The Oscar Senti-meter rates tweets that cite Academy Awards nominees in the Best Actress, Best Actor, and Best Picture categories and scores them on a more-positive/more-negative scale.

What do I like?

1. The timeline control! Move left (earlier) or right (later) to see tweet volume and sentiment scoring for different dates.
2. The listing of selected external events as part of the time control. Some of are quite relevant, for instance, the January 15 Golden Globe awards. Scroll to that date to see the effect on stats for Meryl Streep, for instance.
3. There are at least some linguistic smarts built into the system (although it’s impossible to determine, by looking, how much). “Merryl Streep” and “dragon tattoo” are resolved to the correct, full names.

And what’s lacking, items for the To Do list for next go-around?

4. Provide click-through to underlying tweets, and not just three, all of them or at least a large sample. The current tool is a tease, and it’s not as if the tweets are private or proprietary. But click-through isn’t, alone, quite enough…
5. Show the sentiment classification of the underlying tweets. The LA Times tells us only that “the Senti-meter combs through a high volume of tweets daily and uses language-recognition technology… to gauge positive, negative and neutral opinions shared in the messages.” I want to see for myself how good the technology is.
6. Animate the timeline control. A simple play button would march the display through the range of dates.
7. Let me trace opinion across dates. All that’s needed is a trail attached to what’s now a bouncing dot although to be done effectively, it would be best to focus on a particular nominee and replace the current horizontal axis with a time axis.
8. Let me filter tweets — for instance, to exclude tweets with the #GoldenGlobe hashtag — and dynamically recalculate based on my filtering.
9. Let me explore how sentiment about Actors, Actresses, and Movies is linked. It would be interesting, for instance, to see if and how tweet volume and sentiment about Meryl Streep and the Iron Lady are linked.
10. I want Sorting, Ranking, and Thresholds. Let me sort nominees, ascending and descending, within the Actors, Actresses, and Movies categories, by number of tweets in a given day and over all days. Let me restrict nominees to the Top N or Bottom M based on number of tweets and on positive or negative sentiment. Let me apply a threshold, for instance, show only nominees who have been mentioned in at least 100 tweets.
11. Content enrichment. I’ve never heard of the movie Carnage, nor of actor Joseph Gordon-Levitt. How about a hyperlink, behind each name, for more information?
12. Explain how the system works!

Frankly, I suspect that Senti-meter is classifying tweets at the message rather than at the feature level, that is, for individual Actresses, Actors, and Movies. I’d be happy to learn otherwise, but there’s no indication about method on the Senti-meter site.

I have one, last ask:

13. Find a way to make Senti-meter useful, more than eye candy.

My last point is seemingly a tough one. The best way to make any information-delivery site or utility more useful is by designing it, from the start, to respond to business needs. If you don’t, the majority reaction will be a yawn, So What? (and I don’t mean the Miles Davis version), a few moments on-page dwell time and then the site visitor is on to the next thing. Is there a business need met by the Oscar Senti-meter? There could be, for instance, in predicting box-office and rental demand. But of course we assume that the Oscar-scoping interface is just a demonstration of the possibilities afforded by the technology and the implementation.

The types of interactivity I’ve described would boost Senti-meter’s business usefulness. So would provision of goal-aligned, beyond-polarity sentiment classification. A plug: We’ll discuss these topics and much more at the up-coming Sentiment Analysis Symposium, a conference I organize, May 7-8 in New York. I do come back to these topics frequently. I wrote about them in a recent review of Neurolingo’s beta Sentibet system. Sports betting, Oscar sentiment: Two peas in a pod. Sentibet differentiates Predictions, Feelings, and Wishes. I may think that Tilda Swinton is a great actress but hope that Michelle Williams wins the Oscar and expect Viola Davis to take the prize this year. Surely this granular, focused level of analysis isn’t beyond IBM’s capabilities, USC Annenberg’s research interests, or the L.A. Times‘s information-delivery mandate.

In sum, my review scores the Oscar Senti-meter moderately negative, with lacking outpolling like by 10 to 3. The IBM-USC-L.A. Times system is a worthy nominee, but not a winner. The players have the tools and resources to create something great, both technically strong and usable for a mass audience. I’m looking forward to the sequel.

I offer observations and a bit of analytics history in response to Gary’s question, “Could the great classical music composer Ludwig Beethoven successfully implement business analytics in an organization?,” followed by my answer. Oh, in case you’re wondering: Business analytics is a collection of techniques and tools that model and extract insight from data, applied for business purposes.

Beethoven (1770-1826) was a near contemporary of the founder of analytics, Carl Friedrich Gauss (1777-1855). Gauss invented key methods that underlie analytics although of course he didn’t write software or design business solutions. Gauss’s work, nonetheless, had contemporaneous applications. Foremost of them was gambling, but his larger mathematical legacy comprises the longest of long tails.

