nearly as common as in the most complex tasks. The complexity of information searching does not directly relate to the complexity of work task; their connection is more complicated. This is easily missed in user experiments. Kumpulainen’s (2014) findings show that in routine tasks (simple tasks) searching is actually more complex than in semi-complex tasks, but complex tasks differ clearly from other task types, since, as a rule, searching is explorative in them. Our findings are similar.
Within our work task types, intellectual tasks differ from other tasks because in them even a fourth of search tasks have a developing process and only above a third of processes are single. Interestingly, list processes also form a fourth of search tasks in intellectual work tasks. Support tasks resemble intellectual tasks, because in them, search tasks are quite often (over a fifth) developing, and single processes are rarer than in communication or editing tasks. Thus, work tasks may be connected to searching, but one-to-one connection cannot be expected. It is not self-evident, for example, that a work task that should, by definition, have well- specified performance process, would include only well-specified search tasks, as well.
Information resource use and information needs. Paper IV showed that information need complexity and search process complexity are positively correlated. To some extent, information needs and information resource types are also connected, as analysed in Paper III. Before task performance, the participants reported getting mainly known items from organisational information systems, whereas email, other human sources and the Web were used for finding topical information. In materialised use, the participants’ reports considering email use changed from information aggregates to known items which indicates that actually a single mail or file was enough to satisfy an initially topical information need.
rather typical assigned search tasks. They found that there were over 5 queries per search task, and typically queries were long and even forming full sentences (Wu &
Cai, 2016).
Excluding the issues of age that were in the focus in Wu and Cai’s (2016) study, it may still tell something important about the differences between real-life and experimental searching. Also Aula et al. (2010) found in their experiment that on average, there are 7 queries per search task and each query has on average 5 query terms. Rather multi-query search tasks and long queries in experimental settings have also been reported by Li and Hu (2013). In experiments, assigned search tasks are externally given, concrete snippets of text. Thus, the information need is already in textual, exact form and it can be returned to when searching for the right answer.
Furthermore, the search task exists in its own right, that is, it does not serve any other purposes than stimulating searching. Thus, the search task itself gets all the attention, and the assignment gives static understanding of what it is about. This is quite the opposite of what was found in the present study (however, please note that we analysed search tasks outside the Web, as well). As search tasks and information needs form intrinsically and quickly, they exist only in the head of the searcher. The immediate reaction is to spell a word or two in a search box and try out what can be found. It is likely that the participants have learnt that short queries work well enough. It is important to remember, that if the search task fails, there exist other ways to proceed in the work task the search task was intended to serve in the first place. The goal is not to succeed in the search task but in the work task. These issues may affect that searching for its own right (experiments) is more complex than searching as a part of larger goals.
Though queries occurred rather infrequently in our data, they seemed to have an important role in work task performance (analysed in Data Set A). Forming a query is typically only one option to proceed in the work task; and choosing this option may indicate that it is considered the best option available. However, the searching itself was seldom the main interest in a work task or a search task. The latter means that even search tasks mainly served other purposes than actual information gathering; the point was to move forward in the work task.
The number of queries and search keys. Both datasets showed a slight increase in the number of queries when perceived work task complexity increases.
However, as analysed in Data Set B, query length decreases at the same time. It may be that specific (longer) queries are easier to form in simple tasks, or that the information systems used in simpler tasks encourage issuing longer queries.
In intellectual tasks, there are more queries per search task on average than in other work task types. Queries are shortest in communication tasks. This may partly be explained by the fact that in communication tasks, communication software were used as the main search medium most often, and typical queries in email and especially in instant messaging software are names (normally either first name or last name only).
Within work task types, query lengths decrease clearly with work task complexity in support and editing tasks, whereas the overall trend of increasing number of queries is not actually visible in any of the work task types. However, there is a positive but not statistically significant correlation between the number of queries in search tasks and work task complexity in support tasks. Kim (2006) found that perceived search task difficulty is connected to an increasing number of query reformulations in both factual and exploratory search tasks, which corresponds to the overall tendency of our data. I did not analyse perceived search task difficulty but it is likely to be affected by similar factors as perceived work task complexity.
Wu and Cai (2016) found that closed search tasks (finding the correct answer) feature more queries, and queries are also longer than in other types of search tasks (open-ended and research-oriented). Though not totally analogous, closed search tasks are by definition similar to our simple or support work tasks, and factual information needs. Thus, well-specified tasks seem to entail longer queries, and though search tasks in intellectual and complex work tasks in our data include more queries than other work tasks, simple or support tasks come rather close.
Thus, Wu and Cai’s (2016) findings are to some extent similar to ours. Borlund (2016) found that teachers’ muddled topical (simulated) work task included more search terms and more search iterations than a closed one. This is in contradiction to Wu and Cai’s (2016) findings, but follows our findings in terms of increasing number of queries.
Expertise can be considered having an inverse relationship to work task complexity - expertise in a subject matter is likely to decrease task complexity.
Palotti et al. (2016) found that experts use more search terms per query than lay people when searching for medical information, which in general terms follows our findings. However, they also found that experts’ search sessions include more queries. This is in conflict with our findings. Without doubt expertise is only one factor affecting especially perceived work task complexity, and thus findings based on these two different measures should be compared with caution.
Query types. Four query types were found and analysed in Data Set B. They were figure-only queries (called f), queries formed of pre-defined values (v), and natural language queries with or without proper names (p, proper name; c, common noun). Queries were typically quite specific since only a third of search tasks featured natural language queries without proper names (type c). P-queries were most common overall. However, they were most common in communication tasks, probably because in communication tasks, communication resources are used as the main search medium most often, and queries to communication resources are often names. Query types do not react clearly to task complexity.
However, search tasks with p-queries are rarer in the most complex than in other tasks. This may indicate that forming exact queries in complex tasks is more difficult. A similar trend is seen in intellectual tasks compared to other task types.
Further, growing work task complexity tends to increase the number of query types in search tasks. This holds also particularly for intellectual tasks.
Vakkari et al. (2003) found that search terms become more specific when participants gain more subject knowledge. This is in line with our findings, since subject knowledge can be considered as having an inverse relation to task complexity: The less knowledge about the subject matter of the work task, the higher the perceived complexity is likely to be.
In summary, most queries contain only one search key; perhaps information systems could better support these queries by providing subject facets for browsing the results, for instance. On the other hand, it is possible that people use short queries just because they have noticed that they work well. In the present data, short queries can as well be exact: they represent for example product codes as search keys in internal databases.
A typical search task includes queries with proper names. This calls for special support in handling proper names in search, just as recognising different spellings, providing autocompletion options or query suggestions. In complex work tasks, search tasks with common noun queries are not unusual, and queries are typically shorter than in less complex work tasks. Thus the performance of complex work tasks may benefit from supporting the learning of the searcher so that the query can become more precise. Query suggestions, providing alternative facets in results and implicit relevance feedback can help the searcher to focus her information need and understand the subject of the search better. Supporting search tasks and work tasks in information (retrieval) systems calls for modelling the tasks of searchers, tracing what they are doing, i.e. understanding what their goal is in performing the query, the search task, and even the work task.
Commercial Web search engines may already work well in such search aids as suggesting queries, autocompletion, and correcting spelling mistakes. Organisations may benefit from investing in internal search of their own databases and intranet as well. Easily findable information can streamline knowledge work by saving both time and effort of task performers.
5 Discussion
First, the discussion section briefly deals with the most important and interesting empirical findings and observations made about real-life information searching as an object of study. I make some statements that are marked in bold and explain them further. This is followed by sections on limitations of the study and implications for future research.