Temporal Patterns of Searching in a Public Library Discovery System / Aspects temporels des habitudes de recherche dans l’outil de découverte d’une bibliothèque publique

Abstract

This article reports on an investigation of temporal patterns of searching within the Edmonton Public Library discovery system to identify the temporal trends and aspects of public searching. The specific objective of this study was to examine the temporal patterns of public searching on weekdays and weekends in the months of summer and fall. The findings show that there are hourly, daily, weekday, and weekend patterns associated with the public searching of a discovery system.

Résumé

Cet article présente les résultats d’une étude des requêtes effectuées dans l’outil de découverte de la bibliothèque publique d’Edmonton afin d’identifier les aspects temporels des habitudes de recherche du public. L’objectif était d’examiner la répartition temporelle des requêtes durant la semaine et la fin de semaine dans les mois d’été et d’automne. Les résultats montrent que les recherches du public dans l’outil de découverte suivent des tendances différentes selon l’heure et la journée ainsi que des tendances différentes la semaine et la fin de semaine.

Introduction

The rise of increasingly sophisticated discovery systems in libraries that integrate the library catalogue with digital repositories and collections and social media has created vast volumes of interaction and transaction data that can assist researchers and practitioners in understanding user search behaviour and search queries. In this article, the temporal search patterns using the data from the Edmonton Public Library (EPL) discovery system, a large urban library system in Alberta, are examined to develop insight into the information search behaviour of users of the EPL’s discovery system. This study focuses on the dynamics of time by examining how search queries change over time (hourly, daily, weekly, monthly, and seasonally) and within the context of broader events. The study aims to understand the temporal aspects of a search during the day, week, and months of summer and during one week in the fall. The main research question that this study aims to address is: what are the temporal patterns of a search, if any, in a large, urban public library? Specific research questions related to this main research question include:

1. 1. Are there specific temporal patterns of searching across weekdays during the months of summer and fall?

2. 2. Are there specific temporal patterns of searching during the weekdays and weekends in the months of July and October?

3. 3. Are there specific hourly patterns of searching during the weekdays and weekends in the months of summer and fall?

While the temporal aspects of searching have been studied in the context of Web search engines, digital libraries (Beitzel et al. 2004), and information retrieval test collections (Diaz and Jones 2004), little research on temporal trends has been conducted using public library discovery systems. This article focuses solely on the temporal aspects of user queries submitted to a public library discovery system. The nature, types, and characteristics of the queries, such as popular or high frequency search queries, are covered in a different paper by Tami Oliphant and Ali Shiri that is currently in production. The ultimate goal of this study is to propose potential improvements to the EPL discovery system search user interface to support user search behaviour and to provide insight into the ways in which public library discovery system interfaces can improve user engagement.

Prior research

Information search behaviour

Studying information search behaviour of users interacting with digital information systems has a long history, dating back to the development of the first commercial bibliographic databases and online public access catalogues (OPAC) (Millsap and Ferl 1993; Peters 1993; Mat-Hassan and Levene 2005). Furthermore, examining information search behaviour of large numbers of users as they interact with search engines (Jansen and Pooch 2001; Spink et al 2001; Oliphant 2013), library websites (Ghaphery 2005), OPACs (Lau and Goh 2006), commercial databases (Wolfram 2008), and digital libraries (Jones et al. 2000; Shiri 2010, 2011) has become a popular research area in information and computer sciences. In large part, this is due to the recognition of the importance of understanding users’ search behaviours to design and develop more interactive and engaging information retrieval systems, the availability of larger data sets that track user search behaviour, and the development of new data analysis tools that allow researchers not only to track searches but also to ask new questions. Research suggests that the use of semantic, geographical, and temporal relationships in digital catalogues will lead to significant improvement in the quality of retrieved results (Tochtermann et al. 1997).