Check out text I found, searching on the two, on Beethoven and Gauss, from a blurb describing a 1970 biography of Gauss by Tord Hall:

“Carl Friedrich Gauss (1777-1855) is generally ranked with Archimedes and Newton as one of the three greatest mathematicians that ever lived. His work, in terms of its all-pervasive importance, its painstaking attention to detail, and its completely developed beauty, somehow reminds one of the work of Beethoven, his contemporary and compatriot.”

Gauss himself built on mathematical methods created, before Beethoven’s birth, by figures such as the Bernoullis (Johann, Jakob, and Daniel) and Leonhard Euler. Society in Beethoven’s time was, in a sense, their heritor, just as we are today. If you do any form of (social-)network analysis, you owe Euler. Take a look at slide 11 of my presentation Text, Content, and Social Analytics: BI for the New World. I illustrate analytical modeling with Euler’s famous Bridges of Konigsberg problem. Daniel Bernoulli’s Fluid Equation, in its elegant simplicity, is as monumental as Beethoven’s tortured Grosse Fuge, but unlike a Beethoven string quartet, it’ll keep you in the air next time you fly the friendly skies.

Beethoven’s life also overlapped those of Joseph-Louis Lagrange (1736-1813) and Siméon Denis Poisson (1781–1840), who each invented techniques that are of central importance in aspects of analytics. I would be stunned to learn that Beethoven gave more than a moment’s passing thought to any of these towering figures, to Lagrange or Poisson, to Euler, the Bernoullis, or Gauss, whose stature in mathematics equals Beethoven’s in music.

So, while I like Gary’s thinking and his ability to see something bigger in the work we do every day, I believe his invocation of Beethoven is fundamentally off track. Beethoven admired revolutionaries, but his admiration was that of a Romantic, for the political and literary. He was not an Analyticist (sorry for that word). I see only the faintest ghost of his compositional methods in business analytics processes.

Beethoven’s affinity for the liberator Napoleon, and for literary figures such as Johann Wolfgang von Goethe and Friedrich Schiller (author of the “Ode to Joy,” set by Beethoven in his Ninth Symphony), is telling. So happens that I read Goethe’s Young Werther just a couple of months ago. It was truly hard for analytically-minded me to identify with Goethe’s emotionally overwrought Sturm und Drang hero, to see any redeeming quality in the character’s self-absorbed obsession. Yet Young Werther and similar works were the rage of Beethoven’s age, even while the Analyticists were creating tools that rationalized the world, enabling the Industrial Revolution and modern business methods.

Simply put –

Beethoven lived in analytics’ seminal period, but he wouldn’t have cared a rat’s ass about business analytics. There is very, very little I see in today’s business-analytics world, or that Beethoven would have seen, that truly fits the Romantic’s conception of revolution or of himself.

By contrast –

Occupy Analytics!

Turn business analytics to public good. It can be done. That would be the movement for Ludwig van Beethoven. Dance like a spinning top.

Our tag line remains unchanged:

The symposium bridges technology and business in one of the most exciting applications to emerge in recent years: software that discovers business value in opinions and attitudes in social media, news, and enterprise feedback.

But our program is better than ever. Keynotes kick off the morning sessions:

• Prof. Jan Wiebe of the Univ of Pittsburgh is a sentiment-analysis pioneer. She will speak on Sentiment, Subjectivity, and Sense, providing a leading researcher’s view of the state of sentiment, and where we’re heading.
• Drinking from the Fire Hose authors Chris Frank (American Express) and Paul Magnone (Openet Telecom) are slated to tackle (working title) Emotional Versus Rational in Customer Decision Making.

If you’re sold, click here to visit the registration page.

The Sentiment Analysis Symposium series has drawn steadily growing numbers, from 95 attendees in April 2010 to 120 (capacity) in April 2011 to 154 in San Francisco last November. (The linked text will take you to pages with videos of the last two symposiums’ presentations.) We’re both building on past success and aiming to keep the program fresh, on the leading edge of sentiment technologies and applications. The symposium is not just another Social Media Blah, Blah, Blah show. Instead, it showcases ground-breaking sentiment-analysis applications and provides unique learning opportunities.

Learning is why, in addition to talks by Jan Wiebe and a new presentation by Prof. Ronen Feldman of Hebrew University, who co-founded solution provider Digital Trowel — Feldman will speak on Unsupervised Learning: Insight in Business News, Medical Forums, and Public Filings — we’ll once again have a half-day, day-before Practical Sentiment Analysis tutorial, on May 7, this year taught by Prof. Bing Liu, another sentiment/opinion-mining expert. Liu was featured in a January 26 New York Times article, For $2 a Star, an Online Retailer Gets 5-Star Product Reviews.

