However, there’s an interesting tension between strategies involved with presenting information to users in this fashion. There is the TF/IDF and “PageRank” strategies that Google uses to filter and order (resp.) the relevant pages for each query. These are essentially two different information “spaces” that are used to relate documents to a query. TF/IDF essentially helps to match a set of documents that are “semantically close” to a query, while PageRank helps to identify pages that are close to the “center” of a network of linked pages (Hypothetically, these “central” pages have higher quality information).

In a sense, PageRank is a sort of filter or lens that you can’t turn off. You can’t really tell how much PageRank is affecting the order of items presented, and you also can’t tell if Google has “enhanced” a query by tailoring its results to a more “useful” set of documents.

In fact, most people probably don’t even spend much time following the results exactly as they are arranged. Instead, they quickly scan the list of titles and summaries looking for more keywords that suggest the kind of content they are interested in. In this sense, Google doesn’t even need to provide a perfectly valid semantic or document-link space as long as the user gets something useful in the top 10 results. This method of search is really more about reading, and is therefore a literate form of search.

However, this isn’t typically the way that we find things in the real world. Generally, we do so by using our senses to size up our environment. Is there a good smell in the air? Is there a group of people moving a certain direction? Maybe there’s something interesting that direction? All of these strategies involve notions of proximity/approximation, social awareness, and direction (resp.).

The strategies involve not knowing explicitly what you want or where you want to go, but in manipulating your environment gradually until you identify a source or local optimization of the resource in the surroundings. In order to maximize your efficiency in this environment, you want to have control over how you can move through these spaces.

In the Google example, this may mean going “against” the PageRank position, or only operating in a certain “radius” from the center of the network of document links. Perhaps you don’t want to find something “too popular”, or are looking for something new. Maybe you want to find other complimentary topics or content for the document space you’re already familiar with?

Many of these strategies involve more control over the information spaces than the search portals provide. In addition, they also involve the use of human generated meta-data to a greater extent. Sometimes you want to follow the herd, but sometimes it makes sense not to.

I’ve been interested in how to handle this scenario with a recent paper on “Visualizing Social Links in Exploratory Search” (here’s my author’s copy.).

One of the problems I had was how to overcome the familiarity and reliance on literate search strategies in a spatial interface. In general, if I use the two dimensions of an applet to display document candidates, it often becomes impossible to layout the “snippets” of text necessary to describe the documents in greater detail. We handled this situation by essentially removing these snippets from both the simple “list” interface and the “map”-like interface. The interfaces using maps fared slightly better than the list in this study, but none of them were particularly well liked due to the absence of snippet text.

So… the question is, how do we overcome or accommodate literate search expectations in a spatial environment? Tag clouds? Force directed snippet placement? Abbreviations? It seems like there should be a way to do it.

I’m here in London for the IV08 Information Visualisation conference where I’m presenting a paper titled “Candidate Mapping: Finding Your Place Amongst the Candidates.” This is the product of an independent study I did with Katy Börner. You can grab my author’s copy here, and can see my presentation slides here.

The main thing that people had issue with in my talk was the use of metric analyses with (initially) non-metric data. The argument against using non-metric analysis is that in many cases you can’t compare “apples to oranges.” In my case, I had to define a distance metric between candidates based on the similarity of several issue stances. These issue stances could be simple support/opposes positions, or something “in between” these positions. For instance, some candidates supported same sex marriage, but indicated that they wanted the states to decide for themselves. This is a “weaker” position since it doesn’t indicate the strongest possible endorsement.

The difference in stance similarity is also coupled with issue similarity issues. Are certain issues more important? Should certain issues be included or excluded? These are all questions that have to be answered before any sort of general similarity metric can be created.

So, I followed a set of assumptions. First of all, I assume that all issues had the same “weight”. This is violated if someone doesn’t think that two issues have equal importance (and this happens often, naturally). However, the instructions page indicated that an individual had to provide their response based on how they felt they matched the candidates on all of the issues.

