Data Mining: Text Mining, Visualization and Social Media

ScoutLabs offers 30 day free trial

Having completed their beta phase, ScoutLabs has decided to roll in a 30 day free trial period to their offering. ScoutLabs blog posts is here, and CNet covers the story here. On their plans page, ScoutLabs lists the type of data they analyze: Blogs, open social networks, photo and video sharing sites, Twitter. It is interesting to see that they don’t include any message boards – where a lot of the gold is.

A while ago, I predicted that with the acquisition with some of the major players in the space (Cymfony, BuzzMetrics, Umbria), feature innovation would be seen from the less encumbered entrants. This is exactly what I’m seeing with ScoutLabs (see their feature list page).

To Simplify, Complexify

I remember an interview – years ago – with Benoit Mandelbrot talking about how he arrived at his famous fractal set. He quoted advice he got from his mentor, Julia:

To simplify, complexify!

This was, of course, in reference to imaginary numbers. By casting real numbers into their two dimensional imaginary counterparts, an extra degree of freedom is introduced. I’ve been taking this advice to heart when considering analysing social media for expressions of appraisal (the text mining problem formerly known as ‘sentiment mining’). Much research in this space tries to view the problem in a simple way  - positive and negative words, classifying whole documents, and so on. In this field, complexifying really means working back to a model – a rich description of social context, discourse context, linguistics and psychology – which aims to describe how these expressions of attitude end up in documents. Pushing for this model is already a big win. However, complexification has another great advantage – by proposing a rich model, one can find a principled component of that model to focus on and make valuable, incremental contributions to our understanding of the space.

The appraisal framework is a great place to start, and the work of Argamon et al, and Taboada are great examples of a more principled approach to this space.

Something that I’m having fun with just now is the relationship between appraisal and word sense. While much work in this space centers on building lists of ‘positive’ and ‘negative’ words, examples like ‘strong smell’, ‘strong candidate’, ‘strong personality’, suggest that the adjective (in these examples) requires a fine, sense based analysis.

Update – I found the quote from Mandelbrot via Google’s book search:

GeeYee : Opinion Mining

Bing Liu, a professor at the University of Illinois, Chicago, is one of the founders of GeeYee:

GeeYee is the industry leader in quantifying unstructured text, enabling you to accurately discover and compare a product's or subject's collective features.  With our solution, you can drill down to the most granular level, the positive, negative, and neutral opinions on these features, in their entirety, directly from the online source. 

(note – all companies are ‘the industry leader’, I never know why people write that stuff).

Bing is well known in the sentiment mining community and has done particularly interesting work in the space of discovering which features of a product attract expressions of opinion. For example, discovering automatically that it is the battery life of the camera which people care about.

I’m trying to find out when the company started (I’ve heard Bing speaking quite recently at a number of events and never heard him mention the company; his website hints at a startup ‘still in stealth mode…’). The domain for geeyee.com was registered in 2006, according to whois. They seem to have a demo up here (id required).

Others in this space include Lexalytics, Jodange, Sentimetrix, and Sentiment Metrics.

Twitter, Sentiment, Motrin

An interesting post from Christine Sierra at Lexalytics regarding analysing sentiment on tweets – all about this ad from Motrin:

Personally, the ad seems insulting and dumb. It seems to say: please have a painful lifestyle so that we can sell you drugs.

As this post by Tom Martin suggests, the whole saga (campaign, twitter-storm, campaign pulling) may become something of a case study and certainly an opportunity for accurate social media analysis to really help.

LexaBlog on Sentiment

There is an interesting post on the Lexalytics blog about sentiment. Some of these observations resonate with experiences at BuzzMetrics, though my thinking about has changed quite radically (I’m now much more interested in what is often referred to as subjectivity though I don’t actually believe in the traditional distinction between subjectivity and objectivity). Perhaps the key thing to note here is that Lexalytics are walking the walk, not just talking the talk.

