Professional Documents
Culture Documents
First dwell time on Query SERPs (sec) 6.87 5.83 12.65 10.06 0.009 10.14 5.07 10.64 6.46 0.919
Mean dwell time on content pages (sec) 16.54 11.83 22.62 22.04 0.459 13.65 4.44 14.89 14.37 0.150
Mean dwell time on SERPs (sec) 8.70 5.37 12.14 7.31 0.042 10.27 4.38 10.19 3.82 0.834
Total dwell time on content pages per query (sec) 58.02 61.98 41.21 53.19 0.290 62.58 66.78 36.44 46.23 0.005
Total dwell time on SERPs per query (sec) 22.73 15.35 25.89 16.78 0.410 30.03 13.44 23.52 10.24 0.036
Ratio of dwell time on content pages 0.44 0.32 0.39 0.30 0.411 0.56 0.14 0.45 0.19 0.020
Ratio of dwell time on SERPs 0.36 0.28 0.38 0.23 0.323 0.36 0.15 0.42 0.18 0.222
Average query interval time (sec) 94.04 78.01 87.13 68.19 0.895 99.29 78.12 72.3 60.42 0.012
Average session time (sec) N/A N/A N/A N/A N/A 632.09 421.85 218.46 71.8 0.000
Number of queries per session N/A N/A N/A N/A N/A 7.52 4.77 4.07 2.02 0.006
Participants had roughly the same query interval, total However, after changing queries for several times and
dwell time on SERPs, and total dwell time on content pages when they finished a complete task session, users with time
at this time point. No significant effect has been found of constraint had significant shorter query interval time on
time constraint on the ratio of total dwell time on SERPs to average (72.3s vs 99.29s) and spent rather shorter time on
query interval, nor the ratio of total dwell time on content both content pages (36.44s vs 62.58s) and SERPs (23.53s
pages to query interval. vs 30.03s) per query than those with no such constraint.
Aside from this, while users in TC also spent significant
Whole Session smaller proportion of their total dwell time on content
As showed in Table 1, thirteen of the eighteen measures pages than in NTC (56% vs 45%), there was no significant
had significant differences between users with different difference in the proportion of dwell time on SERPs (42%
time conditions when the behavioral measures were vs 36%). It can be inferred that users’ dwelling strategies
calculated after the whole session completed. Participants changed over a task session.
searched with time constraint generated significantly fewer
queries (4.07 vs 7.52) for each task than their counterparts DISCUSSIONS
experienced no time limit. As for each query, they also We examined the effect of time constraint on users’ search
viewed significantly fewer content pages (2.93 vs 5.14), interactions at two time-stages during search episode in this
fewer unique content pages (1.83 vs 3.02), fewer SERPs study: the first query interval (after searchers issued the
(2.56 vs 3.15), and fewer unique SERPs (1.12 vs 1.21). first query before they issued the second query) and the end
point when users have completed the whole task session.
With respect to time-related measures, there was no
difference in first dwell time on unique content pages or
Comparison of search strategy in the first query
mean dwell time on content pages. Moreover, the once
interval
existed differences of their first dwell time on query SERPs
Our results demonstrated that time constraint did not have
and mean dwell time on valid SERPs during the first query
significant effect on most of the behavioral measures in the
interval were no longer significant at the end point.
first query interval. Particularly, under the two time
conditions, users’ first query interval time was similar, and during the whole session. In general, for the whole search
the number of content pages and SERPs they viewed session, no significant difference has been found of time
during this time period were fairly similar. Considering the constraint on the average dwell time of either content pages
number of content pages and SERPs were in a relatively or SERPs with or without time constraint, but users did
low level (unique content pages: 2.93 vs. 1.76; unique view significant fewer content pages and SERPs for each
SERPs: 1.10 vs. 1.03), it is reasonable for us to conjecture query when there was time constraint. Such results indicate
that searchers took their first queries as the first trials of the that when given time constraint, searchers did not speed up
search system no matter with or without time constraint. It their reading process on either content pages or SERPs;
seems that searchers only explore the search results of their instead, they just reduced the number of content pages and
first queries briefly and would then quickly reformulate SERPs to read. The result also demonstrated that searchers
their queries. were relatively more selective in TC conditions.