Other May 8 speakers include two of our San Francisco symposium panelists, Banafsheh Ghassemi of the American Red Cross (A Multi-Channel Proposition: Customer Sentiment and (Much) More) and Carol Haney, now with Toluna (Tween Pants Cut Too Low!! (or, Combine Survey Research & Social Monitoring to Discover the Unknown)). (Carol’s original title was Tween Butt-Crack: Finding new trends of dissatisfaction by integrating social media data with open-end data, but I asked her to make it both more and less descriptive.) Rich Brown from Thomson Reuters is another repeat, a highly rated presenter at the April 2010 symposium. Brown’s title this go-around is Forecasting Financial Market Response from News and Social Media Sentiment.

New to the symposium, we have Andera Gadeib, CEO of German-based Dialego AG, speaking on Market Research Beyond Sentiment: Differentiating the Engaged and Pleased.

You may have noticed an interest in talks that tell you how to use sentiment analysis and how to combine it with data from diverse online and enterprise sources, and on beyond positive/negative analysis that help you understand and exploit emotion, mood, and opinion in the spectrum of sources. We also focus on User Perspectives in a series of three shorter presentations followed by a panel discussion:

• Can automated sentiment analysis be less accurate than flipping a coin? — Dana Jacob, Yahoo
• Who are the most loyal customers? — Sobhan Hota, Fidelity Investments
• “How Can I Listen If I’m Talking?”: The Power Of Social Media Listening — Frank Cotignola, Kraft Foods

As usual, we’ll have a series of 5-minute lightning talks just before the lunch break:

• Sentiment As A Service — Michael Tupanjanin, Metavana
• Political Sentiment Analysis — Dr. Stuart W. Shulman, Texifter
• Sentiment Visualization — Vsevolod Gavrilyuk, SemanticForce (tentative)
• Data-driven Sentiment Analysis of Financial News — Matt Sommer, MarketChorus Inc.

And we have a new Innovators & Innovation panel to close this year’s program, with Leslie Barrett (The Ladders), Prof. Bing Liu (University of Illinois at Chicago), and Romi Mahajan (Metavana).

Lastly, in order to maximize networking opportunities, we’re once again planning evening-before and post-symposium receptions.

I’m really looking forward to the May 8 symposium. I hope you’ll consider joining us. If you have questions do get in touch, and I hope to see you there.

P.S. Thank you to sponsors Attensity, Lexalytics, and NetBase (there are others pending), whose support makes the symposium possible.

Here are my planned panel intro and starter discussion questions, shared in the hope that they, on their own, will provide insights into questions.

Panel intro

Our brief is to look at “Analyzing the vast amount of readily accessible data that flows constantly across the internet uncovers details, information and relationships that were unavailable a few years ago. This panel will examine methods and practices to glean sentiment from words and text, look at using this data to predict the future and discuss what information social networks can reveal – all accomplished with no limitation on language and on a real-time or near real-time basis.”

There’s a huge amount in that assignment. I count a dozen notions that are worth exploring. Start with “vast amount,” “readily accessible,” “data”,” “flows,” “constantly,” “analyzing… to uncover details, information, and relationships,” “unavailable a few years ago.” Then there’s “sentiment,” “predict the future,” “information social networks can reveal,” “no limitation on language,” “real-time or near real-time.”

That makes twelve notions (putting aside that some of them aren’t even atomic), or perhaps the count is muliplied given the interplay among individual notions. How do we detect “events” in “flows” and use them to “predict the future”? How is “sentiment” “data”? Is it truly “readily accessible,” and is there really “no limitation on language,” particularly when seeking to understand subjective information such as sentiment?

Let’s hear what our panelists have to say on the these points such as these, in particular as relates to today’s theme, Real Time Awareness.

… and Questions

It’s my job as moderator to prompt an interesting conversation. These questions will, I hope, serve the purpose. My expectation and hope, by the way, is that we’ll get through only a few of them. Here they are:

1. Let’s start with sentiment. What role do sentiment, opinion, emotions, attitudes — various forms of subjectivity — play in analyses of the online and social worlds?
2. Say you’re an analyst tasked with some business or research function (and I do include here study and formulation of policy, program analysis, intelligence, political strategy, and so on). There’s lots of information in text: “named entities” (people, place, organizations, and so on), geolocation, events, sentiment and opinion, identity clues, and so on. And then there are those imperatives: “real time,” prediction, flows. Where do you start, that is, what are the most important elements to understand, and the most important technical capabilities to have?
3. We’re interested in social networks. Well, myself, I don’t view Facebook or Twitter as a social *network*. Instead, they’re platforms where networks consist of connected individuals and organizations whose links are rarely limited to any single platform. Certain technologies provide the ability to track individuals across platforms although they’re as-yet controversial. Anyway, to my question for you: How do analysis of content and of networks mesh up? Analytically, how do you match what people say to their actions and interactions? What can be learned from this sort or analysis?
4. Has there been anything really cool, on the language technology front, that has emerged recently? IBM Watson, Siri, Wolfram Alpha? Something else? Only one rule for responses to this question: Please tell us about something other than what you’ve been working on yourself.
5. How important, and how doable, are cross-lingual or multi-lingual analyses?
6. There’s a temporal dimension to our analyses. Information sources capture, and are themselves, events. Patterns both simple and complex emerge from studying sources over time. Even the meaning of information evolves, both because today’s observer has different concerns from yesterday’s and because language changes over time. What’s your view on temporality and temporal analyses?
7. In our discussion before today, some of you wanted to talk about the interplay between technical approaches and social-science techniques. Please tell us about that interplay.
8. How is our field evolving? Where have we been and where are we heading? How do we make our tools more relevant now and more adaptable to emerging needs?

There is a difference. Let’s explore it via the definitions that follow, first of text analytics, then content analytics, and finally content enrichment and where the ensemble takes us. First definition –

Text analytics is a set of software and transformational steps that discover business value in “unstructured” text. (Analytics in general is a process, not just algorithms and software.) The aim is to improve automated text processing, whether for search, classification, data and opinion extraction, business intelligence, or other purposes.

To expand on this definition a bit, to bridge from text to the wider content world:

Text analytics draws on data mining and visualization and also on natural-language processing (NLP). Supplement NLP with technologies that recognize patterns and extract information from images, audio, video, and composites and you have content analytics.

The concept of content enrichment is easy to grasp: Every link in this article — Web links are accomplished via the HTML “a” anchor tag — is a bit of content enrichment. And semantic content enrichment? Marie Wallace puts it this way, focusing on text but with concepts that extend to the broad set of content types:

When I think about semantic enrichment, I see it as transforming a piece of content into a linked data source. In order to do this you do indeed need text analytics for entity and relationship extraction, but you need more than that…. A text analytics engine might recognize that [Marie Wallace] is a person, [Ireland] is a place, and Marie comes from Ireland and annotate the entities/relationships found. However when doing semantic enrichment, I would want to convert those annotations to openly addressable URIs that contribute to the linked data cloud.

URIs are uniform resource identifiers, Semantic Web terminology for IDs, unique within a namespace, that name or locate things. Web URLs (e.g., http://whitehouse.gov/) are a type of URI.

Rather than write my own annotation elaboration, I’ll reuse one from the Web site of Ontotext, a semantic-technology developer:

Annotation, or tagging, is about attaching names, attributes, comments, descriptions, etc. to a document or to a selected part in a text. It provides additional information (metadata) about an existing piece of data.

Semantic Annotation goes one level deeper:

• It enriches the unstructured or semi-structured data with a context that is further linked to the structured knowledge of a domain.
• It allows results that are not explicitly related to the original search.

The earliest specific semantic content enrichment reference I’ve encountered is in an Ontotext paper, Towards Semantic Web Information Extraction, presented at the 2003 International Semantic Web Conference (ISWC). The paper covers work based on Ontotext’s Knowledge and Information Management (KIM) platform, which in turn relies on GATE, the General Architecture for Text Engineering, an open-source text-analysis framework and toolkit, Apache Lucene, and other technologies. The Ontotext folks have other, related papers posted on the company Web site.

The Ontotext materials help explain the role text/content analytics can and should (but doesn’t often enough) play as a Semantic Web generator. The entities, concepts, events, and other features discerned, via content analytics, in text and rich media not only enable smart content; they can also be loaded to knowledge bases (which I won’t get into here, other than to say that systems such as IBM Watson and Wolfram Alpha use them) and Semantic Web triple stores.

There are other solution providers in the content analytics meets semantic annotation/enrichment game. In addition to IBM and Ontotext, they include HP Autonomy, MarkLogic, OpenText, Temis, and the nascent, open-source IKS project. Other vendors offer enterprise-strength building blocks, for instance, SAS via the various SAS Text Analytics components.

I’m sold on this stuff given the business benefits for content producers and content consumers alike. These technologies — and the interplay between analytics and semantics — are key in making sense of the digital universe.

Serafim Scandalos, an exec at Neurolingo, a Greek natural language processing (NLP) specialist, asked my reactions to Sentibet, currently in beta, which performs text analytics on sports-related tweets. Sentibet looks for Wishes, Feelings, and Predictions in tweets related to specific sports and specific contests/games, producing Sentiment Based Forecast (SBF) for each event.

Serafim gave a lightning talk at an October, 2010 conference I organized. Neurolingo’s Mnemosyne platform seemed promising then, and Sentibet seems on the right track now, worth a look. (I do not have any business relationship with Neuroling although I’m hoping Serafim or a colleague will present Sentibet at my next conference, the May 8 Sentiment Analysis Symposium in New York.)