The metric I chose wasn’t based on how much the candidate fit the individual on the “most important issues” to the participant, but based more on how much the candidate and participant lined up with their responses to all the issues.

Furthermore, gradients in the strength of a position are ordinally coded (details in the paper). These follow the Gower Similarity metric, which essentially subtracts the ordinal position of the responses, takes the absolute value, and divides that by the range. This gives each stance a “strength” from 0 to 1. If two candidates differ on an issue, the difference is determined by difference in their gower similarity scores.  These differences can become smaller for issues with more possible responses.

This process essentially imposes a metric space on the non-metric information through the use of Gower Similarity equivalence classes. Unfortunately, the Gower similarity space is still incomplete. This is due to a few candidates neglecting to hold an opinion on some of the issues. This is handled through the use of Sammon Mapping. Sammon mapping seeks to reduce an error function based on the similarity of the entities.

What ends up happening is that candidates are “dumped” into a low dimensional space, and then “moved around” until they preserve the similarities they share with other candidates as best as possible.

For the most part, only a few of the candidates were missing many responsesn, such as Mike Gravel. In total, only 16 out of the 250 issue stances (25 candidates @ 10 issues each) were missing. So, the candidates still had a large amount of relationship information, and this amount of information was enough to ensure that the sammon mapping generated the same lower dimensional configuration under any configuration or order.

If the participant indicates his position according to the layout of map (which is generated specially for each participant and already includes his “actual position”), then the difference in position reflects a difference in similarity between the two positions, and not between every other candidate (as a dissimilarity). The distance between positions (selected by the user, and calculated by Sammon mapping) is not a true metric distance. However, since very few of the candidates are missing data, the Sammon mapping representation is nearly identical to a linear method using only candidates and issue stances with complete information. Therefore, distances as error are still informative. Other approaches for calculating error might involve calculating differences in stress (I’m continuing to look into this).

Finally, as a closing point, the final plot showing all the participants on the image at once is a composite of many different maps. Each of these maps will have slightly different layouts depending on how the participant answered (i.e. the position of the candidates will shift around slightly). However, the shifting is not enough to completely change the general distribution of candidates, and so I chose a single configuration for the candidates, and indicated their general position using a large font.

Finally, the one issue that did not account for with this technique is that some people may have a “centrality bias” with regards to interaction with interfaces, as well as with reporting the overall “centrality” of their opinions. In other words, people may be choosing central positions just to avoid being on the “fringe” for one reason or the other, even though that is in fact their calculated position. I think this is very interesting because in spatial (directional) voting theory (which I reference often) includes a “region of acceptability”( check here for a recent paper explaining some of the concepts) parameter that accommodates this same centrality bias in voters. I.E., even though they endorse a set of ideologies that identifies them as “fringe,” they will not endorse comparable “fringe” candidates. This is an interesting social phenomenon that I’ll have to continue to look out for.

I also had several people come up to me and ask about the paper (also, someone wanted to know more about my groovy laTex-Beamer setup, find more about that here, and then you can check out the source for my presentation (without pictures/plots) here.

Thanks to Joseph Cottam for discussions.

This next time around, I’m going to focus more on research/development issues, and I have quite a few of them. In general, the blog will now concern itself with:

1. Acoustic and social music analysis
2. Cognitive Science and Computational Neuroscience
3. Visualization and dimensionality reduction
4. Web development trends
5. Interaction methods for any of the above

That should be about it. It’s hard to get things going again, I don’t know where to begin.

I can’t help but think that Cookie Monster has the answer to all of our problems on sustainability.

“(There is) a need for hybrid people, who can put together a mean car and pimp it out, too. This is the holy grail of this new generation. Schools are changing slowly to adopt this model of education.”

– John Maeda

In other words, you have to be able to understand what the car can do, and what the people (really) want. I realize now that my whole life has been building up to: “MTV pimp my information system”.

