Data Mining: Text Mining, Visualization and Social Media

Local Search - How Hard Can It Be?

This week, Apple got a rude awakening with its initial foray into the world of local search and mapping. The media and user backlash to their iOS upgrade which removes Google as the maps and local search partner and replaces it with their own application (built on licensed data) demonstrates just how important the local scenario is to the mobile space.

While the pundits are reporting various (and sometimes amusing) issues with the data and the search service, it is important to remind ourselves how hard local search can be.

For example, if you search on Google for Key Arena - a major venue in Seattle located in the famous Seattle Center, you will find some severe data quality problems.

 

Google has confused Key Arena with Center House (another nearby venue) - we can see evidence of this in the photo. Consequently, the Zagat reviews for the two entities are mixed up. It seems that they have also confused the arena with the Seattle Center in general (judging by a reading of the reviews). Finally, the location of the arena (the push pin on the map) is in the wrong location.

Google also links data to reviews found on the web. The first link at the bottom of the g+ page for Key Arena links to Trip Advisor's page for 'Chihuly Garden and Glass'. All told, Google has over-matched at least four entities.

One of the ironies of local data conflation is that landmark entities (like stadia, large complex hotels, hospitals, etc.) tend to have lots of data (everyone knows about them) and lots of complexity (the Seattle Center has lots of things within it that can be confused). These factors conspire to make the most visible entities in some ways the entities more prone to problems.

The State of Hawai'i Demands a New Search Engine

We will soon be embarking on a short trip to Hawai'i. Naturally, I'm turning to search engines to find out about the best beaches to go to. However, it turns out that this simple problem - where to go on vacation - is terribly under supported by today's search engines.

Firstly, there is the problem with the Web Proposition. The web proposition - the reason for traditional web search engines to exist at all - states that there is a page containing the information you seek somewhere online. While there are many pages that list the 'best beaches in Hawai'i' as the analysis below demonstrates these are just sets of opinions - often very different in nature. An additional problem with the Web Proposition is that information and monetization don't always align. Many of the 'best' beaches pages are really channels through which hotel and real estate commerce is done. Thus a balance is needed between objective information and commercial interests.

Secondly, beaches are not considered local entities by search engines. While the query {beaches in kauai} is very similar in form to the query {restaurants in kauai} the later generates results of entities of type <restaurant> while the former generates results of entities of type <businesses that have beach or kauai in their name or associated content>. While local search sounds like search over entities which have location, it is largely limited to local entities with commercial intent.

Finally, there is general confusion due to the fact that the state of Hawai'i contains a sub-region (an island) called Hawai'i.

To get to the answer to my original search query, I reviewed 8 sites which resulted in a search on Bing or Google for the query {best beaches hawaii}. I then reviewed each of these and created a spread sheet tabling all the beaches and whether they were voted for by the site.

Of the 57 beaches that were mentioned on at least one site, the average number of mentions was 1.89. This indicates a general lack of consensus regarding which are the best beaches. In fact, most beaches (38 out of 57) have only a single vote. Consequently, while there might be a set of pages returned by search engines for queries looking for such information, a user will be reading - in isolation - very different opinions with no aggregate view summarizing them.

The top beaches are summarized in the following table showing the beach and the total votes.

Search engines could do a far better job by:

1. Generalizing local search to include any entity which has location, not just commercial entities.
2. Leveraging editorial content (like that reviewed in this post) so that variance may be exposed to the user but aggregates can also be synthesized.

In addition, there is a very large opportunity here in analyzing the content associated with these local entities to determine which beaches are best for different activities, their accessibility, and so on.

Google's Visual Approach to Local Search

I very much like Google's visual presentation of certain answers to local search queries that involve regions of space.

'Pittsburgh'

'98115'

'Fife'

Google's Evolving Approach to Local Search Development

I've been tracking Google's local search experience for a while (I work in Bing's competing local search product). Since February 2012, I've noticed only three variants or changes to the way in which local search results are presented on the main search page. The first is the number of ads displayed with each search. The second is the size of the map accompanying the search on the right hand side. The third is the prominance of the micro-poi (the small dots on the map) in the map design (they are now faded a little to help make the main information stand out).

So, in summary, there is not much happening on that surface - the main Google search results page.

Local search interactions tend to happen in one of two ways. The old-school approach is via a pair of text boxes. The first captures the what (restaurant, bar, sushi) and the second captures the where (Redmond).

This was the UI used in 2004 with the original Google product.

In 2005, Google upgraded this to a single search box - backed by algorithms which could pick apart the what and the where.

Over on Google+, I'm seeing quite a lot of energy being put into the local experience. While there are many Facebook like aspects to the experience (but less of the Facebook UI clutter), it is interesting to note that the query interface has cycled back to the two box approach.

Overall, my guess is that Google is focussing heavily on iterating their experiences on the G+ platform while the SERP (the main search results page) is largely cruising along with few changes.

Livehoods - Behavioural Neighborhood Mapping

I'm late to this, but it is certainly worth posting. A team of researchers at CMU have been working on mining foursquare checkin data to determine behaviourally defined neighborhoods ('livehoods'). They have put together a site - livehoods.org - which showcases their work.

The site has maps for Pittsburgh and Seattle - the work is even more awesome in that it was published at ICWSM this year.

Interacting with the map, as with this example below, shows the boundary of the derived neighborhood. What is interesting in the case below is that the area crosses a natural boundary (the river).

Update: here is the video of their presentation at ICWSM 2012.

Dear Web, Thanks For Using My Screen

In a previous post, I pointed out that while many of us use laptops, desktops and tablets with a widescreen form factor display, many websites fail to leverage much of the space. Some of the commenters indicated that they thought I was suggesting that sites have text running from edge to edge and why this would be a bad idea (thanks for the tip).