Besides the common pattern, participants with and without This result is inconsistent with Crescenzi [9], where they
time constraint had significantly different first dwell time found participants who were given task time limits spent
on query SERP in their first query interval. When there was significant less time on both SERPs and documents. But
a time constraint, searchers had significant longer first they did not explain how the two variables were defined
dwell time on query SERP; that means it took them longer and calculated, so it is not clear whether the results from
time in viewing the search result page and made their first the two studies are comparable. Another difference is that
decision to leave it. Also, searchers spent longer dwell our experiment required participants to record their answers
time on SERPs on average when there was time constraint. in a notebook during the search episode, so participants
It was very likely that users who experienced a time may have to read content pages more carefully, and then
constraint were more selectively and cautiously in viewing cast an influence on their dwell time on content pages.
SERPs and deciding whether to select certain content pages
Another interesting pattern is that searchers had a
to view or try another query. For the dwell time on content
significantly smaller percentage of time spent on content
pages, although no significant difference was found, similar
pages when given time constraint, but they had similar
to dwell time on SERPs, searchers with time constraint still
percentage of time spent on SERPs. Therefore, we could
tended to spent longer time in reading each content page on
infer that searchers had relatively more percentage of time
average and in making their first decision to leave each
spent on the search homepage. Since in our log, the dwell
unique content page.
time on search homepages was the duration of time from
In general, when there was no time limit, users’ search the search homepage was fully loaded till the time when the
strategy in the first query interval were likely to be served SERP for that query was shown, the longer dwell time on
as search trials to get familiar with the search systems and search homepage indicated longer time in formulating
related information in the data collection; however, users queries. It seems that when given time constraint, searchers
who experienced a time constraint behaved more were more cautious or stressed in formulating queries, and
selectively in choosing content pages they viewed and read they might expect to formulate a query that could lead them
each content page a bit more carefully. The results also to good search results directly. Another possibility is that
demonstrated that searchers read search result lists for they were more likely to make typos when there was a time
longer time and then select only one or two content pages constraint, which delayed their time on the search
to read when there was time constraint; while for those homepages. Future research on mouse clicks and
without time limit, they were more relax and were not as keystrokes is needed to better understand such phenomenon.
selective as their counterparts and they tended to make
quick decisions on which content pages to view on the first Comparisons between the two time stages
search result page. This is a hypothetical explanation for We are also interested to see how searchers changed their
searchers’ interactions. Future research would examine search strategies from the beginning to the end of the
search logs in details to understand content of the pages search session. The comparisons showed that when there
that users viewed in different time conditions, or to make was time constraint, searchers reduced their first/mean
use of other equipment, e.g. eye tracker to compare the dwell time on both content pages and SERPs from the first
reading patterns between two time conditions. query interval to the end of the search session (showed in
left side of Figure1). When there was no time limit,
Comparison of search strategies during the whole searchers also reduced their first/mean dwell time on
session content pages, but they increased their first/mean dwell
We also examined users’ behavioral measures at the end of time on SERPs from the first query interval to the end
the search session to describe their general search strategies (shown in right side of Figure 1).
Figure 1 Comparisons of dwell-related measures in TC and NTC at two time stages
In general, as the search processed, users became more comparing to users may encounter in their real life. Future
familiar with the search topic, which allowed them to make research may consider conducting naturalistic study and
a faster decision on the usefulness of the content, and they include higher level type of tasks to investigate the effect of
tended to accelerate their reading process within the search time constraints on search interactions and search
session. This is a consistent pattern no matter whether there experience in real life.
was a time limit or not. However, when it comes to SERPs,
Another limitation is that time constraint was designed as
time limit had different effects on searchers interactions.
within-subject in our experiment. The rationale is to
When searching without time limit, users increased their
compare the difference of search strategies with and
first and mean dwell time on SERPs from the first query
without time constraint and rule out the possible individual
interval to the end of the search session; however, when
differences. We acknowledge that in the within-subject
searching with time limit, users slightly reduced their first
design, one condition the participant experienced may
and mean dwell time on SERPs. On average, when the
influence how they perform in the second condition. In
whole session was considered, searchers with and without
order to eliminate such effect, we asked participants to take
time constraint had similar dwell time on SERPs. Such
a five-minute break after they have finished two search
results indicate that time constraint had more influence on
tasks in one condition. The participants were also told that
users’ dwell time on SERPs during the first query
the goal of this experiment was not to test their search skills
interval—users were more cautious and selective on the
with and without time constraint, but to investigate how the
SERPs for the first query when there was time constraint,
search system could provide assistant for information
but they were more relaxed and might browse search result
search in different contexts. The order of search tasks and
lists orderly when there was no time constraint. As the
time conditions was systematically balanced using 2x2
search processed, users’ interactions became similar, and
Graeco-Latin Square design, to rule out the possible order
the only difference was that users with time constraint
effect.
would select a small number of pages to read.