Temporal queries in libraries

The notion of temporality in the context of searching and retrieval is multi-faceted and may consist of a host of factors associated with time. Alonso et al. (2011) have noted that there are several different ways to express temporal information of the types’ date and time. They distinguish three types of expressions, namely explicit expression (for example, 25 January 2010), relative expression (for example, today), which needs to have some corresponding reference time (21 June 2016), and implicit expression (for example, Labour Day). Several studies have been carried out examining search queries on the Web. For example, Pasca (2008) experimented with one billion Google Web searches to develop a new algorithm for textual indexing of date-related queries containing “when” questions and queries with specified dates to facilitate answering open-ended questions. Conversely, using a collection of scientific articles, Shaparenko et al. (2005) examined the temporal patterns of key scholarly papers and publication years to identify influential papers and authors. They found that temporal information, without the use of citation information, can be used to identify key topics and influential authors.

Temporally dependent queries are defined by Metzler et al. (2009) as queries that are event specific and that change over time such as “new years” and “presidential elections.” They introduce two types of temporal queries: the “explicitly year qualified” query, where a year is specified in the query, and the “implicitly year qualified” queries. An explicitly year qualified query is a query that contains a year such as “Scarface 1932” or “Scarface 1983,” depending on which version of Scarface is wanted. The “implicitly year qualified” query is a query that does not actually contain a year but that may have been implicitly formulated by the user with a specific year in mind. Examples of implicitly year-qualified queries include “Olympics” and “Easter.” Based on their experiment with 670 Web queries, Metzler et al. concluded that more than 7% of queries belong to the implicitly year qualified query category. Shokouhi (2011) stresses the importance of correctly identifying seasonal queries by search engines to ensure that user search results are temporally reordered if necessary. He uses “Halloween” and “Christmas” as examples of seasonal queries that are derived from temporal information needs. He proposes an automatic classification of seasonal queries based on the historical frequency of queries and their conversion into time series. The distribution of queries across various periods and their seasonal peaks provide a good measure for detecting seasonal queries.

In a discussion of social searches and how users’ interests can support the personalization of a social search system, Khodaei and Alonso (2012) provide a temporally aware categorization of users’ social interests as follows: (1) recent interests; (2) ongoing interests; (3) seasonal interests; (4) past interests; and (5) random interests. These dimensions of interests and needs can provide a useful basis for a temporal analysis of queries. Kulkarni et al. (2011) studied 100 queries for popularity changes and query intent changes over time. They developed a model measuring query popularity in terms of decreased query searches, increased query searches, and stable query searches; examined how query intent changes for certain events throughout the year; and identified the following three query intent dynamics:

1. 1. Zoom: query intent zooms in on the event around the time the event will occur and then zooms out post-event.

2. 2. Shift: query intent can undergo a shift (for example, the query “opening day” indicates an information need about the baseball opening season when it is issued at the start of April; however, by the start of May, the query intent shifts to the opening season of boating events.

3. 3. Static: query intent can remain relatively static for recurring events such as “Easter ideas” and questions about taxes.

In a study of time-sensitive methods for term auto-completion, Shokouhi and Radinsky (2012) provide a novel approach to dynamic term suggestion. They adopt a model using time series and rank documents based on temporal aspects rather than on query popularity because query frequency and popularity per se can be misleading. Shokouhi and Radinsky observe that certain types of queries are more popular and recur during the weekends, while others seem to gain more popularity during the week. They conclude that a temporal analysis of queries can significantly improve the ranking of query auto-completion terms.

Whiting and Jose (2014) address the value and importance of auto-query suggestion in reducing users’ cognitive and physical effort, noting that most query suggestion systems rely on past popular queries. They offer a different approach, relying on recent query popularity evidence rather than long-term query popularity analysis and on short-range query popularity prediction based upon recently observed trends. In addition, Joho, Jatowt, and Roi’s (2013) user-centred survey of temporal Web search experience found that recency was one of the major components of participants’ information needs, that continuous and seasonal needs were as popular as recently occurring interests, and that nearly 45% of the users’ needs were related to either long-term interests or seasonal interests. Another study by Joho, Jatowt, and Blanco (2015) addresses the temporal dimension of search queries, yet this work was done in a laboratory setting using search engines rather than using library catalogues. As Alonso et al. (2011) note,

temporal information can be organized hierarchically: Temporal expressions can be of different granularities—for example, of type day (“20 May 2011”) or of type year (“2011”). Due to the fact that years consist of months, and months and weeks consist of days, temporal expressions can be mapped to coarser granularities based on the hierarchy of temporal expressions.

