Data Mining: Text Mining, Visualization and Social Media

Factual's Crosswalk API

Factual, which is mining the web for knowledge using a variety of web mining methods, has released an API in the local space which aims to expose, for a specific local entity (e.g. a restaurant) the places on the web that it is mentioned. For example, you might find for a restaurant its homepage, its listing on Yelp, its listing on UrbanSpoon, etc.

This mapping between entities and mentions is potentially a powerful utility. Given all these mentions, if some of the data changes (e.g. via a user update on a Yelp page) then the central knowledge base information for that entity can be updated.

You can try out the system at this demo site.

There are many challenges in this enterprise. Perhaps most importantly, there is the central issue of matching records. While there are obvious scale and performance concerns (do you really want to match 10 million Yelp entities against 10 million Yelp entities?) the key problem is, when comparing two records determining not if they are identical statements but if they are statements about the same real world entity. This is a challenge because the 'truth' about that entity may change - phone numbers change, addresses change and even names change. This can be particularly troubling with high density chain brands like Starbucks.

Browsing through the data on Factual's demo site, I came across many cases where this matching was overly aggresive (what we call over matches - statements about multiple real world entities being conflated onto a single entity).

For example, this entity:

Wild Ginger Thai Cuisine

124 Washington St
Norwell, MA 02061 US
(781) 347-4687

Is linked to the following:

http://www.yelp.com/biz/wild-ginger-norwell (correct)

but also has links to

http://www.insiderpages.com/b/3715376667/siam-cuisine-hanover

Siam Cuisine

370 Columbia RdHanover, MA 02339

781-826-1115

http://www.dexknows.com/business_profiles/the_make_and_take_kitchen-l801392119/

The Make & Take Kitchen

(781) 659-2440

433 Washington St, Norwell, MA 02061

http://www.insiderpages.com/b/5952045094/maung-thai-restaurant-hingham

Maung Thai Restaurant

74 North StHingham, MA 02043

781-740-1218

http://www.dexknows.com/business_profiles/mcdonald_s-l804749993/

McDonald's

57 Washington St, Norwell, MA 02061

(781) 878-8068

Similarly, his entity:

Wild Ginger China Bistro

3694 Burbank Rd

Wooster, OH 44691 US
(330) 345-6388

is linked to 

http://local.yahoo.com/info-15692932-pizza-hut-wooster

Pizza Hut

1831 Beall Ave, Wooster, OH 44691

(330) 262-6202

The fundamental problem here is that a number of these sites, while they have a 'main' entity on their page, also mention other entities (e.g. for recommending alternatives to the user). However, these recommendations are not guaranteed to be stable (you may get different recommendations at any time). This fundamental problem illustrates why document analysis (understanding the layout of the document and consequently what the document is 'about' through the position of mentions of entities) is key if one wants to accurately mine the web for knowledge of this sort.

  

The Recorded Future is Here

Recorded Future is a new venture which mines the web for statements that are associated with some time expressions. It then uses this corpus to describe the future in various geographies for various topics. In addition to the application of information extraction methods, they also present this information in creative visual displays.

 

The site is plenty full of jQuery goodness, but I did find the newbie experience a little puzzling (how do I navigate to the data visualization? not clear...)

Finally, I loved this quote from a satisfied customer:

"This definitely reduces time in figuring out what may or may not be happening in the future based on what has been happening in the past. It cuts that time in half. "

Advertising Executive

[HT Sundar]

 

The Interpretation of Tables in Texts, 2000

Ten years ago, I submited the final version of my PhD thesis: The Interpretation of Tables in Texts. At the time, there wasn't a huge amount of research going on in the space. Those working in the area pretty much all knew each other and would meet at a couple of conferences, generally in the OCR community as there wasn't much interest in table understanding in other research areas.

Now, there is quite a healthy interest in table understanding due in part to the promise of tabular data being a reasonable way to bootstrap semantic relationships via the large scale mining of the web.

Most recently, I spotted this paper by Finin et al :Exploiting a Web of Semantic Data for Interpreting Tables, WebSci10, 2010 which echoes much of the promise of the 'first generation' of table understanding work by the likes of myself, Dan Lopresti, Thomas Kieninger, Jianying Hu, etc. In fact, the motivating example in that paper:

 
bears a strong similarity to that from my thesis:

with the later also illustrating to some extent the complexity of table semantics.

I'm still very much interested in tabular data. Perhaps as it represents the simplest transition point from textual presentations of information to graphical, or topological representations of information.

For posterity, I've embeded the Scribd incarnation of my thesis below.

2000 - Hurst (PhD Thesis) - The Interpretation of Tables in Texts

Rexa

Rexa, which shares a pedigree with Cora from Just Systems, and which provides a similar product to CiteSeer and Google Scholar is now live, according to John Langford's Machine Learning blog. Andrew McCallum, whom I worked with back at WhizBang, is the PI for this project and comments:

Rexa is a digital library covering the computer science research literature and the people who create it. Rexa is a sibling to CiteSeer, Google Scholar, Academic.live.com and the ACM Portal. It’s chief enhancement is that Rexa knows about more first-class, de-duplicated, cross-referenced object types: not only papers and their citation links, but also people, grants, topics—and in the future universities, conferences, journals, research communities, and more.

Rexa currently provides:
* Keyword search on over 7 million papers (mostly in computer science)
* Cross-linked pages for papers, authors, topics and NSF grants
* Browsing by citations, authors, co-authors, cited authors, citing authors;
  (find who cites you most by clicking “Citing authors” on your home page)
* Web-2.0-style “tagging” to bookmark papers
* Automatically-gathered contact info and photos of author’s faces
* Analysis of research topics, their impact, and how they relate.

Coming soon:
* Much improved coverage of recent CS papers (it’s a little weak now)
* Ability to make corrections to extracted data

Coming later:
* Improved extraction and co-reference accuracy
* Much more data mining
* Broader coverage of more research fields

Rather than seeing our siblings as competitors, we believe that such services are like “newspapers for the research community”, and, just as it is tremendously important that there is not just one national newspaper, we think there should be many such services. This is especially true since increasingly they will do more than simply supply raw information, but also provide subjective analysis, pattern discovery, and predictions.

One of the key challenges that this type of vertical search has to deal with is the task of recognizing variations of named entities. A search for 'Andrew McCallum' serves as a good example of this problem. The first 4 answers refer to the same person but are listed individually. There are some great clues in the space of citation analysis that can be used to help with this problem. All the content in the papers associated with a name will have strong topical affinity suggesting that variations like 'R. Andrew McCallum', 'Andrew McCallum' and 'A. McCallum' are in fact references to the same person.

Something I've been thinking of, and I'd love to see built on top of a system like this, is a social network and commentary space that can continue the topic presented in a paper. Imagine looking up one of Andrew's papers and being able to associate follow on questions with it. Perhaps one of the authors will answer, or perhaps someone cited by that paper will make a comment.

At any rate - congratulations to Andrew in getting this out the door - the feature set looks impressive and I'm looking forward to exploring this more.