This study examined how users interacted on search result
Limitations and Future work pages and content pages during query interval, including
We acknowledge the limitation of this study. The time the first and mean dwell time on different pages, the
constraint we manipulated was five-minute limit for each number of pages viewed and other interactions during
search task. This is somewhat artificial. In real life, query intervals. The results demonstrated users might
searchers sometimes had time constraint or very limited change their search strategies when given severe time
time before deadline to accomplish some search tasks, but constraint, especially in their first query interval. Such
may not be five-minute specifically. The reason we set results also indicate that users’ interactions during the first
five-minute limit was to induce severe time constraint in query interval could reflect their time context. This result is
searching, since some of previous studies did not find much consistent in previous studies (e.g. [21]), which
effect of time constraint when the time limit was set as demonstrated the beginning of the search could indicate
“high” or “moderate” availability. Also, the tasks were of users’ search contexts. Future research may examine users’
only two types in our study: fact finding and understanding, cursor movements and eye gaze movements to fully
which equivalent to the two lower levels of Anderson and understand how users read or browse the search results, and
Krathwohl’s Taxonomy of Learning. This is a limited range further verify some of the explanations in this study.
CONCLUSION international conference on Web search and data
mining (pp. 373-382). ACM.
This study examined the effects of time constraints on
searchers’ information search strategies at two search [7] Byström, K., & Järvelin, K. (1995). Task complexity
stages (first query interval and end point). A user affects information seeking and use. IP&M, 31(2),
experiment was conducted, in which participants were 191-213.
asked to search for two tasks with different time conditions. [8] Crescenzi, A., Capra, R., & Arguello, J. (2013). Time
The results showed users did employ different search Pressure, User Satisfaction and Task Difficulty.
strategies when searching with and without time constraint, Proceedings of ASIS&T 13’.
and their search strategies changed over time within the [9] Crescenzi,A., Kelly, D., and Azzopardi, L. (2015).
search session. Generally, when there was no time Time Pressure and System Delays in Information
constraint, users often took the first query as a trial of Search. SIGIR '15.
search and tended to employ economic-style search [10] Crescenzi,A., Kelly, D., and Azzopardi, L. (2016).
strategy. They spent short time on SERPs, viewed one/two Impacts of Time Constraints and System Delays on
content pages before reformulating the query. When given User Experience. Proceedings of CHIIR’16.
time constraint, users became more selective and cautious [11] Dumais, S. T., Buscher, G., & Cutrell, E. (2010,
in examining the search results, and they spent longer time August). Individual differences in gaze patterns for
on SERPs but only selected one/two content pages to read web search. In Proceedings of the third symposium on
during the first query interval. As the search processed, Information interaction in context (pp. 185-194). ACM.
users with and without constraint both employed similar
[12] Ford, N., Wood, F., & Walsh, C. (1994). Cognitive
search strategy—they became similar in their reading
styles and searching. Online and CD-Rom Review,
patterns on SERPs and content pages, but users with time
18(2), 79-86.
constraint were still selective and only read one/two content
pages per query. Future work will expand the behavioral [13] Ford, N., Miller, D., & Moss, N. (2001). The role of
features set to include mouse movements and keyboard individual differences in Internet searching: an
activities to fully understand users’ search strategies under empirical study. Journal of the American Society for
time constraint. The findings of this study have Information Science and technology, 52(12), 1049-
implications for search system design to assist searchers 1066.
under time constraint and help them search more [14] Fujikawa, K., Joho, H., & Nakayama, S. I. (2012).
effectively and efficiently. Constraint can affect human perception, behaviour,
and performance of search. In The Outreach of Digital
ACKNOWLEDGEMENT Libraries: A Globalized Resource Network (pp. 39-48).
This research is funded by NSFC #71303015. Springer Berlin Heidelberg.
[15] Gwizdka, J., & Spence, I. (2006). What can searching
REFERENCE behavior tell us about the difficulty of information
[1] Aula, A., Majaranta, P., & Räihä, K. J. (2005). Eye- tasks? A study of Web navigation. In Proceedings of
tracking reveals the personal styles for search result the American Society for Information Science and
evaluation. In Human-Computer Interaction- Technology, 43(1), 1-22.
