Text Mining for Beginners, redux

This week a team of Northern Illinois University students and their faculty coach presented their findings after a semester devoted to investigating text-mining from the perspective of a novice.

They have produced a report in which they provide a basic description of text mining itself, including a review of some of the types of procedures used to detect patterns within a very large body text; the types of text available for text-mining work; a discussion of structured, unstructured, and semi-structured data as they pertain to text-mining work; the importance of preparing digital texts (especially those created by Optical Character Recognition Software) for mining activities; and reviews of three well-known text-mining applications: Mallet, Weka, and RapidMiner. Of these, the first two are freely-available open-source software; the third is available in a free demonstration version but requires purchase in order to make use of its most powerful capabilities. These reviews include brief discussions of the types of text-mining activities (i.e., topic modeling, document clustering, sentiment analysis, etc.) that each makes possible. Finally, the report describes the team's activities in using the three applications to perform analyses on sample bodies of text, and the results produced.

I hope to work with the team and their coaches to round the report out into a resource that I can distribute to interested members of the NIU community. We also hope to make it available via Huskie Commons, the university's institutional repository (http://commons.lib.niu.edu ).

Text mining for beginners

I am now Director of Digital Scholarship at Northern Illinois University Libraries. This means that it is now my job to work with faculty members seeking to employ technologies like Geographic Information Systems, text-mining and data visualization - helping those with little experience in such work find a way to put the technology to work.

As I really do not know very much at all about these activities, my job is now an exercise in learning something new. To this end, I have sought some help.

This fall I am working with a team of three Northern Illinois University students and their faculty coach, who will provide me with an evaluation of several open-source text-mining utilities, as well as a more general review of resources available for a scholar or other practitioner who might want to take up text-mining but lacks any experience in the work.

I spent last summer trying to identify and prepare text materials for their use in the evaluation of the utilities, and found very little information explaining how to begin a text mining project - i.e., finding digitized texts, selecting texts for research, and working them into a format suitable for use with the software - available.

I am looking forward to the students' report, and will try to bring their findings to the attention of historians and other humanities scholars who might be interested in text-mining.

Digital/Online Materials and their Place in Historical Scholarship

At the recent meeting of the American Historical Association in Washington, D.C., I made a presentation as part of a discussion session (i.e., not a regular panel - we sat in a circle and talked after very short presentations made by people sitting as part of the circle) exploring digital materials, ranging from blogs and web sites to social media, and the questions that they raise as scholars begin to make use of them as primary sources. Other presenters talked about the future of MOOCs and crowd-sourcing the search for elusive information about a relatively obscure historical figure. I discussed the work of the Digital POWRR project and the challenges presented by the fact that digital objects are generally subject to loss in the relatively short term due to a number of reasons, including hardware and software incompatibility and the degradation of storage media.

One major question that emerged in the discussion was the status of social media materials and other online, digital sources in light of the fact that they are so prone to loss. One presenter at the preceding panel (our discussion group was part of a linked set of two events) described how she had based her work on Pakistani women in part on a web site that no longer existed, apparently because of hacking activities undertaken by parties believing that Pakistani women should not express themselves in this format. The presenter said that she had printed out the sites pages for her own record and thus could document her use of the source. But this made me wonder about the future practice of history.

So, what of digital sources like blogs, web sites, and social media objects like tweets? Digital objects' intrinsic frailty and the complex, easily disrupted nature of the internet used to present them make them fundamentally unreliable as primary sources, at least by the standards developed for the use of analog/paper media materials.  

It seems to me that although history is certainly not a science in any way, historians are similar to scientists in at least one regard. Much like a scientific discovery can only be accepted and confirmed as other practitioners are able to repeat the experiment and yield the same result, historians are accustomed to being able to lay their hands on a paper source cited in a footnote. Manuscripts are usually unique items, but if one travels to the archive and looks in the box and folder number cited, the item will be there. There may be a very small number of copies of a book, but if one is willing to make the trip to the right library, the book will be there. Historians will of course debate a scholar's reading of a source, but the existence of the source itself is fundamental to the discipline. If the item is not there, practitioners may rightly begin to ask questions about the legitimacy of a work citing it.

Many of the participants in the AHA discussion emphasized the need to preserve online digital materials as fully as possible. I certainly concur. But a whole host of problems, not the least of which is the considerable expense involved in the curation/preservation of digital materials, make this impossible. We will have to face that fact that a considerable amount of online digital objects that future historians may want to use as evidence will simply disappear. 

In this situation, several questions occur to me: How will we evaluate work citing online materials that are no longer existent? What if scholars relying on such missing evidence can produce a print-out or other facsimile of the materials? Can we distinguish cases of vanished evidence in which legitimate facsimiles exist from cases of academic fraud?