The sites (or rather pages) below show that it actually is possible to think about this space, and in very simple yet effective ways.

twitter.com/olympics

The BBC's sports coverage

Both use context and mood setting images in support of the textual and other content.

 

Dear Web, Please Use My Screen

We've all been using widescreen desktops and laptop for a while. When will the web catch up?

Urbanspoon:

BBC

Bing

Google Plus

Bing's Evolving Local Search

Recently, there have been a number of announcements regarding the redesign of Bing's main search experience. The key difference is the use of three parallel zones in the SERP. Along with the traditional page results area, there are two new results columns: the task pane, which highlights factual data and the social pane which currently highlights social information from individuals (I distinguish social from 'people' as entities - for example a restaurant - can have a social presence even though they are only vaguely regarded as people).

These three areas can be seen below:

The example above shows another key part of Bing's exploration of the search experience. The results for the query {sushi in seattle} are no longer displayed as a block of structured data inserted somewhere in the stream of blue link web pages. Rather, they are integrated into the results in a manner that leverages both the web presence of the entity (for local entities this generally means the home page of the business) and associates with it the structured data that identifies and characterizes the entity.

The task pane - which appears when the arrow tab is interacted with - holds more detail about the entity and provides affordances for learning more. There are options to compute directions to the location, view the streetside imagery or read reviews.

In addition, for local queries Bing now provides an aggregate view of results.

My team, along with a number of other teams in Bing, have been hard at work bringing this evolution to fruition. Here are some comments on the process:

1. When we provide flat structured data (as Bing did in the past), while we continued to strive for high quality data, there is no burning light focused on any aspect of the data. However, when we require to join the data to the web (local results are 'hanging off' the associated web sites), the quality of the URL associated with the entity record becomes a critical issue.
2. The relationship between the web graph and the entity graph is subtle and complex. Our legacy system made do with the notion of a URL associated with an entity. As we dug deeper into the problem we discovered a very rich set of relationships between entities and web sites. Some entities are members of chains, and the relationships between their chain home page and the entity is quite different from the relationship between a singleton business and its home page. This also meant that we wanted to treat the results differently. See below for the results for {starbucks in new york}
3. The structure of entities in the real world is subtle and complex. Chains, franchises, containment (shop in mall, restaurant in casino, hotel in airport), proximity - all these qualities of how the world works scream out for rich modeling if the user is to be best supported in navigating her surroundings.

It proved to be a lot of fun and technically challenging to get deep into the problem of local search (not to mention the need to throw in some philosophical underpinnings regarding the semantics).

Here is another example of a search result which highlights Bing's understanding of chains of local entities {starbucks in new york}:

Microsoft Surface : Bigger than Windows 8 and Bing

I've been reading the coverage on the new Microsoft tablet launch - the very cool looking Surface.

Over on track // microsoft, the twitter buzz seems pretty hot. I note that the main cluster around the launch has perhaps the highest buzz I've yet seen on the site, peaking at over 4, 000 tweets in an hour.

In addition, the Surface video has 22 mentions in the blog that track // microsoft scans - certainly the highest number of embeds I've yet seen on the site.

 

Recent buzz around Windows 8 and the newest version of Bing didn't peak at these levels.

5 Hidden Skills for Big Data Scientists

1. Be Clear:  Is Your Problem Really A Big Data Problem?

There are many big data problems out there requiring huge compute scale, innovations in computation paradigms, vast storage space and so on. But just because your data takes up lots of disc space does not mean that you have a big data problem. Firstly, your data may be encoded in an inefficient format. XML, for example, can be incredible verbose (all those close tags and human readable text). Secondly, if your data changes over time it may change very slowly indicating that monitoring the difference between data sets is more important that importing complete data sets. Thirdly, you may be processing your information on a legacy architecture designed for low power CPUs or cores. Architecture should be data driven, meaning that you need to deeply understand the informational aspects of your data and not just the size of the data as it comes to you on disc.

2. Communicating About Your Data

Often, in large organization (I work for Microsoft and have worked at IBM in the past), the product requirements for data deliverables are high level. For example: we need these variables to be 99% accurate. This simplistic view of data - that a level of quality can be delivered in a specified time frame - is ignorant of the highly opportunistic nature of processes that improve the quality of data. Consequently, a data scientist needs to aggressively manage the communication about projects which transform and improve data sets. Do as much research as possible to minimize unknowns, but don't sign contracts that involve both time and quality metrics!

3. Invest in Interactive Analytics, not Reporting

When you construct reports about your data products, you are answering a fixed set of questions. This is useful for monitoring, but it doesn't provide a way to get at the unknown unknowns. It is only through interactions with data (often called slicing and dicing) that pockets of interest (problems and opportunities) are discovered. Rich, interactive tools may be perceived as a low priority and never quite got to. Avoid this peril!

4. Understand the Role and Quality of Human Evaluations of Data

When trying to determine how good your data product is, it is often the case that we employ an array of human judges to evaluate a sample of the data. The higher up the management chain you go, you tend to find a higher degree of respect for human judgement. There are many studies, however, that show that human judgements are not always as good as they are cracked up to be. In many cases, machines can do better than humans, they just tend to make different types of errors. On deeper inspection, human errors can be traced to the structure of incentives around the judgement process. Innovate in methods to compare data sets that help distinguish their relative quality without necessarily the expense of human assessment.

5. Spend Time on the Plumbing

How does data get in to your system? How does it flow? Are you sure every bit of information got in? With large scale data loading and processing systems, one doesn't one a small number of failures to tip over the entire run. However, silently failing components can cause big headaches down the line when you are reporting your summary findings. Make sure there are no leaks in your pipeline!