1.06.2007

"sentiment analysis" search on "answers" service, Y! Answers, Live QnA, LinkedIn, Yedda and Amazon's Askville

Answer services are pretty interesting when it comes to finding a high level definition of "sentiment analysis". Answer services are sort of a hybrid between algo search and human-run directories with a different user interaction model. You can ask plain English questions and get plain English answers. Answer services touch on what some have called Social Search, pigging backing on social networking trends, although I am still looking for a good definition of Social Search. What I really like is that answers (a.k.a. results) come from people. It's somewhat along the path of Wikipedia, although the set of info retrieved is made of more diverse answers rather than one by consensus.Answering services do better for sophisticated queries in the form of plain English questions than do 2.1 keywords typed in a traditional search box, even Google's. More importantly the information retrieved is of a different format. Most volunteering to answer these questions do care (or is it the miles?), include subjective in nature opinions and often provide "tone" clues turning Answers databases into wonderful data sets to mine, extract and make sense of "sentiments". Yahoo! has pushed theirs very aggressively, Microsoft Live's QnA was introduced in beta recently. Google recently discontinued theirs. Cannot ask about "sentiment analysis" to Ask although answering plain English questions was really Ask'spop'ing up: Wondir, Yedda, LinkedIn.
vision in 1995 ... just too early to work, maybe. More are now
Overall, LinkedIn did best, by a long shot. Makes sense, my network is made up of people that i) care about Sentiment Analysis ... if anybody does, and ii) care about me ... say, just a tinny biot more than a total stranger. Answers are as good as the people you ask the questions from, right? Well, LinkedIn is all about people you know and just released an Answer service with a really nice, intuitive, minimalist and simple Answers results page. Right on, don’t overdo it. I particularly like to see the names of the contributors in bubbles on the left of the comments.

Hitwise' LeeAnn Prescott reports in an interesting article by Jason Lee Miller that Yahoo!'s got all the answers, well, 96% of them anyways. Y! pushed Answers pretty hard with commercials and more over the past months. I asked Yahoo! Answers what was sentiment analysis, buzz monitoring about 2 weeks ago. the question is now closed. I got 1 answer, not a very good one but thank you anyways. That question closed since. I asked again who knows the most about sentiment analysis, buzz monitoring, reputation management about 16 hours ago and I have not received an answer yet. I have had good experiences before for other questions, though. It really comes down to reaching the masses and creating that Network effect. eBay succeeded, I don't think there will be many Q&A services at the end of the day, or vertical niche ones.

Microsoft Live QnA just shipped in beta recently I believe. It will take them a while to ramp up to reach critical masses of questions, answers, and users for both. Microsoft's got decades of linguistic R&D, though, and could very well catch up by providing better more relevant answers. I asked again what was sentiment analysis and Internet buzz monitoring? I got one spam answer. I still remember to this day an internal R&D presentation about speech-to-text in 1996 when I was doing product planning for Microsoft Works in the Office division. One of the main researcher was talking about how they got 98% there, but the remaining 2% made it still impossible to use in everyday applications. It certainly still feels that way when I have to call American Airlines' automated system to change a flight. It drives me crazy! Then you could say, maybe the French accent's got something to do with that, right? Anyways, I asked Live QnA again about 16 hours ago, and got 3 spam answers. Yahoo!'s doing a better job at filtering spam.

Yedda was not bad, got a couple answers, including one describing IBM's definition of sentiment analysis. The UI is a bit over monetized, but hey, none of them are philanthropic organizations, right?

Amazon's Askville just launched. Frankly, I am scared of Amazon knowing everything about what I read and the questions cycling in my brain ... could turn into some privacy fears. That's true of anybody online actually. Diversify! This said within a couple hours of asking the same question, newbie203175 a very good answer for the 3 terms:

  1. "Sentiment analysis involves classifying text based on its sentiment. Sentiment analysis seeks to identify the viewpoint(s) underlying a text span; an example application is classifying a movie review as "thumbs up" or "thumbs down". [...]
  2. The best buzz-monitoring is a tracking tool a web-developer might be willing to use foranalyzing and monitoring his/her recent web-projects. In order to analyze the popularity of a page regularly, you have to have some statistical data and values [...]
  3. Reputation management is the process of tracking an entity's actions and other entities' opinions about those actions; reporting on those actions and opinions; and reacting to that report creating a feedback loop. All entities involved are generally people, but that need not always [...]"
And finally, I also tried Cha Cha search, interesting model where human beings actually do the searching for you and return the results as is. The answers weren't bad but I still have to do all the work of extracting the info myself.

Sphere: Related Content

Add to Google Add to My AOL

1.03.2007

How are Google, Yahoo!, MSN Live, and Ask doing for a "sentiment analysis" search?