INTERACT 2005 (pp. 1058-1061). Springer Berlin [16] Heinström, J. (2006). Broad exploration or precise
Heidelberg. specificity: Two basic information seeking patterns
[2] Anderson, L. W. & Krathwohl, D. A. (2001). A among students. Journal of the American Society for
taxonomy for learning, teaching and assessing: A Information Science and Technology, 57(11), 1440-
revision of Bloom’s taxonomy of educational 1450.
objectives. New York: Longman.
[17] Liu, C., Liu, J., & Belkin, N. J. (2014, November).
[3] Bates, M.J. (1979). Information search tactics. Journal Predicting search task difficulty at different search
of the American Society for Information Science, stages. In Proceedings of the 23rd ACM International
30(4):205–214. Conference on Conference on Information and
[4] Bhavnani, S. K. (2002). Important cognitive Knowledge Management (pp. 569-578). ACM.
components of domain-specific search knowledge. Nist [18] Liu, J., Liu, C., Cole, M., Belkin, N. J., & Zhang, X.
Special Publication. 1001, 48109-1092. (2012, October). Exploring and predicting search task
[5] Borlund, P. (2003). The IIR evaluation model: A difficulty. In Proceedings of the 21st ACM
framework for evaluation of interactive information international conference on Information and
retrieval systems. Information Research, 8(3): 1–34. knowledge management (pp. 1313-1322). ACM.
[6] Buscher, G., White, R. W., Dumais, S., & Huang, J. [19] Jiang, J., He, D., & Allan, J. (2014, July). Searching,
(2012, February). Large-scale analysis of individual browsing, and clicking in a search session: changes in
and task differences in search result page examination user behavior by task and over time. In Proceedings of
strategies. In Proceedings of the fifth ACM the 37th international ACM SIGIR conference on
Research & development in information retrieval (pp. [26] Thatcher, A. (2008). Web search strategies: the
607-616). ACM influence of web experience and task type. Information
[20] Kuhlthau, C. C. (1991). Inside the search process: Processing & Management, 44(44), 1308-1329.
information seeking from the user's perspective. [27] Topi, H., Valacich, J. S., & Hoffer, J. A. (2005). The
Journal of the American Society for Information effects of task complexity and time availability
Science, 42(5), 361–371. limitations on human performance in database query
[21] Liu, C., Liu, J., & Belkin, N. J. (2014). Predicting tasks. International Journal of Human-Computer
search task difficulty at different search stages. In Studies, 62(3), 349-379
Proceedings of the 23rd ACM International [28] Vakkari, P. (2001). A theory of the task-based
Conference on Conference on Information and information retrieval. Journal of Documentation, 57(1),
Knowledge Management (pp. 569-578). ACM. 44-60.
[22] Liu, J., & Belkin, N. J. (2014). Personalizing [29] Vakkari, P., & Hakala, N. (2000). Changes in
information retrieval for multi-session tasks: relevance criteria and problem stages in task
examining the roles of task stage, task type, and topic performance. Journal of Documentation, 56(5), 540-
knowledge on the interpretation of dwell time as an 562.
indicator of document usefulness. JASIST, 66(1), 58– [30] Weenig, M. W., & Maarleveld, M. (2002). The impact
81. of time constraint on information search strategies in
[23] Liu, J., Gwizdka, J., Liu, C., & Belkin, N. J. (2010). complex choice tasks. Journal of Economic
Predicting task difficulty for different task types. Psychology, 23(6), 689-702.
Proceedings of the American Society for Information [31] White, R. W., Ruthven, I., & Jose, J. M. (2005). A
Science & Technology, 47(1), 1–10. study of factors affecting the utility of implicit
[24] Savolainen, R. (2006). Time as a context of relevance feedback. In SIGIR '05. ACM, New York,
information seeking. Library & Information Science NY, USA, 35-42.
Research, 28(1), 110–127. [32] Wu, W-C., Kelly, D., Edwards, A. and Arguello, J
[25] Slone, D. J. (2007). The impact of time constraints on (2012). Grannies, Tanning Beds, Tattoos and
Internet and Web use. Journal of the American Society NASCAR: Evaluation of Search Tasks with Varying
for Information Science and Technology, 58(4), 508- Levels of Cognitive Complexity. In Proceedings of the
517. 4th Information Interaction In Context Symposium
(IIiX'12).