The Sentibet interface is nice enough although a refocusing on sports contests with available analyses is in order. It’s off-putting to see the “SBF info not available yet!” and “Optional empty text…” messages that now populate many screens. Tell me what you know, not what you don’t.

To get to analyses, try the Most Tweeted and Finished Games areas. (Be patient: Response is far slower than it should be, typically 8 seconds to bring up a page. Something to work on.) Clicking through a particular match…

The Prediction/Feeling/Wish categorization is conceptually quite interesting. It reflects the richness of human emotion, that we may have a variety of different purposes and meanings when we express seemingly similar feelings, attitudes, and opinions. That you hope Manchester United will beat Arsenal doesn’t mean you think a draw won’t be the outcome.

Similarly, I really like the Team A (Home)/Team B (Away)/Draw categorization. It’s goal-aligned and therefore far more useful than the positive/negative/neutral sentiment classification that’s typical of the Twitter-sentiment toys that are far too common.

The best part is that you can view the tweets behind the dual classifications, and you can apply filters to select only tweets in certain categories. I was skeptical of the analysis displayed: 2% of tweets on an Arsenal-Manchester United match Wished for a draw?! (Who hopes for a draw?) Well, SentiBet got it right. Witness:

@darrenpage1983 revenge weekend spurs bt man city or atleast don’t get stuffed 5-1 arsenal man u would like to c a draw there as never want the scum to win

@Slickon_CFC Good Morning guys. Its super sunday today and i’m hoping for a City win and a draw between United and Arsenal. #Top3

@annemacey02 @Betherz_BCFC I want my spurs to thrash manc and a manu arsenal draw :p

Note a few points:

• Accurate co-reference that correctly sees “Manchester United” – “man u” – “manu” as one and the same.
• Ability to distinguish two different matches in a single tweet.

This is strong text analytics.

It’s a good sign when an app leaves you hungry for more, for more data, for additional analysis. Sentibet does. Within the Finished Games section, it would be interesting to be able to correlate scores and forecasts, to understand the correlation between tweet aggregates and match outcomes. How well do Sentibet forecasts predict final scores? It’s a Neurolingo work-in-progress as explained on the Sentibet Post Game Analysis page.

It would be even cooler to connect forecasts and betting lines, that is, bookmakers’ odds. As the expression goes, “The proof of the pudding is in the eating.” If you consider sports betting as a business, they payoff isn’t in scores predicted, it’s in your gambling winnings. Financial-market traders exploit the price-expectation gaps, and so do winning gamblers. Maybe another to-do for Neurolingo?

Serafim Scandalos tells me Neurolingo is seeking financing to accelerate service development, also that Sentibet is simply one application of the underlying Mnemosyne platform, meant to demonstrate possibilities to potential customers and investors. Judging from Sentibet, Neurolingo seems like a good bet.

Stephen Arnold, “Thy wit is a very bitter sweeting; it is a most sharp sauce,” well served in to the goose (as Arnold characterizes himself) himself. Steve, if you want to be paid for your work, ask up front. The payback you offer Temis, a text-mining solution vendor, says far more about you than it does about the company you target.

In the words of a marketing exec at a semantic-analysis company, not Temis, “What a super childish article.”

Arnold’s article is no dissection of the fine points of competitive market positioning. It deals broad cutting strokes (“Sectumsempra — For Enemies”?) in an attempt to hack away, well, much more than just the credibility of a vendor’s public relations. Arnold brackets his article introduction of Temis (in a lede buried four paragraphs in, as deeply as his hatchet blade) around a Jean Genet quotation,

“I recognize in thieves, traitors, and murderers, in the ruthless and the cunning, a deep beauty – a sunken beauty.” (The Thief’s Journal)

WTF? Temis is (likened to) thieves, traitors, and murders!?

What spurred this attack, which also devotes several paragraphs to an attempted take-down of Temis’s client, the American Society for Microbiology?

Arnold received a press release “attributed to an individual identified as Martine Fallon [sic]” that he characterizes as spam.

Arnold writes, “I considered that Martine Fallon may be a ruse like Betty Crocker.”

WTF redux!

For the record, it’s Martine Falhon. She is marketing and communications manager at Temis. Her full rap sheet is publicly accessible on LinkedIn. She’s a quite pleasant person, actually. I met her when I keynoted Temis’s 2009 user conference (which, by way of disclosure, was a paid gig, and Temis was one of seven sponsors of my Text Analytics 2009 market study).

Oy. A vendor sends a release to an analyst in its space — with an unsubscribe option that Arnold was able to use! — and Stephen Arnold blows a gasket. Yes, Arnold states he “previously asked the firm’s public relations expert, who seems to be more inclined to spam than research, to cease sending me meaningless spammy news releases. My request was ignored.”