In this study, our specific focus is on the volume of search and query traffic across weekdays, weekends, and times of the day. In other words, the notion of temporality operationalized in this study refers to hourly, daily, and weekly patterns of searching in a public library discovery system. While numerous studies have been conducted on seasonal and temporal queries in the context of Web search engines and test collections, the research on temporal aspects of queries in public library search systems is scarce.

Methodology

The following three research questions were addressed in this study:

1. 1. Are there specific temporal patterns of searching across weekdays during the months of summer and fall?

2. 2. Are there specific temporal patterns of searching during the weekdays and weekends in the months of July and October?

3. 3. Are there specific hourly patterns of searching during the weekdays and weekends in the months of summer and fall?

To explore these research questions, the EPL permitted access to their search query data. The summer months (21 June (first day of summer)–21 September 2014) and the week of 1–6 October were purposively sampled to explore and analyze searches by day of week and to determine if there are temporal trends and patterns in weekdays and weekends. The EPL serves a metro area population of approximately one million people in Edmonton, which is the capital city of the province of Alberta in Canada. The EPL is 101 years old, a prominent member of the Canadian Urban Libraries Council, and one of Canada’s most well-known public library systems. It has received numerous awards and recognitions for its innovative and inclusive approach.

BiblioCommons and Google Analytics

The EPL makes use of BiblioCommons as its front-end interactive catalogue that integrates various digital resources such as e-books, catalogues, accounts, and recommendations. The EPL website and discovery system provides a single search bar with three options provided as “catalogue (default),” “website,” and “articles.” To gather user interaction and search-related data, the EPL uses Google Analytics. Google Analytics gathers various statistics about the EPL users, their geographic location, their search terms, and what devices and browsers were used when searching BiblioCommons. According to Clark, Nicholas, and Jamali (2014, 193), “Google Analytics . . . has already been successfully used by researchers for the study of user behaviour, website effectiveness, and Web traffic and has been recommended by all these researchers.” Indeed, Google Analytics has been used to measure the effectiveness of an academic library website (Turner 2010), to measure the effectiveness of a faculty’s personal websites (Plaza 2009), to evaluate library instructional materials available to distance students (Memmott and deVries 2010), and to inform library decision making (Paul and Erdelez 2013).

For this study, we selected the following two sets of user queries:

• • summer data set: user queries gathered by Google Analytics from 21 June–21 September 2014 and

• • fall data set, including:

  • ∘ user queries gathered by Google Analytics from 6–9 October 2014 (weekdays) (7,611 queries) and

  • ∘ user queries gathered by Google Analytics from 10–12 October 2014 (weekend) (4,320 queries).

The initial focus of the project was on queries submitted during the summer of 2014. However, to conduct an in-depth comparative analysis of weekday versus weekend search patterns, the search queries posed to the EPL discovery system during the week of 6–12 October were purposively sampled.

The EPL has set up a separate profile for BiblioCommons using Google Analytics, which means that one can determine if searches were emanating from the library catalogue, the library website, or the articles field. In other words, the EPL has set up a BiblioCommons profile that allows Google Analytics to gather data from just the library catalogue search field. Furthermore, Google Analytics distinguishes between “visits without a site search” and “visits with a site search,” meaning that Google Analytics parses those visits that include an internal website search and those that do not. However, one limitation to the data is that Google Analytics does not capture Internet protocol addresses of users for privacy reasons. Therefore, it is not possible to provide detailed analyses that geo-locate temporal queries to a specific area or neighbourhood in Edmonton.

The data sets

A meta-analysis of users’ search behaviour studies, sponsored by the Joint Information Systems Committee and the Online Computer Library Center, concluded that “a large, random sample of specific demographic groups of information seekers should be identified in order to identify how individuals engage in both the virtual and physical worlds to get information for different situations and contexts” (Connaway and Dickey 2010, 5). In this study, we made use of a relatively large sample of searches conducted on a public library discovery system.

