Data Mining: Text Mining, Visualization and Social Media

Search Doesn't Work: Story 4

I really enjoy reading Infosthetic - a blog about information visualization. When reading it, I often wonder - how does the author discover these interesting visualization stories and images? One way in which I find new and interesting data graphics is via image search. But am I using the best image search engine? I typed 'the best image search' in to Google and got this as the top result: Reviews of the Best Image Search Engines. Sounds like it is right on the money - until you realise that it was written on July 31st 2001. Things have changed a lot since then. Interestingly, that date of this article is not on the page itself, but on the page linking to it.

The second result is: The 5th Annual Search Engine Watch Awards a little better, being dated March 31st, 2005. The third is hard to date, bu the fourth is here, dated May 2006.

What is going on here is that some information online is dated. It is possible to approximate the date of a page via its ingest date (the date that the crawler discovered it), though this is pretty shaky at best. There may be information on the page that helps - though as we saw with the first example, that may be on another page on the site. In addition, older articles may well accumulate more inlinks, boosting their static ranking, thus compounding the problem.

I appreciate that I could use better terms. However, I'd certainly like the search engine to be sensitive to timely information, and to factor that in to its model of relevance.

See previous posts on search problems.

Search Doesn't Work: Story 3

I come from a small town - a very small town - in Scotland called Elie. A search for 'elie, fife, scotland' on Google Maps produces a very strange result - a point in the middle of Elie Road outside St Andrews and a long way from the village. The same search on Google produces a page about the village. The mapping system's assumptions about intention - even though it interfaces a far more structured data set - appear to be wrong whereas the combination of relevance and static ranking in the document search system produces an appropriate response.

The problem here is that search engines (or Google at least) has no real ability to interpret queries. The paradigm of the main stream search engine is to treat everything as tokens (strings of characters) and to rely on behaviour of such things as page rank and statistical models of relevance to conjure results that appear to be meaningful. Because the interpretation in the mapping system and the document system are different, in this case it fails to produce equivalent interpretations of the same query (and the same intention).

The solution lies in (can you guess) natural language processing.

See here for more on search problems.

Search Doesn't Work: Story 2

NLP means many things. To me it means Natural Language Processing. To others it means neurolinguistic programming. When I search for the bare term 'nlp' in Google, I just get results with the second sense - same for other search engines. If I search for 'William Cohen', the first result on Google is for my friend Prof. William Cohen and the second for the other chap. If I search for 'Michael Jordan' I get similar results - the first page contains pages for both Prof. Michael Jordan and the sports star.

So why don't I get this for NLP? Why no mixture of results?

I can, of course, add in some terms. 'nlp hurst' returns results that are about Natural Language Processing due to my blog and other related content. 'nlp smith' doesn't do this - not too surprising due to the frequency of the name. However, this disambiguation is only happening via the very literal constraining term - note that the adverts served for the Natural Language Processing results are for neurolinguistic programming exclusively. Opps!

A more pertinent constraint might be 'nlp ai' (Natural Language Processing and Artificial Intelligence). Not surprisingly this produces far better results. But - hang on - still adverts for neurolinguistic processing on Google.

Word sense disambiguation is a core requirement for a search engine. The problem - the same text having more than one meaning - can certainly be reduced by the user. However, it seems that there is a great amount of scope that could be explored on the interface side. Google is definitely aware of the problem, which is why results for ambiguous names produce multiple sense results pages, but they (and the other major engines) are way behind systems like Vivisimo's Clusty which produces appropriate results for the NLP problem.

See previous posts on search problems.

Search Doesn't Work: Story 1

It is always informative to look at cases where one simply can't get the results one is looking for via a standard text based search interface. In part, such examples help dispel the aura of infallibility that surrounds some of the most popular search engines. In addition, these examples help us understand what is broken and what new paradigms are required to get beyond the current cul-de-sac that is search technology.

Here is an initial story: I have plenty of information describing the political geography of China. I have data describing the relationship between populated places (cities, towns, etc.), their local administrative regions, their co-ordinates, and so on. What I need is a list that will help me map these transliterated terms (e.g. Beijing) into their Chinese language counterparts (e.g. 北京). This turns out to be next to impossible for most search engines. I say next to impossible as I'm assuming that the data exists somewhere, and I'm assuming that there is some (possibly eccentric) search term that will surface it. I want to be able to express a description which generalizes the page I want to find (e.g. 'chinese city names in chinese').

One creative way to possibly solve this is to list a few examples (e.g. '北京 beijing 上海 shanghai') but, of course, that only works if you know the names (it does produce some helpful results, but doesn't really find the data I am after).

Let's call this the descriptive search problem - I want to describe the data set I am looking for. In this case, the data is relational, and I want to capture the type of elements (Chinese city names in Chinese and Chinese city names in English) and the relationship between them (alignment of some sort).