“Bad, bad, bad public-relations expert,” I say. “Get back to your research. They need your help in the labs.”

Everything copacetic now? Well, not quite. There’s more to Arnold’s complaint. His complaint’s true core is not spam PR. It is here:

“What fascinated me is that Temis asked me to facilitate an introduction for them to a $1.2 billion company’s president. I did this and moved on. I assumed in the manner of French cultural norms that I would be rewarded with entrecote. Wrong. My reward has been spam.”

It’s about money. Temis didn’t pay Stephen Arnold for freely offered advice. Temis didn’t pay Stephen Arnold for freely offered advice so Temis gets slammed, for supposed spam, for claims Arnold questions, for being French. Yes, for being French. Arnold writes,

“I quite like Expert System SA in Bologna, Italy, and Bitext in Madrid, Spain. Great food, interesting culture, and -– nota bene –- no spam. One has to get the semantics correct. No spam from Italy. No spam from Spain. Hmmm. There’s a cultural message perhaps?”

There is some, slim attempt at substance in Arnold’s article. Arnold questions Temis’s claim to being “the leading provider of Semantic Content Enrichment solutions for the Enterprise.” Then he writes, “Leading? Semantic content enrichment. What’s that?” and asks, “What about outfits like Access Innovations, Concept Searching, Expert System SA, Smartlogic, and more than 75 other firms in the semantic space.”

Of course, it’s legit to question, but where’s the logic in offering specific other providers as contenders when you claim not to know what they’re contending for?

In case you doubted: Arnold’s list of contenders shows that he’s not completely clueless about “semantic content enrichment.” That said, of the companies he listed, only Expert System is squarely in that biz.

Myself, I describe the term functionally in a sponsored newsletter article I wrote for OpenText (a former consulting client): “Semantic content enrichment adds value to online information by tagging topics and providing context-sensitive links.” Content enrichment is a significant capability for digital-publishing platforms, for instance, MarkLogic’s. MarkLogic’s Open Enrichment Framework came out in 2008, integrating text-analysis products from companies that include both Temis and Access Innovations.

I (and OpenText, MarkLogic, and Temis) don’t stand alone is seeing enterprise value in semantic content enrichment. I’ll further point you to Barry Graubart, who cites a number of user organizations in a blog entry about a Software & Information Industry Association seminar last June (which I attended),
Semantic Technology Driving Real Revenue for Publishers.

In the end, my only quibble, as an industry observer, with Temis’s leadership claim is that I see the company as “a” rather than “the” leading semantic-content-enrichment provider. My issue with Stephen Arnold’s Temis slur, while I don’t have a dog in the fight, is far more significant. I find Arnold’s vindictive payback quite distasteful and even destructive. It calls out for a response. I hope I’ve adequately answered the challenge.

I know of no open-source (software) tools dedicated to sentiment analysis. Instead, a variety of open-source text-analytics tools — natural-language processing for information extraction and classification — can be applied for sentiment analysis. They include –

- Python NLTK (Natural Language Toolkit), http://www.nltk.org/, but see also http://text-processing.com/demo/sentiment/

- R, TM (text mining) module, http://cran.r-project.org/web/packages/tm/index.html, including tm.plugin.sentiment.

- RapidMiner, http://rapid-i.com/content/view/184/196/.

- GATE, the General Architecture for Text Engineering, http://gate.ac.uk/sentiment/.

I’m sure you can also find UIMA-plug-in annotators for sentiment — Apache UIMA is the Unstructured Information Management Architecture, http://uima.apache.org/ — also sentiment classifiers for the WEKA data-mining workbench, http://www.cs.waikato.ac.nz/ml/weka/. See http://www.unal.edu.co/diracad/einternacional/Weka.pdf for one example.

I bet someone’s doing sentiment with the Stanford NLP tools, http://www-nlp.stanford.edu/software/, although my understanding is the maximum-entropy classification isn’t the best approach for sentiment. I’m no scientist so I won’t go into this.

Then there’s LingPipe, which can be characterized as pseudo-open source. See http://alias-i.com/lingpipe/demos/tutorial/sentiment/read-me.html.

Powerful, I can’t say. Where machine learning is involved, a lot will depend on your training set.

Note that the tools above work on textual sources. There may be open-source tools out there for information extraction from non-textual, sentiment-bearing sources such as speech (with the outputs fed into a classification engine such as some fo the above), but I haven’t looked into them. If you know of any, or have additions for my list above, please send me a note (grimes(at)altaplana.com).

Want to catch up… or stay ahead? Check out the Sentiment Analysis Symposium, May 8, 2012 in New York; also the May 7 Practical Sentiment Analysis tutorial, to be presented by Prof. Bing Liu.

