I hate to go back to what I was saying earlier. I understand that the algorithm is still under development, but if we understood what the algorithm for defining a popular tweet was, we would be able to understand why Osman is seeing the results that he has found here.
On 2 April 2010 22:48, Osman Ali <[email protected]> wrote: > Hi: some questions regarding the popular tweets feature. These are all > based on a user-end interaction with search.twitter.com. > > 1. Is popular search moving to home page search at any point or is it > just staying on search.twitter.com? I wanted to know if the home page > will just be showing recent tweets as it does now. > > 2. What is the "refresh rate" if you will of popular tweets over a > given time period? What does "recent retweets" mean because there are > instances where that applies to minutes and others where the tweets > generating the popularity are hours or even a day old? As in, when > does one Tweet replace another tweet as a popular result for a given > query. This seems to vary across different search queries with some > being more stagnant than others. > > 3. Is there a "drop-off" point for the popular results for a given > query after the numbers of retweets subside that the results are taken > off or do they just stay there indefinitely until replaced? > > 4. Do you have a way of determining when irrelevant popular tweets > have been maliciously set as the result for a given query and will > they be actively taken off? > > Here are some examples. > > For the past twelve hours or so a search for "#nowplaying" has yielded > the following: > > http://twitpic.com/1cokv3/full > > It almost seems as if the use of the popular tweet ranking in this > case actually takes away from the real-time search experience by > having dated tweets at the top of the queries. Once new results arise > for the query, these tweets are gone which gives way to a real-time > feed but initially these popular tweets are there and in this case > none of the leading tweets that are "popular" actually have the > utility expected from a user (i.e., users seem to use the > "#nowplaying" search as a way to scroll through tweets like > "#nowplaying Clocks - Coldplay). > > It seems that for a query like this where users generally just scroll > through the list of results as a way to discover new music or even > other users and thus generally do not retweet individual tweets that > often for the query, there is a level of stagnation where old tweets > linger for hours as opposed to more active queries where individual > tweets gain attention. > > Additionally, the popular search ranking in this case has no way or > weeding out artificial results that are not query related. The top > result from the image above is actually an advertisement being run by > a company that contains a link leading to this: > > http://twitpic.com/1coml8/full > > A search for instances of the two most popular tweets using the text > content of each one to show recent instances where they have been > retweeted shows the following results: > > http://twitpic.com/1colcd/full > http://twitpic.com/1com2o/full > > The most recent tweet of both the first and second most popular tweet > is by the same account. Here's a look at that account, which seems to > be set up for a very specific purpose. > > http://twitpic.com/1comde/full > > There's also this account which retweets the top "#nowplaying" query. > The account is also clearly set up for a very specific purpose and is > retweeting a tweet that does not give users the expected query result > for a "#nowplaying" search. > > http://twitpic.com/1comgd/full > > This seems to show that the method of curating popular tweets may need > to take into account the retweet versus the output rate for a given > query otherwise the system can easily be taken advantage of to give > users irrelevant results based on similar tactics. That is, for the > "#nowplaying" search, even though thousands of tweets are being > outputed by users, not many of them are being retweeted so all it > takes is for someone to alter that dynamic as is the case here and > because of the "refresh rate" for the query, irrelevant results can > sit at the top for an extended period of time. > > But in cases where a twitter users with involved followers starts a > discussion, the popular tweet ranking not only picks up on that > effectively but it also allows new users to the conversation to have a > reference point as is the case here (with the user's tweet being > listed first and then legitimate popular tweets right after): > > http://twitpic.com/1coyhy/full > > And in cases where there is news or an event, it seems that retweeting > of a legitimate news source for a query leads to an instance where a > news source is immediately available as a result for the query. > > http://twitpic.com/1cp7sh/full > > In both of these examples, "recent" is truly recent as it happens > almost instantaneously. > > However, these are still instances for even the more popular queries > where using retweets over a given period of time as a measure of > popularity can lead to irrelevant query results unless the refresh > rate of the "retweet count" somehow takes into account the fact that a > single instance of numerous retweets might not necessarily be relevant > a fixed amount of time later. In efffect, the popular tweets feature > can lead to instances where the content being generated about a topic > has moved beyond what was popular and being retweeted even a few hours > ago. So you end up with a part of Twitter that is actually slower than > the rest of it. Here's an example of that using a popular query: > > http://twitpic.com/1cp61k/full > http://twitpic.com/1cp7pt/full > > The first image shows the query results and the second a search of the > text contained in the third most popular tweet. Although this is a > query for which there is always a tremendous amount of output, the > third "most popular" tweet about is one that a content search reveals > the last retweet to have been ninteen hours ago. Additionally, the > content of the tweet is not relevant to a current query about the > topic. This indicates that there is sometimes no direct relationship > between "recent" number of retweets and current output of content. > > Osman Ali > @namsoila > > > -- > To unsubscribe, reply using "remove me" as the subject. >
