Tips to avoid hitting rate limits for my movie monitoring application.

[Rahul]

Hello,

I am building an application that monitors tweets about movies(for now
with... other interesting things planned). I have my id whitelisted
but I want to avoid overusing it.

The challenge that I face is that ideally I want to make full use of
the opportunity to retrieve 100 tweets per call and for that I need
information on the frequency with which users are tweeting about a
movie and then set my call frequency (to call twitter search api)
accordingly so that I maximize the number of tweets returned per call
or atleast.

Since I presume there is no way to know what frequency is someone
tweeting about a movie - I need help is what is the best way to
optimize for such a situation.

The challenge is complicated by the fact that users tweet about
different movies at different rates and the rates generally decrease
overtime.

I have tried combining searches - but the challenge is that lets say I
search for

(Movie A OR Movie B)
(Movie C OR Movie D)

it could be the case that people tweet about Movie A & B a lot and
litle to none about C or D or there is a combination in which they
continue to tweet about A but not about B - So I still can end up in a
situation where I am not optimizing my calls. Also situations such as
Oscars can dramatically change what people talk about even about
movies out months ago.

I have thought of writing something such as a variable frequency
caller that can check the frequency of tweets for the last 3 calls in
order to appreciate the frequency of tweets for a given search and
then continuously vary the time between calls so that I can get as
close to 100 tweets as possible in a call.

Any ideas suggestions that can suggest ways to alleviate the above
will be highly appreciated.

Thanks
Rahul.

[Mark McBride]

This sounds like a perfect use case for the streaming API.  The rate limits there are different, but in general more permissive. And because you're doing primarily OR queries, the current track functionality seems sufficient.

  ---Mark

http://twitter.com/mccv

[M. Edward (Ed) Borasky]

What would make use of Streaming for this use case a *lot* easier
would be if Twitter would export to the API more detailed information
about the Trending Topics. For example, I'd like to see more topics
than just the current number displayed, and tweets per unit time
(hourly worst case) for each topic. I'd like to see at least the Top
100 and maybe even the Top 1000! This seems to me to be an easy task -
you've got to do the computations anyway, right? Heck, with pages /
cursors, you could send the whole table out and let people do their
own cutoffs.

For example, over the weekend, the Trending Topics were,
understandably, dominated by the Oscars. That's ten or twenty right
there, by the time you factor in the fact that Farah Fawcett got
ignored in the memorial, ten pictures nominated for best picture, ten
Best / Best Supporting actresses, ten actors, etc. Throw the perennial
Justin Bieber and Lady Gaga into the mix and it's clear there's
interesting and useful information further down the list. Why should
we have to monitor Streaming and do our own topic analysis and
filtering, or subscribe to some service with Firehose access?

--
M. Edward (Ed) Borasky
borasky-research.net/m-edward-ed-borasky/

"A mathematician is a device for turning coffee into theorems." ~ Paul Erdos

[Rahul Dighe]

Hello,

Correct me if I am wrong but doesn't the streaming API has limitation that allow me to only track 200 keywords.. and also with the added caveat that -

Track keywords are case-insensitive logical ORs. Terms are exact-matched, and also exact-matched ignoring punctuation. Phrases, keywords with spaces, are not supported. Keywords containing punctuation will only exact match tokens. Some UTF-8 keywords will not match correctly- this is a known temporary defect.

If this is the case how will the api track keywords such as "The Hurt Locker" or "The Blind Side"?

Thanks
Rahul Dighe

[Mark McBride]

This is correct.  The general advice is to choose the most specific keyword to track (probably "locker" and "blind" in this case), then run an additional layer of filtering on your side.  There are higher access levels available that grant you more than 200 keywords to track.

  ---Mark

http://twitter.com/mccv

[Rahul Dighe]

thanks - I need to put more thought into this - I am inclined to feel that at the moment that the search api will probably deliver better resuls - as the cost of filtering thousands and thousands of records for even something as basic as a movie called "New York" or "Independence Day" split into independent words will probably be cost intensive and might end up being looking for a needle in the haystack.

Having said that I think Twitter has surely come up with this API with good thought - it's just needs further analysis from my end with regards to whether the cost of filtering outweigh the benefits from getting real time streaming resuls.

thanks
rahul.

[John Kalucki]

We'd like to offer phrase search, or at least AND search on the Streaming API, but we've had other priorities recently.

Note that Search is not intended for repeated automated keyword queries, and that Search results are filtered for relevance. If you need all the Tweets, or if you need them in real-time, the Streaming API is the best answer. The Search API is mostly intended for complex, historical backfill, ad hoc, and direct-display-to-user queries.

-John Kalucki
http://twitter.com/jkalucki
Infrastructure, Twitter Inc.

[Rahul Dighe]

thanks john - I have not considered the implication of search results being returned by relevance - I will give the streaming API a shot -