Phrase search in Movies and TV programmes

Subzin is a fun site - you can search the scripts of many Movies and TV programmes to find the a phrase which you remember, but can't quite remember which Film/Show it comes from.

Phrases are specified with the usual double quotes "", otherwise words are search as is. Stop wording is obviously not done, which makes sense if you are providing search for words actually spoken. Unclear as to the type of indexing being used.

Advanced search provides the user with the opportunity to search for a phrase in a particular title, and to restrict search by popularity, Genre, Type (Film/TV) and language (but English is only available at the moment).

(h/t: David Thompson).

Image Search by Example

Relevance feedback is a very useful tool to either modify or expand queries according to user needs, or to provide an example with which to start a search. The former tends to be used in text retrieval, the latter in image retrieval.

An example on the web is TinEye, provided by a company by the name of Idée.

It works by finding an exact copy of an image you specify, either via upload or via a link. It does not find near images, the result is binary - either the image matches or it doesn't. Some small differences are allow for example, searching for Mona Lisa in curlers will bring up the original version of the Mona Lisa. One interesting part of the service is that if you provide a web page, it will load the images, and allow you to click on them to conduct a search. The service is useful for tasks which require known item search - e.g. I've got this image, has someone used this before on the web, I need an original image.

(h/t: Tanya)

Non-English Video Search

Found a couple of sites which provide non-english language video search, RuTube and YouKu, both obviously 'inspired' by YouTube (which I tend to use rather a lot).

More details of RuTube and YouKu are available.

Can't enter Russian or Chinese queries, so I tried some English ones.

I tried queries 'putin' and 'manchester united'. The former yielded this rather amusing video called 'Putin on the Ritz' further down the hitlist. Found lots of relevant stuff for the latter.

I tried queries 'Uighur' and 'manchester united' on YouKu, found more interesting stuff on the latter, music videos mostly on the former - the area being something of a sensitive subject at the moment.

Bottom line - English queries seem to work quite well on these systems, and its clear that the services have thought about cross language issues to some extent.

Music Search

Downloaded Spotify on my work machine ages ago, but only just got to installing it on my mac at home. This works extremely well with the Bowie/Berlin query set, which I can't show because you have to log in, but it has all the tracks, and its really quite an impressive service. Searches on the set provide not just the album, but also extra tracks e.g. covers. Only problem is that my version of little snitch keeps popping up with messages, and if you don't accept the connection, the track won't play! An independent review of the service is provided by David Toube.

Medieval Image Retrieval

University of Cambridge provide access to a set of Medieval images, and provide commentaries on them. Basic search is - very basic.

You can choose in advanced search:
1. Media: type of painting used oil, tempura, gold and medium such as paper book.
2. Cycle: don't know what this is.
3. date: 11th to 16th Century
4: Title: of painting, any word or exact phrase.
5. Keyword: search restricted set of keywords e.g. Christ in Majesty, Evangelists, Majesty
6. Image number (if you know it).

Doesn't seem to be a full text search on the commentaries unfortunately.

Thesaurus Search

Thsrs finds short words from longer versions. Heres the list I got for 'establishment':

constitution
formation
organization
organisation
institution
governance
brass
ecesis
validation
action
activity
beginning
body
commencement
construction
proof
start
structure

If you ask why, you'll hopefully find the answer here.

Video Search

Corbis Motion is a website that provides access to various types of video including royalty free videos. The database contains short clips of people doing things, heres one for a search on football I did. Clicking on a video loads a new tab (or age) provides information on the clip e.g. length, frame set, what camera the video was shot on etc . Mousing over a clip plays it, which is unique.

Local Search

UpMyStreet is a local search facility, based on postcodes, and provides the same kind of information you can find in the yellow pages. With UpMyStreet you type in the postcode and information in your area is then provided e.g. here's one for near where I work. You can then browse the categories. With Yellow pages you can either browse the categories available, or search on the postcode refining by categories and/or company name. Searches bring up a map with companies locations. Quite an interesting difference in the strategies: is it better to provide a more focused set of information at the outset or more functionality in search with access a big list of categories?

Music Search

Gracenote is a service based on CD meta-data which has been made into a business. More details from wikipedia, and relevant info on CDDB, based on this format. Hmm. not sure about this! Here's my attempt searching this system with the bowie/berlin query set.

• David Bowie Low. Didn't find the album in the first page hitlist, but there were albums with hit records, so partial relevance was established. Retry. Did find the album but it was lower down the the, in fact 'low'. Groan.


• David Bowie "Heroes". Same hitlist as low query. Retry. 3rd hit down.


• David Bowie lodger. Same hitlist as low query. Retry.


• Iggy Pop lust for life. Album is 2nd hit. Retry.


