The anatomy of a trending topic on Twitter and the rate at which online interests change are the focus of a new report released by the mico-blogging site's researchers.
In an effort to scientifically analyse the real-time nature of the site, the study tested how quickly terms and phrases appear and disappear in real-time search queries on a platform like Twitter. The result is a study titled: “A Study of 'Churn' in Tweets and Real-Time Search Queries”.
Twitter research scientist Jimmy Lin says: “We see that the most frequent terms in one hour or day tend to be very different from those in the next - significantly more so than in other content on the Web. Informally, we call this phenomenon churn.”
According to the study, the growing importance of real-time search brings several information retrieval challenges - one being the issue of rapid changes to term distributions for search queries.
The study examined all search queries from October 2011, and found that 17% of all top 1 000 search query terms disappear or “churn over” on an hourly basis. A further 13% drop off after a day.
“During major events, the frequency of queries spikes dramatically. For example, on 5 October, immediately following news of the death of Apple co-founder and CEO Steve Jobs, the query 'steve jobs' spiked from a negligible fraction of query volume to 15% of the query stream - almost one in six of all queries issued,” says Lin.
Algorithm adjustment
According to Lin, the research shows that local and global news breaks on Twitter and users flock to the service to find out more information. As a result, Twitter says it is constantly improving its relevance algorithms.
It was also noted that the rate of churn was significantly increased when a trending topic was removed on Twitter. According to the study, the typical lifespan of a trending topic is no longer than an hour.
“In the context of term churn, rapidly-unfolding events - such as natural disasters or political unrest - are of particular interest. In such scenarios, term frequencies may change significantly over short periods of time as the discussion evolves,” says the report.
“Trending topics churn 'naturally' at a rate that is slower than hourly, such that removing those events from the distribution increases overall churn. The time that a particular topic trends is a function of many factors: for news, factors include significance of the news event and interactions with competing stories; for internet memes, the lifespan of a particular hashtag is often idiosyncratic.”
The researchers also looked at the implications the rate of churn has for ranking in a real-time search engine. One of the possible conclusions drawn from the data is that a simple approach to account for changes in term frequency over time on a platform like Twitter may not be sufficient.
“Additionally, it is clear that, at least during major events, sub-hour updates to various collection and query term statistics are essential,” says Lin.
“Although query churn is consistently high, during major events it can further increase dramatically, as queries change minute by minute. In fact, to maintain accurate collection statistics requires frequent term count updates - in intervals of five minutes or less, according to our data.”
Share