#TwitterSearch

Social networking sites are not just places to maintain relationships. They are also valuable information sources. Previous posts (Online QuestionAsking, Asking v. Searching) have discussed how people find information using social networks by asking questions. This post explores the other way people find information using social media: by searching preexisting social media content.  For example, if someone wanted to read a lot of snarky comments about the use of CamelCase in hashtags, they might go to Twitter and search for "#nowthatchersdead."

Despite the fact that people issue over a billion queries to Twitter every day, we don’t actually know much about their behavior when they do so. Of course, we didn’t used to know much about people’s behavior during web search either. Much of what we take for granted about web search (e.g., that people issue one or two word queries, don’t use advanced operators, and often search to navigate to websites) was once surprising because our frame of reference was library or legal search. For example, Jansen et al. observed in 1998 that many Excite users misused the Boolean AND operator by not capitalizing it, when anyone looking at the logs today would assume the lowercase “and” was intended to be used as a conjunction. Of course, fifteen years ago 50,000 web queries seemed like a large dataset to study – that represents just a few seconds of traffic on a major search engine now!

So when we look at microblog search behavior, we see differences that make sense in the context of Twitter but are somewhat surprising coming from web search. These include:

• The information needs we bring to Twitter are very different than those we bring to web search. Twitter is used primarily to find social content and events, while web search is used to find facts or navigate to sites.
• Twitter queries are used to monitor a topic, while web queries are used to learn about a topic. Individuals on Twitter issue exactly the same query over and over again just to see what is new. What’s more, the topics that we monitor are all the same. Everyone is monitoring #nowthatchersdead today, and tomorrow we will all monitor something new.
• People can use specialized syntax when they search. A strong lesson from web search is that nobody uses advanced search operators. I know I rarely do, despite how intimately I work with search engines. But hashtags and @references to people are common in the Twitter search logs as a way to reference particular memes or people. Exposing the operators in the content creation as well as the search interface helps users gain familiarity with the specialized syntax.
• While people use web search to return to previously viewed content, Twitter search is almost never used to find previously viewed tweets. This is probably because it is really hard to do. Think of a tweet you saw yesterday and now try to find it. Hard, right? The ability to express context and meta-data in our Twitter queries might help with this.

By studying and understanding these differences, we can not only improve people’s microblog search experience, but also identify opportunities where lessons from microblog and web search can be combined to create a better overall search experience. 


Related papers:
J. Teevan, D. Ramage and M.R. Morris. #TwitterSearch: A Comparison of Microblog Search and Web Search. WSDM 2011.

S. Kairam, M.R. Morris, J. Teevan, D. Liebling and S.T. Dumais. Leveraging Social Media to Support Search over Trending Events. ICWSM 2013.