This application displays political issues and concerns by aggregating anonymized queries submitted to the U.S. version of the Yahoo! web search engine. For a given week, political queries trending that week are identified.
Assigning leaning to a query
We look into all the queries landing on hundreds of top U.S. political blogs annotated with a political leaning.* Subsequently we assign a leaning to each query, proportional to the number of times it lands on a blog with a particular leaning. The resulting data set consists of classified queries, made by hundreds of thousands of users. Queries are ranked such that the most ‘polarizing’ queries will be shown first.
Retrieval of relevant political queries
Several steps ensure that the queries are considered relevant by many users. After removing navigational queries and normalizing for spikes in the query volume, we filter the data to only retain queries made by multiple users, resulting in a click-through to at least three political blogs. Besides, we remove queries with a very high query volume which only result in click-throughs to very few political blogs. This last step prevents a particular blog setting the agenda.
How not to interpret the results
Q: Does your tool show that a particular political side only searches for the displayed queries?
A: No, it does not. The tool emphasizes polarizing queries, which (almost) exclusively lead to click-throughs to blogs of a particular leaning. Many queries have a high volume and lead to click-throughs to blogs of both leanings. Such queries show up further down in the ranking. Also note that the notion of ‘trending’ in a given week is based on volume deviations from the norm for this query. A query that every week has a large volume would not be listed as trending.
Queries are ranked based on their assigned proportion to a particular leaning. There are rankings for the left and right side of the political spectrum. Clicking a query opens a new window containing the current search results for that query, restricted to the blogs of a particular leaning.
Apart from a global ranking based on a total of nine months' data, specific queries can be searched. If a match is found, all queries containing the search will be shown and ranked. When trying ‘obama’ one will see queries like ‘obama accomplishments’ to be more on the left-side of the political spectrum, and ‘obamacare’ to be more on the right side of the political spectrum. Note that the order of the terms (or prefixes) matters when issuing multi-term queries.
Note that very similar queries might be displayed in a ranking. This is because in the current version of the application we have not done any grouping of queries, i.e. we left all queries intact.
To provide some context for each query we have mapped them to the Wikipedia articles to which they are most relevant. By clicking the "W" next to the query the relevant Wikipedia article is shown, together with the categories to which the article belongs.
To guarantee the privacy of our users, queries containing personally identifiable information were removed. Furthermore, only queries which are submitted by multiple users and with a click-through to multiple blogs are shown.
The relation to the offline
In order to ground our methodology, we validated our data set with voting polls of the 2010 U.S. House of Representative elections. It turns out that people visiting blogs of a particular leaning, say right, are more likely than random chance to live in a ZIP code that voted accordingly, say Republican.
Looking into the queries resulting in a click-through to political blogs is believed to produce an insightful description of the nation's political preferences, or political mood. This tool will not predict who is likely to win the next presidential elections, but will show which issues are deemed important by different sides of the political spectrum by looking into the interplay between searchers and political bloggers. As such it does not only describe what political bloggers deem important, but also what users find interesting to look into.
By looking into collective search histories, Political Search Trends thus allows for detailed insights into topical political concerns. Moreover, highly partisan (groups of) queries can be sifted out, e.g. queries containing “obamacare”, which were likely to land on right-leaning blogs.
* The blogs and classifications used are obtained by merging those described in the article "A Tale of Two Blogospheres: Discursive Practices on the Left and Right" by Yochai Benkler and Aaron Shaw with the blogs listed in the Wonkosphere Blog Directory.
Political Search Trends by Yahoo! Labs detects political queries trending in a given week and displays the proportion of times a search query results in a click-through on a political blog of a particular leaning. Blue stands for left-leaning blogs, and red for right-leaning blogs. The ranking shows polarizing queries first, rather than high volume ones. For more information see the about page.