This hack is an extension of Vik Singh’s qna service, which finds answer using the popular phrases in the top search results for a query. I do image search for the best answers and blend them with the regular image search results. The hack is a basic prototype and natural language image search gets triggered only for questions (queries including who/what/which).
Below is a quick comparison of search results obtained by askBoss, Google image search and Y! image search for query: who is batman in the dark knight?
With Yahoo Boss APIs and a deployment platform like Google App Engine, building a decent search service is pretty easy. I could finish this hack within a few hours by using Boss Mashup Framework and App Engine. Apart form the qna service, other popular Boss API/app engine integrations include 4hoursearch aka YUIL.
Ignoring stop words, doing word stemming, etc. can really change the meaning significantly. For eg, searching Apples on google, returns results mainly for Apple Inc.
Multi-lingual search: Web being driven with focus on US market, problems of the rest of the world (especially eastern world) do not really get sufficient attention. Today’s web search experience does not have multi-lingual features! I spent couple of years during my masters at Media Lab Asia,IIT Bombay, under Prof. Krithi, with people working on Multi-lingual search for project aaqua.org. Multi-lingual search works pretty good here. Try searching “onion” or “कांदा“, you get identical results :). Such a search experience on the whole web, will be awesome!
Treatment of Symantec data: Lot of standard formats have emerged like RDFs, microformat, RSS, etc, but still they are treated in almost same way as other web pages.
Personalization & Data mining: There are a few signs of google personalizing the results. But, nothing significant yet!
Multimedia search: None of the search engines is doing a great job here, which is attributed to complex and computationally expensive image processing. But, pretty significant research is up for the same in Google, Yahoo and Microsoft. A recent publication in WWW08 from googlers suggested a concept ImageRank, similar to PageRank which can actually work well.
Some of the cool innovations in todays search:
Improved UI/visualizations: UI innovations are the most prominent amongst all. Here are a few set of examples:
Openness/APIs: Google/Yahoo have been pretty open in terms of providing search APIs, applications, etc. Want to experience google search in a terminal: Try out www.goosh.org
Specialized searches like: local/maps: Local/maps and other focussed searches like publication search, patent search, etc. are doing pretty good. Directions are now available in India also with Yahoo Maps the only provider
What else can be tried on search?
There are couple of things things which I think can work for search, but we need to overcome spam problems for these:
WikiSearch: Allow users to tag/rank search results. Something like digg/delicious for keywords…
Push based update notification model: Search results are not uptodate. Even for popular pages they lag by few days. Introduce a push based model, something like blog.gs, it can help?
Disclaimer: All opinions are solely mine and and do not necessarily reflect the opinions of my employer.
Yesterday, a friend of mine was looking for a way to download original photos from one of his sets on flickr. Since we could not find any flickr downloader which can run on Linux, (for windows you can try : http://flickrslideshow.fateback.com/), I quickly wrote a script using flickr APIS. The sed and awk power made it v. easy
Below is the script:
SET=”<Set - id >” # Enter SET ID here from which photos have to be downloaded, for eg. SET=”72157604130281022″
APPKEY=”" #Your APP key here, get one from http://www.flickr.com/services/api/keys/
curl “http://api.flickr.com/services/rest/?method=flickr.photosets.getPhotos&
api_key=$APPKEY&photoset_id=$SET&extras=original_format&per_page=500″ | sed ‘~s/title=”[a-zA-Z0-9_ :)?(.]*//g’ | awk ‘/id=/ {print “http://farm3.static.flickr.com/”$4″/”$2″_”$8″_o.jpg” }’ | sed ‘~s/\(server=\|originalsecret=\|id=\|”\)//g’ > p
wget -i p
rm p
Early this week, Yahoo acquiredIndexTools, an analytics company. IndexTools offers tools for monitoring and analyzing websites.
The interesting thing, which happened the very next day (10th April) after the acquisition, was an email send by Google Analytics team to all (or many) analytics users notifying about the benchmarking feature, which was launched on March 5th!
The email stated:
“We are writing to let you know about a change in our service offerings. If you have logged into your account recently, you may have noticed that you can now choose to share your Google Analytics data. … We’re also happy to announce industry benchmarking as the first new feature available …. Benchmarking lets you compare your metrics against industry verticals….”
I feel the email, on the next day of acquisition was not a coincidence, rather Google wanted to try its best to avoid any chunk of users migrating to Yahoo/IndexTools. If Google just had to notify the analytics users about the service change, they must have done this a month back.
Flickr videos come to users with few restrictions:
1. Videos can be uploaded only by pro members though anyone can see them
2. Videos can be atmost 90 seconds long
3. No more than 150 MB per video. (Well, I do not see any 90 sec video going beyond 150MB)
The restrictions are more seen as a way to avoid illegal videos coming in and infringing copyrights. The goal is not to have another youtube but rather a place where you can upload videos that you have created.
