No Clicks, No Problem: Using Cursor Movements
to Understand and Improve Search
Jeff Huang
Information School
University of Washington
chi@jeffhuang.com
Ryen W. White
Microsoft Research
Redmond, WA 98052
ryenw@microsoft.com
Susan Dumais
Microsoft Research
Redmond, WA 98052
sdumais@microsoft.com

ABSTRACT
Understanding how people interact with search engines is
important in improving search quality. Web search engines
typically analyze queries and clicked results, but these ac-
tions provide limited signals regarding search interaction.
Laboratory studies often use richer methods such as gaze
tracking, but this is impractical at Web scale. In this paper,
we examine mouse cursor behavior on search engine results
pages (SERPs), including not only clicks but also cursor
movements and hovers over different page regions. We: (i)
report an eye-tracking study showing that cursor position is
closely related to eye gaze, especially on SERPs; (ii) pre-
sent a scalable approach to capture cursor movements, and
an analysis of search result examination behavior evident in
these large-scale cursor data; and (iii) describe two applica-
tions (estimating search result relevance and distinguishing
good from bad abandonment) that demonstrate the value of
capturing cursor data. Our findings help us better under-
stand how searchers use cursors on SERPs and can help
design more effective search systems. Our scalable cursor
tracking method may also be useful in non-search settings.
Author Keywords
Cursor movements, clicks, implicit feedback, Web search.
ACM Classification Keywords
H.3.3 [Information Storage and Retrieval]: Information
Search and Retrieval–selection process, relevance feedback
General Terms
Experimentation, Human Factors, Measurement.
INTRODUCTION
Understanding how people interact with Web sites is im-
portant in improving site design and the quality of services
offered. The Web provides unprecedented opportunities to
evaluate alternative design, interaction, and algorithmic
methods at scale and in situ with actual customers doing
their own tasks in their own environments [19]. Such stud-
ies typically involve measuring clicks which can be ob-
tained easily at scale. However, they fail to capture behav-
iors that do not lead to clicks (e.g., which items are attend-
ed to, in what order, etc.) or subjective impressions. Gaze-
tracking studies with participants present in the laboratory
can provide more detailed insights but on a smaller scale.
In this paper we consider how mouse movements, which
can be collected remotely on a large scale, can be used to
understand richer patterns of behavior.
We focus on understanding cursor activities in Web search
behavior. People conduct Web searches to satisfy infor-
mation needs. Their interaction with search engines begins
by issuing a search query, then reviewing the search engine
results page (SERP) to determine which, if any, results may
satisfy their need. In doing so, they may move their mouse
cursor around the page, hovering over and possibly clicking
on hyperlinks. Small-scale laboratory studies have ob-
served participants making many uses of the cursor on
SERPs beyond hyperlink clicking [1,21,25]. These uses
include moving the cursor as a reading aid, using it to mark
interesting results, using it to interact with controls on the
screen (e.g., buttons, scroll bars), or simply positioning the
cursor so that it does not occlude Web page content. How-
ever, studying such behaviors in small-scale laboratory
settings is limited in terms of what inferences can be made.
Tracking mouse cursor movements at scale can provide a
rich new source of behavioral information to understand,
model, and satisfy information needs. Recent research has
shown that cursor movements correlate with eye gaze
[6,13,25,26], and may therefore be an effective indicator of
user attention. We believe that cursor data, like click data
[18], can provide signals that reveal searcher intent and
may be useful in improving the search experience. Cursor
data can be used to complement click data in several ways.
First, cursor data can be captured for uncommon queries
where strong indicators of relevance such as result clicks
may occur less frequently or not at all. For example, ana-
lyzing click logs for a query that has been issued several
times but never clicked may provide limited relevance in-
formation, but cursor behavior on the SERP associated with
the query may provide insight about relevance. Second, in
cases of so-called good abandonment [20], where the con-
tent on the SERP satisfies the user’s information need di-
rectly, a search result click may be unnecessary. Thus the
lack of a click should not always be interpreted as a search
failure. Cursor behavior may help in distinguishing be-
tween good and bad search abandonment.

Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise,
or republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee.
CHI 2011, May 7–12, 2011, Vancouver, BC, Canada.
Copyright 2011 ACM 978-1-4503-0267-8/11/05....$10.00.

