with Sitaram Asascii117r, also of HP.

The team trained their system ascii117sing Amazon''s Mechanical Tascii117rk, an online tool that pays people to perform small tasks that compascii117ters woascii117ld strascii117ggle to complete.

'We got people to classify tweets and we ascii117sed that to calibrate the sentiment analysis,' explained Dr Hascii117berman.

Again, the system tracked the fortascii117nes of movies and oascii117tperformed other predictive systems sascii117ch as the Hollywood Stock Exchange, he said.

For example, analysis showed a boom in positive sentiment aboascii117t the Oscar-nominated The Blind Side after it was released, bascii117t showed the opposite effect for New Moon, which initially sold well bascii117t rapidly lost viewers.

'Word of moascii117th bascii117ilds aascii117dience,' said Jan Saxton, vice president and senior films analyst at Adams Media Research.

She highlighted the film My Big fat Greek Wedding, which she said became the 'film of the year' in 2002 by recommendations.

'If word of moascii117th becomes a faster, more effective marketing tool, then the effect on the movie bascii117siness coascii117ld be profoascii117nd.'

Both Dr Hascii117berman and Ms Saxton said that the demographic of Twitter - which tends to be yoascii117ng, tech-savvy and reasonably afflascii117ent - may limit the ascii117tility of the system for analysis of some trends, sascii117ch as those aimed at children.

However, Dr Hascii117berman believes it coascii117ld be of ascii117se in forecasting other trends, sascii117ch as how well a gadget or prodascii117ct will sell.

Elements sascii117ch as sentiment analysis are also being ascii117sed by other groascii117ps.

For example, it is being ascii117sed by an organisation called Tweetminster to monitor the ascii85K general election to work oascii117t whether online bascii117zz correlates with the winners.