But wait. Wasn’t 2011 — weren’t 2010 and a few years before that — for those in the know? I think so, and I think 2012 will keep up the pace, seeing the text technologies and solutions adopted, directly and indirectly (embedded in applications) by significantly more users than ever before.

The question originated with my friend Tom Anderson, who collected and published responses from a variety of industry figures to his Next Generation Market Research blog. I was late in answering myself. Tom will add my response, but I’ll also post directly here:

The easiest prediction to make with confidence about 2012 text analytics is continued strong market growth — my estimate is 25% on a base that likely topped $1 billion globally in 2011 — as uptake expands throughout the enterprise and as the technology becomes a must-have value-booster for broad-market survey, social/media analytics, and CRM platforms.

With less certainty: We may look back on 2012 as the Year of Question Answering, of the deployment IBM Watson/Apple Siri-type technologies to respond to enterprise and consumer information-access needs ranging from customer (self-)service to medical diagnosis, as a semanticized replacement for tired old search.

And there are signs, from market leaders such as SAP and IBM and from innovative start-ups alike, that 2012 will be the year of effective data fusion across database and text (a.k.a. “unstructured”) sources. Business can’t, won’t, wait for prescriptivist, rigid Semantic Web approaches but is instead applying analytics to the job, to discover the connections that make for truly rich data. You need analytics to operate in real time, to keep up with the data torrent. Many of those efforts will incorporate information mined from audio (speech), image, and video sources as a evolution from text analytics to content analytics picks up speed.

Check with again a year from now and we’ll see how 2012 panned out!

How does this bit of theory play out in practice, among people engaged in real-world customer relationship management (CRM), marketing and market research, and business intelligence (BI)?  I polled three industry figures to find out: Banafsheh Ghassemi, VP, Marketing – eCRM & Customer Experience (CE) at The American Red Cross; Marshall Toplansky, president of “mass opinion business intelligence” vendor WiseWindow; and Next-Generation Market Research guru Tom Anderson, who heads Anderson Analytics. I posed them three questions, exploring the path to enterprise-scale sentiment analysis.

My  first question gets at a basic question, essentially, is sentiment analysis worthwhile? The responses address ways sentiment analysis complements established methods and channels, in time frame and as a cross-check.

Seth> What has your sentiment-analysis experience been like?  Have you gained new customer or market insights, and have you been able to do anything new, anything you couldn’t have done without sentiment analysis?

Marshall> “The big revelation to us has been the volatile nature of both consumer sentiment and the business metrics they are indicators of.  When you use traditional marketing research to understand sentiment, you are dealing with long time frames.  Research does a good job of identifying long-term trends.  But, people are living increasingly in the short-term.  They capitalize on market moves quickly and have a set of short-term tactics.  Mass consumer sentiment from online sources, unlike marketing research, has been able to identify these tactics and measure results in the kind of fast timeframes contemporary businesses require.  You could never have even designed a survey in the time it takes to get in and out of moves indicated by sentiment.”

Banafsheh> “[Let me tell you about our] in-kind donation scenario. In recent years, due to harder economic times, people have shown more interest in contributing through unsolicited in-kind donations rather than money.  Due to cost structures associated with storage, transportation, and delivery of such contributions we are unable to accept such donations.  We have seen some negative emotional reaction to this on our social networks.  However, when we look for context in other channels where the same interest is voiced, such as calls to our public inquiry line, we have seen that we could do better to proactively communicate and educate the public in why we don’t accept these donations.  The proactive awareness could very well minimize, if not eliminate the negative sentiments that come with a preconceived expectation.”

Tom> ”There’s no doubt that sentiment analysis has been useful on several projects and that we have gained market insights thanks to being able to segment and code data based on various sentiment approaches. Overall usefulness really depends on the data though, which is different on a a case by case basis.

 ”Most of our data contains a lot of fields other than unstructured data, so we’ve been lucky to have a lot of options from the beginning. Sentiment becomes a lot more important when you are looking at data that is less rich, like Twitter which more or less is just 140 characters with a data stamp.”

On to Q2, how sentiment relates to other data elements. My one-sentence summary of the three answers is, There’s clear correlation, but you don’t want to make too much of it. But read the responses for yourself.

Seth> How are your organization and clients matching online or social sentiment or enterprise feedback with information from other data sources?

Banafsheh> “I wouldn’t say we ‘match’ the information, but look for context in our other more structured (e.g., CRM) and richer data sources (e.g., email which is extremely rich). [The in-kind donation scenario illustrates this.]

Marshall> “We have seen a good deal of client interest in correlating online consumer sentiment to a number of important business metrics.  For instance, in one case, correlation analysis found that changes in online sentiment relating to product problems was found to be a leading indicator of call center volume.  In another, changes in weekly sentiment related to product quality were found to lead weekly stock prices.  And, in another, sentiment related to a leading musical group was a strong indicator or changes in sales of music.