So, about my opinions on opinion. I think one has to look more broadly at ‘subjective’ statements, not just the corner called ‘sentiment’. The reason being that you need to understand all the interactions between belief, speaker performance and listener perception and context. Appraisal theory is a good place to start, but it doesn’t include the broader social contexts discussed in the LexaBlog post.

As for subjectivity and objectivity, I don’t believe that we can every really say that any statement is objective. Certainly we can talk about the intentions (explicit or implicit) of the speaker to either be objective or appear to be objective. That being said, we might recognize certain surface forms as being evidence of the intention to be or appear to be objective or subjective, but it is not clear that that really helps. If I say

“John flew to New York.”

I might appear objective, or at least more objective than if I said:

“OMG, I think John might have gone to New York!!!”

but that has no relation to whether or not John went to New York.

Opinion Mining and Sentiment Analysis; Pang and Lee, 2008

Congratulations to Bo Pang and Lillian Lee for getting their monograph on subjectivity published. The authors have been good enough to put the pdf version of the book online (here). The page for the book itself, from which you can order your copy, is here. The abstract:

An important part of our information-gathering behavior has always been to find out what other people think. With the growing availability and popularity of opinion-rich resources such as online review sites and personal blogs, new opportunities and challenges arise as people can, and do, actively use information technologies to seek out and understand the opinions of others. The sudden eruption of activity in the area of opinion mining and sentiment analysis, which deals with the computational treatment of opinion, sentiment, and subjectivity in text, has thus occurred at least in part as a direct response to the surge of interest in new systems that deal directly with opinions as a first-class object.

This survey covers techniques and approaches that promise to directly enable opinion-oriented information-seeking systems. Our focus is on methods that seek to address the new challenges raised by sentiment-aware applications, as compared to those that are already present in more traditional fact-based analysis. We include material on summarization of evaluative text and on broader issues regarding privacy, vulnerability to manipulation, and economic impact that the development of opinion-oriented information-access services gives rise to. To facilitate future work, a discussion of available resources, benchmark datasets, and evaluation campaigns is also provided.

Lillian Lee AAAI Talk on Subjectivity

Briefly, Lillian Lee’s invited talk at AAAI 2008 on subjectivity is available online. I’ve not yet had time to look through it, but am hoping it looks beyond the simple lexical approaches that have been the mainstay of this field for way too long at real models of opinion and subjectivity.

FlixPulse: Twitter Movie Reviews

More on this later; for now take a look at Nathan's post.

Opinion Mining Startups

Jodange has just released a couple of interesting Google gadgets which relate opinion to the presidential race. Jodange specializes in sentiment mining, deriving tuples of the form <holder, opinion, target>. Jodange's technology is built on ongoing research from Cornell's computer science department. You can play with the gadgets on their homepage.

There are a number of startups out there which are all focused on this one piece of technology which is often associated with social media analysis:

• Sentimetrix : Provides sentiment mining technologies to social media analysis shops.
• Sentiment Metrics : A competitor to Visible Technologies and others in the enterprise facing space.
• OpinMind : OpinMind has a new look, but its ad targeting system is differentiated by their opinion mining technology.
• Summize : "Summize connects you with the opinions of millions of people on zillions of products. We scour the web for the latest user reviews and blogger discussions, surmise their sentiments, and summarize them here."

In general, I see a number of reasons to be impressed with Jodange. They have a strong research background (investing in the science behind automated analysis is a key commitment often overlooked by companies in this space - see an earlier post on this); their formulation of the opinion mining problem is simple, but contains enough structure to ensure they focus on key questions (especially the problems of associating the opinion holder and the target of the opinion with the opinion itself); they have rolled out some simple but useful user experiences (again, UX is often overlooked - why hire a UX person when you could augment your sales staff?).

Can't Get Enough Blews?

I've been collecting all the online versions of ICWSM 2008 papers in this post. Our paper on the Blews system (of TechCrunch and Slashdot fame) can be found here.

If you haven't yet, please put your ICWSM 2008 paper online somewhere and I'll link to it. Note that we are working with AAAI to get all the paper put up freely on the conference site. This will happen, it's just taking some time.