The research questions that we ask are: (i) to what extent
does gaze correlate with cursor behavior on SERPs and
non-SERPs? (ii) what does cursor behavior reveal about
search engine users’ result examination strategies, and how
does this relate to search result clicks and prior eye-
tracking research? and (iii) can we demonstrate useful ap-
plications of large-scale cursor data? Answers to these
questions help us determine the utility of cursor tracking at
scale, and ultimately inform search system design and im-
prove the experience for users of search engines.
RELATED WORK
One line of related research has explored the use of cursor
movements, clicks, and gaze as implicit indicators of inter-
est on Web pages. In early work, Goecks and Shavlik modi-
fied a Web browser to record themselves browsing hun-
dreds of Web pages [11]. They found that a neural network
could predict variables such as the amount of cursor activi-
ty on the SERP, which they considered surrogate measure-
ments of user interest. Claypool et al. [7] developed the
“curious browser,” a custom Web browser that recorded
activity from 75 students browsing over 2,500 Web pages.
They found that cursor travel time was a positive indicator
of a Web page’s relevance, but could only differentiate
highly irrelevant Web pages. Surprisingly, they also found
that the number of mouse clicks on a page did not correlate
with its relevance. Hijikata [15] used client-side logging to
monitor five subjects browsing a total of 120 Web pages.
They recorded actions such as text tracing and link pointing
using the cursor. The findings showed that these behaviors
were good indicators for interesting regions of the Web
page, around one-and-a-half times more effective than ru-
dimentary term matching between the query and regions of
the page. Shapira et al. [27] developed a special Web
browser and recorded cursor activity from a small number
of company employees browsing the Web. They found that
the ratio of mouse movement to reading time was a better
indicator of page quality than cursor travel distance and
overall length of time that users spend on a page.
In the search domain, Guo and Agichtein [12] captured
mouse movements using a modified browser toolbar and
found differences in cursor travel distances between infor-
mational and navigational queries. Furthermore, a decision
tree could classify the query type using cursor movements
more accurately than using clicks. Guo and Agichtein also
used interactions such as cursor movement, hovers, and
scrolling to accurately infer search intent and interest in
search results [13]. They focused on automatically identify-
ing a searcher’s research or purchase intent based on fea-
tures of the interaction. Buscher et al. investigated the use
of gaze tracking to predict salient regions of Web pages [2]
and the use of visual attention as implicit relevance feed-
back to personalize search [4].
Another line of research examined the relationship between
eye gaze and cursor positions. An early study by Chen et
al. [6] measured this relationship in Web browsing by re-
cording 100 gaze and cursor positions from five subjects
browsing the Web. They showed that the distance between
gaze and cursor was markedly shorter in regions of encoun-
tered pages to which users attended. Liu and Chung [21]
recorded cursor activity from 28 students browsing the
Web. They noticed patterns of cursor behaviors, including
reading by tracing text. Their algorithms were capable of
predicting users’ cursor behaviors with 79% accuracy.
More recent work has focused on the relationship between
cursor and gaze on search tasks. In a study involving 32
subjects performing 16 search tasks each [25,26], Rodden
et al. identified a strong alignment between cursor and gaze
positions. They found that the distance between cursor and
gaze positions was longer along the -axis than the -axis,
and was generally shorter when the cursor was placed over
the search results. Rodden et al. also observed four general
types of mouse behaviors: neglecting the cursor while read-
ing, using the cursor as a reading aid to follow text (either
horizontally or vertically), and using the cursor to mark
interesting results. Guo and Agichtein [14] reported similar
findings in a smaller study of ten subjects performing 20
search tasks each. Like Rodden et al., Guo and Agichtein
noticed that distances along the -axis tended to be longer
than the distances along the -axis. They could predict with
77% accuracy when gaze and cursor were strongly aligned
using cursor features.
The research presented in this paper extends previous work
in a number of ways. Our analysis of the cursor-gaze rela-
tionship (Study 1) involves more search tasks than prior
studies, compares SERP and post-SERP Web pages, and
confirms earlier results with a large study using the same
SERP layout that we use in the remainder of the paper.
More importantly, we develop a scalable approach to cap-
turing cursor data that enables us to analyze real user activi-
ty in a natural setting for more than 360 thousand searches
from an estimated 22 thousand searchers (Study 2). Finally,
using two case studies, we show how cursor data can sup-
plement click data on two search-related problems.
STUDY 1: GAZE-CURSOR RELATIONSHIP
We begin by replicating and extending prior laboratory
experiments on the relationship between gaze and cursor
activity using the same SERP layout deployed in our large-
scale cursor study (Study 2, see Figure 2). Study 1 also
involves more tasks and participants than prior laboratory
studies, and measures the relationship between gaze and
cursor position on SERP and on non-SERP pages.
Data
We used a Tobii x50 eye tracker with 50Hz tracking fre-
quency and 0.5° visual angle on a 1280 × 1024 resolution
17 inch monitor (96.42dpi) and 1040 × 996 resolution In-
ternet Explorer 7 browser. Cursor and gaze coordinates
were collected in an eye-tracking study of 38 participants
(21 female, 17 male) performing Web searches. Participants
were recruited from a user study pool. They ranged in age
between 26 and 60 years (mean = 45.5, = 8.2), and had a
wide variety of backgrounds and professions.