“To our minds, this is not surprising.  It seems obvious to us (with plenty of hindsight) that when you have hundreds of thousands of people expressing their preferences, their actions will tend to follow.”

Tom> “It depends on the project, there have been many so we’ve probably done just about everything at least once. We’ve merged LinkedIn data with survey data, CRM data with social media and survey data, and call center data with matching operator notes with field technicians notes, to name just a few.

“I think though that there is a misconception out there among some that it’s a good idea to pipe all text data into one source and make comparisons across data. Having worked with quantitative data for many years I can tell you that often times those cross comparisons sound better in theory than in reality. I think it’s far more important to think about the various resources you have, identify the most important ones and look at them individually. After that is done, you will have a better understanding of what can be gained by merging the data.”

My last question was intended to be practical and forward looking –

Seth> Do you have any guidance for folks who are new to sentiment analysis or who have been using sentiment technologies only for social-media analysis?

Banafsheh> “First, I would say using sentiment technologies with social data is valuable in any Voice of the Customer (VOC) program toolkit to the extent that it is not the only data source used. Otherwise, the insights produced will be very narrow and limited insights as it would be if any one of the other data source was used. If you examine each one of your touchpoints and channels where free format VOC is captured, such as surveys, call-center notes, customer e-mails, etc., in isolation, you will very likely find sentiment patterns in each that is more slanted in one direction than another.  For example, you may find your surveys show more positive sentiment, your letters may show more neutral, and your Twitter feed more negative, and so forth.  So if you only took one source you will draw incomplete conclusions.

“Second, these tools are still fairly challenged when it comes to social ‘parlance,’ the abbreviations, excessive snark, emoticons and so forth.  So be aware of those limitations and recognize that the outputs will still require human intervention and validation of the results.  Recently we compared coding using SPSS’ text analytics with the coding manually provided by Red Cross volunteers and found poor agreement. The text analysis software coded 26% of the positive comments as positive. The software was unable to assign a sentiment to more than half.  The 21% of positive comments coded as negative by the SPSS software were generally related to the use of words describing the seriousness of the hurricane or the extent of the damage.  Examples ‘hit very hard by Irene,’ ‘Many blood drives were forced to close,’ or ‘Instead of whining about the dud, how about joining the Red Cross.’ There was modestly more success with negative comments, where 53% were coded correctly by the software.”

Marshall> “My advice is about WHO should be adopting the use of sentiment technologies.  Give it to the people that run real metrics in the company.  If you let market research people control this tool, you will move too slowly.  If you give it to the social media engagement group, you will get no strategic value from it.  Give it to the operations people, who have to create better forecasts and design real-time key performance indicators for the business.  This is where the real strategic and tactical value lies for sentiment analysis.”

Tom> “In regard to sentiment generally, the tendency is to compare machine coded sentiment to human coding. The problem is most of us are so far removed from human coding that we forget just how inaccurate it can be. We also seem to forget that a lot of data really is neither negative nor positive.

“Personally I don’t like to compare human coding to machine coding, we find that we can do so much more with machine coded data. If companies are looking at text analytics and sentiment analysis as just a way to cut or eliminate human coding costs, they’re not understanding the true benefits.

I’ll split out part of Tom Anderson’s answer, regarding social media analysis, which characterizes as a “pet peeve.” Tom’s thoughts on this particular topic –

Tom> “Everyone is talking about social media analysis. In reality though 99% of what’s being called social media analysis is just Twitter or Twitter plus blog data. This represents only about 10% of the population, and for those of us who blog or are on Twitter, we know just how ‘special’ this population and messages we propagate are. A lot of it is definitely rather promotional in nature.

“I’m not saying Twitter and blog analysis has no value. It’s good to understand what drives online buzz and consequently some of the Search Engine Optimization (SEO) efforts. However, until Facebook tears down their walled garden (I think it will happen soon), we’re not seeing what most regular people are saying. Until then more focused research among brand enthusiasts on special discussion boards will probably remain most useful. Sadly, relatively little of this is actually being done. Those projects we’ve done in that area have been rather successful, and sentiment analysis certainly was a critical component in all of them.”

I had posed my questions to Marshall Toplansky, Banafsheh Ghassemi, and Tom Anderson in connection to the November 9, 2011 Sentiment Analysis Symposium. Banafsheh was a panelist, Marshall gave a lightning talk and his company was a sponsor, and Tom would’ve attended if he hadn’t had a schedule conflict. The conference went really well. See for yourself: Videos are online at sentimentsymposium.com/SS2011w/presentations.html. And please do revisit the symposium site in mid-January, when information on our May symposium, the 3rd New York symposium, should be online.