• David Bowie the man who sold the world. Same hitlist as low query. Retry.


• David Bowie hunky dory. Same hitlist as low query. Retry.


• David Bowie Aladdin Sane. Same hitlist as low query. Retry.


• David Bowie ziggy stardust. Same hitlist as low query. Retry.

I did loads of initial searches and got identical results for the Bowie queries, before I realized that I could refine the query to 'album' or 'track' - it seems to automatically assume you're looking for artist. I therefore did all the searches again, this time on with the 'album' tag chosen labeled 'Retry' above. Retries are with the album name only. Results were a lot better doing this. Clicking on the albums allows the user to buy the track (from itunes) or view the song lyrics. Otherwise a pretty basic service, with no audio options I could fine.

Music Search

Been a while since I've looked a music search engines, so I thought I'd better have a look at a few more. Here a system called 'Audiobaba', with searches using the bowie/berlin query set.

• David Bowie Low.


• David Bowie "Heroes".


• David Bowie lodger.


• Iggy Pop lust for life, (0 hits for either query)


• David Bowie the man who sold the world, (0 hits).


• David Bowie hunky dory, (0 hits)


• David Bowie Aladdin Sane.


• David Bowie ziggy stardust.

Impressions. It didn't like my queries above, and returned nothing for three queries before I gave up. I then tried just the name which came up with a list of albums songs and films to which Bowie was associated with. Links to queries I tried are presented above. It only plays samples of the songs currently. Without the ability to refine the query in another way (e.g. by artist), the service is rather limited. Searching on just one element made the queries rather ambigious sometimes (e.g. see heroes query above). There is a way to refine the query by choosing 'mainstream', 'normal' and 'independents' which is only available once you have done a search. I'm not sure of the distinction between 'mainstream' and 'normal' (these terms are not explained on the website), but I assume that the scheme is partly based on label type ('independents').

If you can get a query to work, the results then become very interesting. The service is billed as a recommender search engine, and clicking on a hit from a search brings up a list of similar songs based on the analysis of the audio, according to the FAQ.

Image Search by colour and example

Idee is a company that provides image search software, and they have a number of products including TinEye, which is a firefox plugin to provide query by example search for images. They provide a number of examples on various data sets (Alamy and flickr):

1. Search by colour on Alamy or flickr sets.
2. Visual search by example using an image from the Alamy set.
3. Visual search by example, using an uploaded image, either directly or using a URL on the Alamy set.

This is rather fun, enjoy! They have a blog.

Related: Hermitage museum provides similar search on their art collection, by colour and by layout, which are both great fun to use! Both provide demo's. This particular system is based on IBM's QBIC system.

(h/t: David Thompson)

Visual Search Engines

There are a number of visual search engines around, which instead of providing a list of hits in a linear fashion (like google), different metaphors are used to represent the retrieved pages. Here are some examples:

• http://www.kartoo.com/ - this is a search engine which uses Cartography as its basis, and uses map like visualisations to represent the importance of pages, and he relations between them. Nice idea, but its not clear how the metaphor works.


• http://www.search-cube.com/ - uses a cube to represent results, which you move around with either the up, down,left right keys or shift/mouse movement. Thumbnails are shown as tiles on each face of the cube - thumbnails can be pages, images, videos.


• http://quintura.com/ - presents a tag cloud, together with a linear list of results.


• http://www.searchme.com/ - presents screenshots of pages, which the user can navigate to either the left or right. Search terms are shown on the retrieved screenshots. Video Demo..


• http://redzee.com/ - much the same as searchme, but with more pages to left and right displayed.


• http://middlespot.com/ - presents a set of page screenshots in the same way image search engines work, but these are thumbnails of the retrieved pages. One page at time is expanded and highlighted.


• http://viewzi.com/ - presents results as tiles, can either view text versions or screenshot thumbnails. To change view you navigate left and right. Video Demo.


• http://www.sbrows.com - presents a list of thumbnails in the lower part of the browser, and a screenshot of the page being inspected is presented in the middle (and larger) section of the browser. Navigation of the list is done left to right. This is a beta service, and doesn't appear to be working properly; thumbnails clicked on don't seem to match the screenshot when I used it. Video Demo - no sound.

Basic problem with all these systems - how do you represent a multi-dimensional space in 2D or 3D representations? All the the services here are either meta-search engines, or are another type of interface to one search engine (e.g. Google). Its difficult to know what some systems are using as their underlying search services.

Roger K Summit and Dialog

Rather long, but see here:

Roger founded Dialog

Music Search

Midomi is a service which allows you to do music retrieval by humming (query by humming)!

Unfortunelty I don't have the equipment to do this either at home or at work, but it does have a search button, so I tried the usual dataset on that:

• David Bowie Low.


• David Bowie "Heroes".


