May 24, 2008

Powerset ( Mathematical Pun )

Lot has been said about Powerset since its inception and blogosphere has not spared it either after its release in the recent past. Search is one of the toughest problems - given the magnitude and the sheer plethora of the information around. Though most of the blogs concerning the Powerset launch deal with how the management ended up having all the NLP grads(from Stan etc) and paid them 100k+ to develop this platform in SF; i will try to share some personal experience with this engine and how my expectations hasn't been satiated still.

Powerset was touted to be a Q&A engine with 'human' intelligence inbuilt into it. I was expecting some practical applications of Discourse theory too here. But neither were 'visible' in the experience that i had.

This blog is by no means a comprehensive review of the entire engine and its product strategy and the market segmentation brought on by Powerset, but sheer reflections of some emotions(read 'rants').

read on...

--> I was confused for a minute when i first saw the page result(search-hits) after doing the search for 'gandhi'. Now you need to understand when i say this that, i am referring to a 'typical' lane man who wants to search for some recipe or some term. There are just too many UI elements which can easily confuse your grand-mom.

--> I did a search for 'gandhi' and the results that i got were nowhere related to Gandhi(noun) but to the movie and the music bands and etc (in the tabbed bar above the Factz). Hence, i am not going to share what happened when i did a Q search (you may check other blogs for guys who did 'ask' question and the answers were a sheer 'mashup' of all the 'associated' keywords)!

--> The Factz were just too simple and i could not much gather anything from it, except for a solo fact which stated 'Quit India Movement'. The Factz look more like Powerset has simply searched on the n-grams gathered from all of the Wikipedia articles which has a 'rough' semantic association with the word 'gandhi'.

--> The widget on the right of the page (with delicious/reddit/et al links) is simply not required here. Atleast this could be made to bookmark the link with the search-word, but it instead bookmarks the powerset.com site. Anyway, I find this as a criminal waste of screen real estate.(But on a different note, this can be accommodated now, as Powerset has just launched and would be just interested in getting themselves bookmarked the max number of times - somehow, i don't subscribe to this PoV).

--> At the bottom of the page, the section on "Exploring the following pages on Powerset" is something that is simple but yet elegant. I would have preferred to see this result than seeing the Factz.(This is something similar to Sponsored Searches found in many other sites and corporate portals).

--> When you click on any of the result-hits, then the link opens in a new page ; which has an Article Outline on the right. Again, i simply would not even look at this - frankly speaking i did not even observe this, probably it was in my blindspot. Also, the way in which it slides when the page is scrolled - nice effect , but does not amaze me.

Concluding remarks:
# It looks like the entire is just an Alpha release , am sure that the Powerset team can collate its entire '(in)organic intelligence' and take this product to the next level; in the current release its a sheer play with ngrams and some named entity recognition (NER). Am sorry to say, but this does not look like the much-anticipated-Google/Live/Ask-Killer; in fact, not even closer to any of the latter. I would prefer Clusty.com to Powerset.

# Indexing all the wikipedia articles using Lucene/Solr, some NER and using the faceted searching would end up having almost similar results. I somehow opine that using simple algorithms and heuristics often ends up in having 'almost-similar' results than spending millions of USD and thousands of man-hours on developing a product that would be a step advanced than the former.

# Some bloggers have stated that Powerset can be a probable fit in the Enterprise Search space - but still solr+some-nlp-lib would be better of there (or use Google CSE ;) )

# Get the footer of the page properly aligned!!! It looks really bad in Firefox - The text "Privacy Policy" and "Except where" overlap badly.

......... And its not necessary that all sites have to be Javscript powered to give it a 'cool' look&feel .. simplicity rocks!