Remember, I am on a quest to find a good high level definition of "sentiment analysis", nothing too granular, and definitely related to Internet buzz monitoring, reputation management. Overall, results are pretty good doing a quick comparative analysis across the main algorithmic engines. Google and Yahoo! provide the best results. yahoo!'s 404 SL is annoying but the Answers and Related searches integration makes up for it. Ask comes next, actually doing better than MSN Live. Keep in mind that "relevance" is very much a personal measure of perception that goes beyond results #1 to 10. Both advertising and organic results are somewhat dynamic, so don't send me hate mail telling me you are not getting the same results. At the time I did these searches, about 15 minutes ago, these are the results I experienced.

  • Google's got 7 relevant organic results and 1 relevant sponsored link. the fastest experience, in & out.
  • Yahoo!'s got 7 relevant organic results as well and 2 relevant sponsored links, although one is dead, 404. In reality, when users click on it and get a 404, they mentally penalize the engine, so it's worst than not having that link of course. Yahoo! is also suggesting some related searches at the top of the SERP, but I am not impressed. This said, Y! Answers is pretty well integrated to the Web search results page.
  • Microsoft Live's got 3 relevant organic results and no relevant sponsored link what-so-ever. I guess it will take them a bit to catch up with Google and Yahoo!'s SL coverage. Live also integrates QnA although not allocating as much real estate as Yahoo! is.
  • Ask's got 5 relevant organic results and 1 relevant sponsored link.

Can you find a better results page?

Sphere: Related Content

Add to Google Add to My AOL

"sentiment analysis" on Microsoft Live Search

A search for sentiment analysis on MSN Live serves some interesting results as well. TrendIQ is at the top, definitely well indexed with most engines. The second result is not as relevant, but the third is, the paper from Cornel discussing professors Claire Cardie and Lillian Lee, "working on sentiment-analysis technologies for extracting and summarizing opinions from unstructured human-authored documents. They envision systems that (a) find reviews, editorials, and other expressions of opinion on the Web and (b) create condensed versions of the material or graphical summaries of the overall consensus." MSN Lives's last result for [sentiment analysis] is a full deck, "Sentiment Analysis of Blogs" by Paula Chesley from the Linguistic Department of University at Buffalo. Pretty nice, covering definitions, examples, mining and summarizing architecture, and a whole lot more about classification approaches.

Sphere: Related Content

Add to Google Add to My AOL

"sentiment analysis" search on Yahoo! Search

First, Yahoo! returns different sets of results whether you "quote" the query "sentiment analysis" or not; the set of Sponsored Links changes because no advertiser seems to be targeting the exact phrase.

At first sight, most of the organic results, sponsored links, "Also try" related searches features, ... relate to the financial markets' sentiments. Nice, related to "sentiment analysis" but not what I am looking for. You're probably wondering how could Yahoo! or anybody else know? Well, I have been doing searches on the topic for a while, repeatedly. Can't the engines learn from my behavior? Not yet I guess.

The 3rd. sponsored link - Biz360 - seems interesting, a lead-generation landing page offering a white paper download titled "Use Competitive Intelligence Media Monitoring to Improve Your PR Positioning". Not about Internet sentiment analysis in the context of smart linguistic-based data mining, though. Biz360 is more like a news clipping service from briefly browsing the Web site.

The 4th. sponsored link URL navigates to a 404 page. I have been seeing less and less of these, though, since Greg Notess started tracking 404 rates by the major engines in the 90's. The company - TrendIQ - is a good result, though. TrendIQ's home page does mention doing sentiment analysis studies. Interesting positioning : Discovering Trends Hidden in the Internet. I'll have to get back to it and learn more. TrendIQ is then listed twice in organic results, linking to deeper, more interesting pages, including "TrendIQ Sentiment Analysis" with a case study covering the 2004 elections in the United States and another.

The 1st organic result is not bad, links to the Wikipedia's "Sentiment" page, where you can disambiguate your query, including branching out to "sentiment analysis: automatic detection of opinions embodied in text or speech". Problem is that Wikipedia is pretty light on the topic, still. On top of that, the actual "sentiment analysis" page is also listed a bit lower. That's a bit Wikipedia overkill. Yahoo! has a couple of other results about Lillian Lee's paper "A Matter of Opinion: Sentiment Analysis and Business Intelligence".

The most interesting result is probably a white paper, Recognizing Contextual Polarity in Phrase-Level Sentiment Analysis, by Theresa Wilson, Janyce Wiebe from the Department of Computer Science at the University of Pittsburgh and Paul Hoffmann. Will have to read that as well in more details. Abstract: "This paper presents a new approach to phrase-level sentiment analysis that first determines whether an expression is neutral or polar and then disambiguates the polarity of the polar expressions. With this approach, the system is able to automatically identify the contextual polarity for a large subset of sentiment expressions, achieving results that are significantly better than baseline."

Overall, not bad, but far from really good results. Better up-front disambiguation features would have helped filtering out financial topics and focus on data mining -type Sentiment Analysis results.

Sphere: Related Content

Add to Google Add to My AOL