Twitter Paper of the day: Searching Twitter: Separating the Tweet from the Chaff

Within the millions of digital communications posted in online social networks, there is undoubtedly some valuable and useful information. Although a large portion of social media content is considered to be babble, research shows that people share useful links, provide recommendations to friends, answer questions, and solve problems. In this paper, we report on a qualitative investigation into the different factors that make tweets ‘useful’ and ‘not useful’ for a set of common search tasks. The investigation found 16 features that help make a tweet useful, noting that useful tweets often showed 2 or 3 of these features. ‘Not useful’ tweets, however, typically had only one of 17 clear and striking features. Further, we saw that these features can be weighted as according to different types of search tasks. Our results contribute a novel framework for extracting useful information from real-time streams of social-media content that will be used in the design of a future retrieval system. Read more . . .

This is the meat:

Results

Tables 3 and 4 give us an overview of the codes, grouped by category, that were derived from the data, for both the useful and not-useful collections, respectively. We saw four key reasons where the content of the tweet was directly useful. Some contained facts (e.g. times or prices) or increasingly common knowledge (e.g. problems with the iPhone). Others contained direct recommendations, or relayed insights from personal experiences. We also saw two types of tweets that the user found to be amenable, ones that were funny and ones that shared the searcher’s perspective (e.g. Apple products are good or bad). We also saw two codes that focused on whether tweets were geographically or temporally still relevant (e.g. tweets in British prices). We also saw a key theme of trust, where users reported approving of trusted twitter accounts and recognising trustable avatars for those accounts. Also, links to authoritative or trustworthy websites were frequently recognised. Other links were also important, whether they provided more detailed information, rich media, or services (e.g. buying tickets).

There were also five key reasons that the content of tweets was not useful for the searcher. First tweets were frequently vague or introspective (for the author), or were quite directly not relevant by topic. While some tweets showed potential, it was easy for tweets to be too technical for the reader (containing jargon) or to contain errors (e.g. malformed URLs). There were 3 other reasons for tweets to be badly constructed: containing dead links, spam-style content, and being in a foreign language.