Data Mining: Text Mining, Visualization and Social Media

Speech Interfaces to Search

In  Yahoo accused of poaching speech engineers CNet says:

  A California court has temporarily barred about a dozen engineers hired by Yahoo from working on interactive speech technology at the search engine after the company they recently left filed a lawsuit accusing Yahoo of trying to steal trade secrets.

A Santa Clara County Superior Court order issued on Wednesday said speech technology company Nuance Communications was likely to prevail on the merits of the case regarding misappropriation of trade secrets and unfair competition, among other claims. The order also said Nuance would suffer "great and irreparable injury" if the employees were not blocked from doing work on speech technology at Yahoo.

Maybe Barney was right about talking to our search engine tomorrow.

 

Searching for Conversational User Interfaces

Barney writes that:

I agree with Ron Kaplan's view at last week's Accelerating Change Conference that Conversational User Interfaces will come of age in over the next 5-10 years.

I've recently been doing a literature survey on models of agreement and disagreement in dialogue and found very little out there other than this paper which is close, but not quite what I am after (even when extending the survey off net and speaking to linguistitians). This page on Rhetorical Structure Theory (RST) is a good place to start, as is this page on Discourse Modeling, but I'm looking for research that focuses on the act of agreement/disagreement, and light-weight processes for detecting cues in text. For example, imagine looking for agreement in email exchanges.

It turns out that searches for linguistic issues are especially hard. The term 'agreement' has a number of special meanings in linguistics (agreement of features, agreement of labellers in labeling tasks). MSN search does a better job here than Google in surfacing the notion of speaker agreement in the query 'theories of discourse agreement.' But lo, Clusty aced this problem. A query on 'discourse model agreement' produced a cluster (the first) for which the first result (via citeseer) was:

Dialog Act Modeling for Conversational Speech (1998)  
   Andreas Stolcke, Elizabeth Shriberg, Rebecca Bates, Noah Coccaro, Daniel Jurafsky, Rachel Martin, et al.

with the abstract:

We describe an integrated approach for statistical modeling of discourse structure for natural conversational speech. Our model is based on 42 `dialog acts' (e.g., Statement, Question, Backchannel, Agreement, Disagreement, Apology), which were hand-labeled in 1155 conversations from the Switchboard corpus of spontaneous human-to-human telephone speech.

Ok - so it is not about agreement/disagreement in text, but it does look promising.

For these types of non trivial search, the generic solutions of Google, et al are simply not working. There is a long tail of search problems that mine above, but they seem to have little impact on the product direction of search companies. I wonder if there is a relationship between the complexity of a query and the likelihood of click-through on ads which means that the optimal point for Google is closer to serving very simple searches rather than really doing a good job at hard search problems.

On a side note, Google also does poorly with 'NLP' which it believes is largely about Neuro Linguistic Programming, not Natural Language Processing. There may well be a lot more pages on the former, but  a search for 'Michael Jordan' shows that Google does attempt to provide results on the first page for various meanings of the search phrase - perhaps they haven't figured out how to do this for abbreviations and acronyms.

I need to look into Rollyo...

Ping Crisis

Kevin Burton suggested that I fix my weblogs.com ping graph so that it was actually readable. Ping servers have been a key feature of the blogosphere, but there is something fundamentally worrying about them. As comprehensive coverage is a key battleground for blog search engines and portals, how can the currently free and (almost) open services offered by weblogs.com, blo.gs, etc. resist greater and greater claims on their value? Secondly, as this graph of weblogs.com suggests, as the number of pings grows, how can underfunded services like weblogs.com be considered reliable?

Edelman gets some negative buzz, but NoingNoing doesn't get it.

NoingNoing (a blog I am unfamiliar with, but which gets reasonable readership) posted some pretty negative comments about the Edelman/Technorati survey:

In an email sent out to Technorati subscribers this morning, bloggers were offered the "opportunity" to fill out Edelman's survey. Why?  So Edelman can get a better idea of how they can persuade bloggers to dance to their tune (we're not suggesting Technorati are going to try and blog content).

It's no secret that PR people are essentially paid to lie, embellish, and filter out any information that goes against their client's interests.  So the uncontrollably democratic nature of the blogosphere poses a problem for the PR industry.  They're left swinging in the wind when millions of people are speaking their minds online without including the official line as dictated by a PR professional.

One has to view the blogosphere, and the ability to observe it and mine content from it as an information system. Imagine two extremes. In the first, I don't know what you want, so all I can do is to persuade you to buy my stuff. In the second, I know exactly what you want, so I can create a perfect product for you.

That is the proposition. If you believe it, then you can see any attempt to understand the information system as a good thing (would you like a better computer to update your blog from?) Remember, we are in the first five minutes of everything here, so attempts at figuring out what is going on are bound to seem problematic.