Money and wealth in these games are pretty similar to money and wealth in the real world. You get more of it, and your sphere of influence is increased, both in what you want to accomplish in the game world, and in how people respond to you in the social world. What most companies do is artificially “limit” how wealthy or powerful you can be inside these games… this effectively ‘caps’ what a person can do, and arguably limits the economy. For some games, this is fine. Having a powerful or meaningful economy is not what they’re about… but that’s changing.

Several years ago EVE online launched. It’s a space based trade and warfare game with essentially little to no ‘caps’ on any of the economic or social organizational schemes. Furthermore, the entire player base is in one ‘world’, rather than a platform of separate worlds isolated on different servers. What this does is create the potential for massive growth, and for individuals to hold incredible power over others in the game. Soon enormous corporations were formed, and wars were fought over in game resources and facilities. Deals were made, friends were betrayed, and tables turned as the stakes in the game grew higher and higher. In fact, one of the most powerful methods to gain advantage in the game is in spying on people or engaging in social engineering outside of the game.

This article talks about one individual and his experiences doing this, and some of the observations he made. His most interesting observation has been on how the developers have been unfairly helping some of the people in the game. While one could see the spying and social engineering aspect of the game as par for the course in any market economy, the “deus ex machina” aspect of devs interfering directly with player affairs is almost like something out of Greek mythology.

While the infraction basically amounted to “insider trading”, the implication is interesting. Can game companies keep developers from interfering? What happens when the developers can make more (real money) by interfering in the game than they can through their day job? It’ll be interesting to see what kind of market strategies emerge when the actions of both “God and Men” must be considered.

What is unique, novel or innovative about this recommendation mapping technique?

First of all, it is important to realize the technique is comprised of two (arguably) novel components: The network visualization/embedding process, and the interaction process.
MDS as a general visualization technique for networks is not novel. However, it is important to realize the technique I’m working on does not use the “entire network” of network elements for visualization in a conventional sense, but rather a small subgraph. Conventional subgraph visualization techniques tend to treat the subgraph no differently than the global network. However, my technique handles the node weighting and embedding process much differently. It does so by weighting nodes based on their participation ratio, akin to the notion of entropy in the field of information theory. The resulting visualization emphasizes any significant local features of the subgraph, rather than showing features that are part of the global network (features consistent with hubs, etc.)

The “interaction feature” of my work is also fairly new, although perhaps not completely novel. Even though the embedding is different, this technique suffers the same “occlusion” problem prevalent in MDS. The way I handle this is through a method of dynamic node “repulsion” according to direct user interaction with a mouse or cursor. This method allows the user to spread out dense clusters of nodes without the use of an interaction modality (zooming, clicking/dragging, etc.). While this technique won’t work at any arbitrary density, it has proven useful for “nuisance” occlusion that occurs with sets of 2,000 or so network nodes. In general, I think that bringing both of these techniques together holds a lot of promise, especially in the field of music recommendation (my focus).

Why is the visualization more informative/useful to a user than a ranked list of results?

This is an argument I’m going to back up with a quantitative study of user behavior. However, the general idea is that these embeddings are a far more rich source of information than a simple list of items. The user will have a better understanding of higher level thematic contexts (clusters/shaped features) that exist in the results with this technique, and these features will be specific to the context that the user is investigating. This type of clustering information obviously cannot occur in lists, and is valuable both for identifying specifically relevant information, as well as determining parameters to include/exclude in subsequent searches. As a side note, the main problem here is that people seem to want “categorizations” to guide their search. I think this is an idea whose time has more or less passed, especially in the realm of exploratory search. For one thing, categorizations of novel and important content is often not present or possible. Categorizations are useful as condensed partitioned generalizations of content, but they can become obstacles or shibboleths to exploratory search. Music information retrieval suffers from this problem. Genres are the obvious categorical descriptors for music. However, an ignorance of the arbitrary terminology and labels of genres can impede an individual that is trying to find music they like. Furthermore, genre labels like “rock/pop” that most tracks are labeled with are virtually useless as categories, since they define such a broad array of music (try filtering on rock/pop using my recommender interface… most of the time it’s not that useful). In the end I’m arguing for a more free-form association method for the underlying data. This sort of data is easy to collect (analyzing sets of associated songs in playlists), arrange into essential data structures (networks of songs/people/artists, etc.), visualize (using many network viz tools), and arguably presents a clearer picture of how our society understands and responds to popular music as a cultural artifact. It’s important to think of these networks as aggregates of global listening behavior. As an example, consider another comment:

… Can you explain to me why “Snoop Dogg” is more like “Kayne West” than “Nelly” ? – they’re all Africa American rappers

The answer is: In this context and at this particular point in time, Kanye West is associated more strongly with Snoop Dogg than with Nelly on associated playlists and listening behavior.

Wouldn’t it be better to give users control over this rather than just munging everything together?

One of the things I’d like to add to these interfaces is more control over non-dimensional/non-relational pieces of information related to the individual tracks. This information includes the obvious features of track, artist, album, year of release, genre, bpm, etc. These would be pretty useful, especially since they would give folks some sort of “bearing” on the results using words that they are familiar with. However, searching/filtering by “genre/year/artist” is not necessarily a new concept, and therefore it hasn’t been a focus for me in the implementation of this interface.

However, I think it’s important to realize that the many of the separations and distinctions we make between songs are by and large arbitrary, both when it comes to how we associate songs on our playlists, and when we analyze and compare the underlying acoustic features of songs. Therefore, the notion of “control” over search/exploration using these features is tenuous at its core, since it can end up limiting information retrieval as well. Enforcing categorical descriptors in an exploratory tool is “hardwiring” them into its use pattern. What happens when these categorical descriptors change? The notion of “rock and roll” is very different now than it was 50 years ago.

An alternate method being utilized by many is the “folksonomy” approach towards categorization. This method creates associations between arbitrary songs based on terms that individual users have applied, rather than partitions/hierarchies of songs based on categories prescribed by experts/authorities. However, this method suffers from many of the same problems as genre descriptors. Terms don’t do any good unless people know what they mean. Furthermore the meaning of terms can change independently of the content they describe, and the meaning of the content can change independent of the terms used to describe them. Twenty years from now, I will wince as all my favorite songs growing up are labeled with the term “oldies”. I’m not arguing for “stability” in the method that songs are considered and indexed, but rather that the we not rely wholly on a layer of abstraction (categorical or even “folksonomical” terminologies) for indexing songs, and that the notion of exploring “song associations” via network associations is worthy of further research.

The user does have a surprising amount of control over what information is returned given the underlying network neighborhood extraction model. Even though there are millions of songs, the dominant selection patterns for playlists fall into a much lower set of “trends” at any given point i ntime. Identifying which pattern the user subscribes to (via their playlists), and then giving them a “local view” of the related data (the unrealized neighbors with strong connections) gives them likely candidates for a valid recommendation. People are like snowflakes in that they’re all different… they just often happen to be “more” similar to a distinct group of people… at least in music listening behavior. In fact the similarities they share are the basis for the evolution and dynamics of genres. In the end, these genres will reflect the attitudes and attentions of our culture… in other words, the genre will always follow the culture. As a side note Jay-Z and Linkin Park (hip-hop and modern rock/metal) realized that they shared the same fan base, and decided to do a cross over album. I think this is a perfect example of the genre “coming to the people”, rather than vice versa.

As a final note, criticism always serves the important role of illuminating the critics point of view. Several other articles were pointed out that either support or relate to some of the ideas I’m getting at. For instance, this paper uses MDS as a user interaction analysis technique for information retrieval. Even though they don’t use their MDS technique as a visualization interface, they make a claim for prominent “user profiles of behavior” that exist among the innumerable choices that are possible.

This paper uses a scatter plot of images returned from an image retrieval result, with differing layout and interaction strategies. The most interesting result (for me) was the poor qualitative user response given for their fish-eye lens approach, and the high qualitative user response given for their slider approach. Since my method is very similar to fish-eye, this is something I’ll need to test as well.

