One of the most important features in a search engine is query
auto-completion (QAC). QAC is the first service through which users
interact with a search engine to input their search intent. In 2014,
global users of Yahoo search saved more than 50% of keystrokes when
submitting English queries by selecting QAC suggestions.
QAC provides and updates a query suggestion list based on each new
character typed by a user in the search box. The suggestion list is
ranked by considering different factors, such as most popular completion
(historical frequency counts from query logs), time (breaking news or
recent popular queries), location, context (user’s previous queries),
personalization (user’s profile), and click modeling (user’s past click
behaviors).

The aforementioned approaches use only certain relevance features and do
not fully take advantage of users’ preferences such as user-QAC
interactions. Suppose a user dwells on a suggestion list for a long time
without selecting the top-ranked query, it indicates that the user
intent might not be satisfied by the provided query suggestions. That
wealth of implicit negative feedback has not yet been fully exploited
for designing QAC models. Our findings suggest more accurate search
results when redesigning QAC to include a more general “(static)
relevance-(adaptive) implicit negative feedback” framework.