While there is an increasing amount of interest in evaluating and supporting longer \search sessions", the majority of research has focused on analysing large volumes of logs and dividing sessions according to obvious gaps between entries. Although such approaches have produced interesting insights into some di erent types of longer sessions, this paper describes the early results of an investigation into sessions as experienced by the searcher. During interviews, participants reviewed their own search histories, presented their views of \sessions", and discussed their actual sessions. We present preliminary ndings around a) how users understand sessions, b) how these sessions are characterised and c) how sessions relate to each other temporally.
Information Retrieval (IR) specialists are becoming
increasingly concerned with users who continue to search
beyond a few queries or a few minutes1. Although
Information Retrieval, and even Interactive IR, evaluations are well
known, research is recognising situations where people
continue to search after nding seemingly useful results [
Most research into extended search sessions, described in
detail below, has focused on analysing search engine logs [
Presented at EuroHCIR2013. Copyright c 2013 for the individual papers by the papers’ authors. Copying permitted only for private and academic purposes. This volume is published and copyrighted by its editors. however our research has focused on using such methods to better understand real extended search sessions. This paper begins by rst summarising literature on sessions and then describes our research methods and preliminary ndings about extended search sessions. 2.
Although investigations into web sessions can be dated
back to around 20 years ago (e.g. [
In 2006, Spink et al [
A clear de nition has also been cited as an important
challenge in other research. While focusing on \revisitation"
behaviour, Jhaveri and Raiha [
When focusing on searching, rather than web sessions,
some use the concept of a \query session". Nettleton et al
de ned a query session as at least one query made to a
search engine, together with the results which were clicked
on and other user behaviours as well [
To summarise the di erent approaches used to de ne
sessions, Jansen et al. provided a summary of the three most
representative strategies [
The methods summarised in Table 1 are primarily focused
on temporal and topical boundaries, but other research has
shown clear challenges to these strategies. Mackay et al, in
2008, examined tasks that frequently occur as multi-session
tasks, where something thematically consistent occurs over
multiple sessions [
To understand and characterise real extended search
sessions, we employed similar interview methods to Sellen et
al. [
Participants began by providing their web history and they were advised to edit their history in advance should they wish to keep some logged activities private2. These logs were gathered by importing their search histories to Firefox (if not already there), and creating an XML export using \History Export 0.4"3. This log was then structured and preliminarily processed using a) automatic methods to nd search URLs, and b) manual investigation to nd possible sessions to discuss in the interview. After providing demographic information, participants spent around 20 minutes examining the structured printout of their history, using a pen to mark sessions. These sessions, unless duplicates of prior sessions, were written onto separate cards for later sorting until around 20 cards were produced. Each card had a number, a title, activity purpose, included history items from the history list and also whether it has been completed successfully or not; an example is shown in Figure 1.
The remainder of the interview involved rst open, and
then closed card sorting. Open card sorting allowed the
participants to classify and group the sessions according to
their own ideas, whilst closed card sorting allowed us to
make sure the following dimensions were considered:
purpose, for whom, with whom, location, duration, di culty,
importance, frequency, and priority. This exercise was to
help explore the session feature in a more detailed way. For
example, studying frequency helps to nd out the most
frequent sessions and elicit the pattern of user's web activity.
2Although this means we have likely missed common search
sessions, like the lengthy adult sessions observed by Bailey
et al [
This paper describes our preliminary analysis of the rst phase of the study, which involved 11 interviews. Phase two, which is still under way, involves a slightly re ned methodology to capture more information about topics that emerged from the initial analysis described below. A more comprehensive analysis of both phases will be published later.
Based on our preliminary investigation, some potentially
interesting results relating to perceived duration, time of
day, and use of queries were found. We considered each of
these below according to two aspects: activity goal and
activity context. For activity goal, we used Sellen et al's [
There were 216 sessions in total and 19.6 sessions per person have been studied thus far, as shown as Table 2. Amongst these, 94 were longer than 5 minutes, 99 featured search and only 9 sessions were unsuccessful.
All participants mentioned that activities with the same purpose and subject should be grouped into one session, as shown in Table 3. In addition, 8 of the 11 suggested that similar tasks happened in di erent time periods should be classi ed as a single session, rather than them being temporally connected. Some participants said that they always kept the browser windows open when doing long-term tasks. Finally, 1 participant advised that they care about the emotion involved within these web activities, even when they were doing the same task, such as \buying a pair shoes". In particular, this participant indicated that one topically consistent session should be divided between two disappointingly unproductive and excitingly productive phases.
As duration is one of the targeted dimensions, all participants were asked for their own de nition of what constitutes a \long session". 45% of participants de ned the session where the duration is more than 5 minutes, whereas 27% went with over 30 minutes, 18% more than 1 hour, and 1 participant chose over 2 hours.
Because participants rst de ned what they considered to be a long session, and then later sorted their sessions into length categories, we investigated the di erence between sessions that met their de nition of long, and ones they remembered as being long during the card sorts. Participants frequently grouped `de ned short' sessions as long and vice-versa. Consequently, we investigated both `overestimated' and `under-estimated' sessions in addition to `dened long', `long', `actual long', `de ned short`, `short', and `actual short' as given in Table 5.
Firstly, considering activity goals given in Table 6, the number of `information-gathering' sessions de ned as long was 5 times as that of those `de ned short', as was the same with `browsing'. On the contrary, the number of ` nding' sessions de ned as short was 1.5 times the number de ned as long. Overall, nearly 70% of ` nding', 42% of `informationgathering', 60.7% of `browsing', 50% of `transaction', and 85.5% of `email' sessions de ned as long were overestimated by users. Moreover, under-estimation occurred with ` nding', `information-gathering', and `housekeeping' although over-estimation was more frequent with ` nding', `browsing', `communication', and `email' sessions.
Figure 3 shows that most the `information-gathering', ` nding' and `housekeeping' sessions seem to occur between 10:00 and 16:00 whilst more `browsing', `email', and `communication' activities were done between 22:00 and 0:00, which was labelled \before bed time". Additionally, there is a peak around 14:00, in which more ` nding' and `informationgathering' happened rather than other kinds of sessions. Finally, at 23:00, general `browsing' is most prevalent.
Figure 4 shows that most of the `serious-leisure' sessions occurred between 18:00 and 22:00. Most of the `work' activities happened between 11:00 and 18:00, which seems to t in within a typical working day. In the time `before bed', the most frequent activity is `casual-leisure'. 4.4
In Figure 5 below, sessions with more search queries tend to be classi ed as `de ned long', `long', and `actual long' than those with fewer queries. An interesting observation is that what the user de ned as a long session features a relatively low average number of search queries compared with `long' and `actual long' sessions. Equally, sessions de ned as `short' by the user actually feature relatively more queries compared to `short' and `actual short'. This may indicate that the user did not consider the number of queries performed when de ning the duration of sessions and failed to realise the e ect of this behaviour.
Although this paper only describes a preliminary analysis of over 200 sessions from 11 participants, we have begun to see some potentially interesting early ndings. Initially, participants varied greatly in their opinions about their own sessions, with some matching topical divisions, some temporal divisions, and some a combination of the two. The majority of participants judged \long sessions" as being longer than 5 minutes, but many had inaccurate recollections of the length of sessions. Long sessions were typically a mix of casual and serious leisure that often involved information gathering and browsing behaviour, while the majority of work related sessions were typically short. We also noticed that some of these activities may also be related to certain times of the day. All of the ndings will be further explored after phase two of the study, but early insights suggest that real extended search sessions could be more accurately modelled based on additional factors such as: time of day, activity goal, activity context, and number of queries. 6.