Currently, there is no sure-fire algorithm that will tell whether or not a certain song is good or not. Some may argue that Hit Song Science does this, but in actuality all it is doing is correlating the new songs to songs that have come before it, and determining if the new song is similar to a previously popular one. Besides discouraging novelty, this method is also completely proprietary, so in the end you’re relying on someone’s word that the computer produced the results… sort of a shady arrangement, and perhaps easily taken advantage of by unscrupulous labels.

So, in the end, even with correlation metrics like “Hit Song Science” people need to decide whether a song is good or not. Sometimes they have to decide as a group (listening to it when partying together), and sometimes they can decide it on their own (like I like to do with my huge headphones). Furthermore, it’s generally accepted that certain people are better at identifying good music than others.

This last point is interesting, because I know of about a dozen different ways of valuing artists (reach metrics, album sales, ticket sales, DVD’s, etc.), but there’s very little in the way of metrics for valuing individuals who find the good music early and often. I’ve taken to calling these individuals music mavens, which is the common term around myStrands, no doubt thanks to Malcolm Gladwell who used the term prolifically in “The Tipping Point”.

I’m putting together a version of a “Maven Metric” algorithm that well attempt to do this, and hopefully address the pitfalls of such a goal. Secretly, I hope that it can cause a “put up or shut up” movement in music review circles. All you would need to do to become an authority according to the “Maven Metric” is listen to good music early and often, and not pen long winded articles that are mostly self-righteous exercises of erudite prose.

Knowing that Chris is a fledgling expert in cybersecurity should prompt you to select option B, but maybe with reservations. After all… why would he expose the flaws publicly rather than directly to the appropriate committees?

Well, his actions seem to expose “security” (most notably airport security) as a troubling new dimension for our society to deal with… Chris would argue that the notion of security is being used as an “instrument” of the government for the indirect control or distraction of a population, rather than its assumed purpose of protecting the population. He calls this particular context “security theater“.

Under the assumption of “security theater”, the government and it’s various security committees would have no direct interest in acting on the advice of security experts… instead the security measures taken would function mainly as a “scare tactic”. In other words, the government would rather spend more money on armed security guards to shake down suspicious looking grandmas and expensive bomb sniffing machines than to fix simple technical vulnerabilities with easily duplicated boarding passes. Many security experts believe that “security theater” is indeed occurring, and are attempting to expose security threats through other channels.

The true test of whether or not “security theater” is occurring is watching how the government responds to Chris’s actions. The government’s first action was to send in the FBI, seize all of his passports and computers, and call for his arrest.

Now, the government is changing its tune… not only dropping their case (seemingly) but calling for Chris to be in direct talks with the Department of Homeland Security itself.

Is the government’s intentions sincere? Or have they realized that the biggest show in “security theater” is now “back stage” and are trying to cover their actions? I would wager that they can’t afford to be exposed as “the bad guys” in the current political situation.

Anyways, I think this puts Chris in an enviable role, even though he’s a bit disadvantaged in the short run. As America learns to deal with increased homeland security fear, I think we need more than one voice offering their perspective. If we’re going to be truly secure, we need experts like Chris exposing our flaws publicly rather than privately. Perhaps we should realize that we’re never really truly safe, and come to terms with it rather than trust in security theater.

I’ve read some articles by Larry
Lessig talking about alternatives to supporting artists (not just
supporting the RIAA/labels), but I can’t find them anymore.  I’m also
aware of a couple of other sites (such as sellaband)
that are trying some innovative models for supporting artists.  I also
find it interesting that Nielsen is including
iPod usage as a new rating dimension. 

It seems that a couple of things will come to pass… There won’t be a single
monolithic source for music (such as iTunes) because it won’t be able to
control the information, and therefore will lose economic viability. 
Artists will have to derive income through means other than through the sale of
their content… a scary thought… but artists receive very little from their
own digital content currently anyways.  Take Weird
Al Yankovic for instance.  It seems that the time is right for a new
model for artist compensation.