DuckDuckGo Search Statistics
Axibase has been tracking DuckDuckGo search statistics. It is a very interesting case as DuckDuckGo shares its core operational metric with minimal delay.
DuckDuckGo is a new search engine that has a focus on privacy. With DuckDuckGo there are no personalized search results based on location or past click behavior and search history. None of that information is gathered or stored, which means DuckDuckGo always shows users the same search results if the search terms are the same. DuckDuckGo also emphasizes information from the best sources rather than the highest number of sources, which often results in higher quality and more relevant search results.
With increased concerns over privacy due to recent world events, DuckDuckGo has been gaining ground on its competitors. In the last two years, DuckDuckGo has experienced a huge spike in popularity, partially due in result to the United States government surveillance program that came to light. The service has grown by over 600% since then. Now DuckDuckGo is becoming a mainstream search engine, primarily due to its focus on user privacy. Read more about the story surrounding the NSA surveillance program here.
You can learn more details about DuckDuckGo in this article.
The main DuckDuckGo metric that the company tracks is the amount of search queries per day: https://duckduckgo.com/traffic.html
Search queries (direct queries) are driven by humans. This activity at a large scale is subject to common patterns such as a strong weekly cycle, which means patterns can be forecast with a high degree of accuracy.
The Holt-Winters algorithm, which is built into the Axibase Time Series Database, is designed to detect trends over periods of time. ATSD iterates various Holt-Winters algorithm parameters to arrive at a combination of settings that produces the most accurate forecast.
We used Axibase Time Series Database forecasting features to forecast the amount of daily search queries in DuckDuckGo. DuckDuckGo was approaching the 10 million search queries per day barrier, and ATSD was able to forecast the exact day when DuckDuckGo surpassed this threshold for the first time. Read more about this event here.
For the above forecast, automatic parameters were set in the Forecast Engine, allowing the system to match the best parameters to historical data. However, if the Holt-Winters algorithm was to run using static parameters, the system would not be able to forecast abnormal events such as news publications or one-time events (such as NSA PRISM disclosures).
This is where daily recomputation of forecasts in ATSD shows its strengths. Recomputation allows ATSD to determine new periods and new trends continuously. New forecasts are generated every night, taking into account all the recent developments and trends that have occurred in the time series. This kind of approach results in high forecast accuracy and dynamic response to abnormal events and new trends.
In ATSD’s forecasting engine it is also possible to exclude special events using Calendars to minimize their impact on stationary periods and trends.
Visit axibase.com to learn more about the Axibase Time Series Database, Data Visualization, Data Forecasting, and Internet of Things.