• David Bowie lodger.


• Iggy Pop lust for life.


• David Bowie the man who sold the world.


• David Bowie hunky dory.


• David Bowie Aladdin Sane.


• David Bowie ziggy stardust.

Some searches found the relevant albums, pretty well all the relevant song were found. Seemed to work quite well. Wish I could try the query-by-humming though....

Posts in the series are available here.

Film Search

The Internet Movie Database (or IMDb for short) provides information on films either past, present or future productions.

Heres some searches for a few films with one actor (David Bowie in that guise):

• David Bowie the man who fell to earth (search produces exact result).


• David Bowie merry christmas mr lawrence (first match is approximate, and relevant).


• David Bowie the hunger. (no relevant results on list)


• 300 (first hit item is relevant).


• brazil (first hit item is relevant).


• citizen kane (first hit itme is relevant).


• decadence (very ambiguous an exploratory query, most hits would be relevant in some way).


• watchmen (2nd hit under exact match on title).

Overall its a very useful service and I use it all time. The results list comes back with all kinds of results produced in different ways, so in general you can pretty well find what you want.

Posts in the series are available here.

Video Search

Veoh is an online video website with search facilities. You can also search channels, groups and people for videos. Like other video websites (YouTube, JumpCut) it provides facilities for you to upload your videos. Lets see how we do with the 'Berlin Bowie' query set (on videos).

• David Bowie Low. (no relevant results)


• David Bowie "Heroes" (first hit was song).


• David Bowie lodger (hitlist looks the same as Low).


• Iggy Pop lust for life (no relevant results).


• David Bowie the man who sold the world. (no relevant results)


• David Bowie hunky dory (hitlist looks the same as Low).


• David Bowie Aladdin Sane (slightly different results from Low i.e. different order, but still nothing relevant).


• David Bowie ziggy stardust (first result presented is relevant).

The results on these searches is generally very disappointing, if this is anything to go by then YouTube doesn't need to fear much about the competition!

Posts in the series are available here.

Video Search

JumpCut is a website which allows people to create videos and upload them for searching. This appears very different from YouTube (a service I use all the time) in that the users create most of the content (rather than post content from third party sources e.g. music video's).

The results on the 'Bowie Berlin' query set are therefore interesting:

• David Bowie Low.


• David Bowie "Heroes".


• David Bowie lodger.


• Iggy Pop lust for life.


• David Bowie the man who sold the world.


• David Bowie hunky dory.


• David Bowie Aladdin Sane.


• David Bowie ziggy stardust.

None of the searches brought back any video's of relevance to me, and many of the hitlists look very much the same. Each hit records the views, plus love's, clearly a mechanism for obtaining relevance from the user. Probably used in recommendations. Service appeared a little slow on some of the searches, I had to start a couple again.

Posts in the series are available here.

Mobile Search

Taptu is a search engine that presents results in a format acceptable to a mobile phone. They have a very simple interface for a mobile phone.

Search can be done on the following sources: All, Images, Videos, Music, Web, Artists, Blog Posts, Desktop Web, Movie Info, News Stories, Song lyrics, Songs, Travel Guides, Videos, Wikipedia.

One thing that is clear here, it doesn't use location or contextual information to refine the search as yet, its just a search for a mobile phone.

Here's some results on he 'Bowie Berlin' query set:

• David Bowie Low.


• David Bowie "Heroes".


• David Bowie lodger.


• Iggy Pop lust for life.


• David Bowie the man who sold the world.


• David Bowie hunky dory.


• David Bowie Aladdin Sane.


• David Bowie ziggy stardust.

The result is generally in there somewhere (note: all searches were done on the desktop).

Posts in the series are available here.

Geographical Web Search

I suspect everybody has used 'local search' on Google Maps at some stage or other, but its not the only type of search you can find on the web.

Here's a list of services I have so far:

Placeopedia - Geolocates wikipedia articles.
Wikimapia - searches user descriptions of places and displays the results on a map.

I tried the search terms strawberry fields. Search didn't work on Wikimapi :(
- so I tried random place, so it does have data!

Both of these services are using Google Maps.

I've put these on one page, but will add more Web Search services of this type as I find them or people send them to me.


Posts in the series are available here.

Spanish Web Search

In a recent article in the Communications of the ACM, I picked up some URL's for non-English including Spanish.

Here's a list of Search Engines I gleaned from that article:

Quepasa (appears more like an online community - very slow!)
Orange

I tried the query spain european champions. Orange appears to index UK pages as well.

I've put these on one page, but will add more Web Search services of this type as I find them or people send them to me.

References

Chung, W. (2008). Web searching in a Multilingual World. Communications of the ACM, Vol. 51, No. 3, pp 32-40.

Posts in the series are available here.