Full disclosure: I work for Intelliseek which has previously published with Edelman - but read this blog to get my full story.

Technorati/Edelman Survey

Technorati and Edelman have created an online survey:

in an attempt to better understand how blogging and traditional PR intersect, and what bloggers think about communication from mainstream companies.

(according to Niall Kennedy, Edelman). The results will be out in a month - should make for good reading.

I hope that they include some discussion of the methodology. I'm particularly interested in the issue of self-selection: how the results are biased to profiling the type of people who elect to take such surveys. One of the promises of measuring communication and publication in the OPM (online personal media) space is the reduction of this problem. Evaluations and white papers are being thrown out at a considerable pace with, I'm afraid, a worrying lack of sound methodology.

The white paper will get a lot of attention from bloggers, just as comScore's Behaviors of the Blogosphere did. The reaction to that publication was not entirely positive due to a lack of methodology and ultimately some very dodgy looking results.

(It is a shame that both Technorati's and Edelman's blogs don't support trackbacks.)

Battle of Movie WOM

After Serenity bloggers were forced to include a link to the movie site in order to see a free screening of the movie, the movie site became the top link in today's BlogPulse rankings. I still think that their strategy is less than perfect - Jackson's web based diaries of King Kong production are far more engaging and will, I believe, ultimately be more effective. Serenity's approach doesn't understand the concept of a relationship with bloggers. That being said, have a look at these graphs:

Update: And here is the relevant alexa chart:

As the Kong site is about production, it is not surprising to see it peak during production. I'm guessing that in the run up to the movie the Kong is King site will start to get more buzz and pick back up, along with the more traditional movie site which hosts the trailer and 'official' stuff.

 

Vertical Search

Barney posts a summary of a panel he moderated on vertical search. One of the main takeaways is that vertical search has come to mean specialized search. One thing I would add to his comments is that due to that specialization, the nature of search changes (or can change) dramatically. Intelliseek, for example, provides analysis over online personal media (boards, blogs, etc.) Our 'searches' are extractions of brands and products and our results are anything from documents (e.g. board posts), time series, etc. to measurements of the polarity of users statements (favourable or unfavourable with respect to some brand or product or feature).

Existing interfaces use the simple command line (text field) to capture the query and then iterate passively with the user over sessions of results and further queries to get to what is needed. Specialised domains can start off with radically different interfaces - e.g. a map of the Caribbean or a distribution of house prices displayed geographically. In addition, they can be far more explicit about iterating with the user to get to the desired results space.

Serenity Blogging? No thanks.

Gclectic pointed me to the free screening of Serenity (Firefly, canceled, blah blah,...). I registered then got this in email:

Congratulations! You are one of the lucky bloggers to be chosen and confirmed for the screening of SERENITY for the time, date and the number of guests that you have requested. Please note, this confirmation DOES NOT guarantee you a seat at the screening.

To significantly increase your chances of getting into the screening, you MUST do the following:

· You MUST include the film’s synopsis on your blog (synopsis below) and you MUST link your blog to the SERENITY website (which has the trailer and production notes) http://www.serenitymovie.com and featured artwork. After you have screened the film, please discuss it on your blog. Please provide us the links to all of your blog posts on SERENITY at serenity@gracehillmedia.com

· Print out and bring a copy of this confirmation.

· Arrive at the theater AT LEAST 45 minutes before the show begins.

· Upon arrival at the theater, please find a UNIVERSAL PICTURES representative and inform him or her that you are part of the SERENITY BLOGGER BONANZA. The Universal rep will then instruct you as to what to do next.

· DO NOT bring in a camera or a cellular phone that takes pictures. They WILL be confiscated, and you will NOT be allowed into the screening.

· Have a great time!

Great - so I have to show up 45 mins ahead (not likely) and I might not even get in? All those links that bloggers 'must' provide - is that like making a link farm out of bloggers in exchange for tickets? It's like they almost got it about engaging bloggers to generate word of mouth, but just messed it up at the last mile.

I will have a great time though, thanks to Netflix, Sony Wega and my Ikea sofa.

BTW, Gclectic, I was excited when I heard of this from you, but after the registration demands, the feeling passed.

I'm amused that my quoting the email resulted in a link, but I'll leave it in there anyway...

Farewell Jeeves

RIP.

Search site Ask Jeeves is getting rid of the iconic valet that has been its companion since its earliest days.

Citing "user confusion" over what the butler character represents the search site has said that Jeeves will soon be phased out.

Via BBC.

Ping Volume

The graph below shows the volume of pings received by pingomatic. It is interesting to compare this with the trend I posted on earlier for weblogs.com. These two together suggest that there is an upwards trend for pings, but that weblogs.com is failing to capture it (assuming a lot of generalities about the pingers).