FRAUDSTERS beware, sports nerds rejoice: a new search technology that turns graphs and charts back into raw data could make it easier to spot research cheats and tally sporting statistics, among other things.
Computer scientists Lee Giles and Prasenjit Mitra have worked with chemist William Brouwer at Pennsylvania State University to develop a system that can extract information from illustrations in research papers. 鈥淎cademics frequently report important findings in graphs,鈥� says Brouwer. 鈥淲e want to see the data liberated from these images.鈥� That would allow others to reuse the data, perhaps to compare results.
Now that most scientific papers exist in digital form, it should be possible to get a computer to gather such data automatically 鈥� rather than unfortunate students having to measure printed charts with rulers. In practice that has proved difficult, Giles says, because it is hard to teach a computer to understand what constitutes a graph and interpret it appropriately.
Advertisement
The Penn State team claims to have cracked the problem with Linux-based open source software that can 鈥渃rawl鈥� a database of papers looking for the salient features of graphs 鈥� 2D plots containing combinations of lines, curves, data points, axes and caption text. Once a plot has been recognised, an algorithm works out the units on the axes and then pulls the data from the graph, presenting it to the user as a table ().
Brouwer is using the system to extract data about molecular structures from nuclear magnetic resonance spectra, and the team hopes other scientists will find plenty of other applications. Since the system could in principle be used to extract information from any chart, it could also prove useful for non-scientists, particularly if integrated with internet search.
For example, 鈥渁 sports fan could take a web page where statistics have been published in graphical form, and use our tools to quickly make a spreadsheet,鈥� says Mitra. That would make it easy to compare how different players or teams have performed without having to gather detailed statistics all over again.
And scientific fraud-busters could use the technology to 鈥渄etermine statistical similarities鈥� between data sets, says Brouwer. It was duplicated graphs that tipped off investigators to one of the most infamous scientific frauds of recent years; the discovery in 2002 that Bell Labs researcher Hendrik Sch枚n had fabricated apparently ground-breaking research in electronics (New 杏吧原创, 5 October 2002, p 4). 鈥淚 would welcome any tools that facilitate the detection of plagiarism,鈥� says David Williams, editor of the journal Biomaterials, who last year complained about plagiarism in scientific papers.
The system might also prove useful for spotting researchers who re-use the same data to stretch a single piece of research work across multiple publications.