Two different time periods were chosen for this study, namely the summer months of 2014 and one week of October 2014. To manage and contain this project, three consecutive months (21 June–21 September) that constitute a temporal “season” were selected, with the first day of summer being 21 June. The summer months (21 June–21 July; 22 July–21 August; 22 August–21 September) were purposively sampled for data collection for a variety of reasons. These data were the most current when this research project was undertaken. Furthermore, in Canada, the kindergarten to Grade 12 school year ends in June and begins again after Labour Day in the beginning of September, and it was possible that the transition from, and back to, school influences search queries posed to the EPL’s catalogue. During the summer months (including the end of June after school is let out), public libraries offer many programs and events for children, teens, and families, and the general public partakes of festivals, movies, cultural events, travel, gardening, and other interests that might also be reflected in search queries.

Table 1.

Total search sessions, unique searches, and searches for the summer months

The data were collected and analyzed using both quantitative and qualitative approaches. Google Analytics was used to collect the data and to carry out preliminary analysis (for example, the number of searches per month), but further analysis required the data to be exported and analyzed using Excel and SPSS Statistics. The data set consisted of 1,123,152 total searches over the course of the summer season (table 1). A description of the concepts of a session, a unique search, and a search is useful for understanding table 1. A search session is a group of interactions that take place on a search engine within a given time frame. A session may consist of multiple searches. Unique searches are those that consist of unique search terms, excluding multiple searches on the same keyword during the same session. The searches mentioned in table 1 include all of the searches conducted regardless of identical or frequently used terms.

The total unique searches resulted in a different number from the total searches because of daily, multiple searches of words or phrases such as “Harry Potter “ or “Fault in Our Stars”—two popular books. For example, on Monday, 7 July, there were 70 searches for “Harry” and 39 searches for “business.” Each one of these 70 and 39 searches would be counted in Google Analytics as a search, but they would only be counted once as a unique search. Furthermore, total sessions refer to the number of discrete sessions by users. For example, a single user may have multiple sessions, and many sessions will include multiple searches. This data set was used to conduct statistical analyses to identify and analyze patterns of searching across various months of the summer.

To understand the user’s search behaviour better, a second data set, comprised of purposively sampled queries from the week of 6–12 October, was collected and analyzed. October is an interesting month for this kind of study as it is the second month of fall and has occasions such as Thanksgiving (in Canada) and Halloween. Given the scope of the project, a small sample of queries from the second week of October 2014, including the weekdays of 6–9 October and the weekend of 10–12 October, were analyzed for comparative purposes (table 2). The rationale was to observe whether and how the search patterns of users shift from weekdays to weekends.

Table 2.

Unique searches for the week of October

Data analysis

Combinations of qualitative and quantitative analysis methods were used to analyze the data. Statistical analyses were conducted on the number of searches by weekday. The analysis of variance (ANOVA) test and the Tukey’s honest significant difference (HSD) test were conducted for the data set of summer 2014. Tukey’s HSD test—or the Tukey–Kramer method—is a statistical test similar to the t-test and is a single-step multiple comparison procedure. It can be used on raw data or in conjunction with an ANOVA test to find means that are significantly different from each other. These two tests were mainly used to observe differences among searches conducted on weekdays and weekends. For the hourly analysis of a select number of days in June and October, Google Analytics profiles were used for those days, and a visual examination of hourly and weekly patterns was conducted. Given the small sample of searches conducted in one week in October, Google Analytics was used to examine the temporal patterns for that week as well as for individual days of the week. An hourly analysis was conducted to study if there were specific patterns of searching during the day.

Findings¹

Statistical analysis of search data for summer months

One of the key research questions in our study was to determine the variation in number of searches in the library discovery system across days of the week. Were people searching the discovery system most frequently during the weekends or weekdays? Table 3 shows the total sessions and searches by day of the week for the entire three-month summer season in which the study was carried out. For example, in the three-month period, there were 172,954 total searches carried out on Mondays, whereas the total number of searches carried out on Fridays was 155,539. The above analyses clearly demonstrate that the general public tends to conduct a larger number of searches during the weekdays and a smaller number of searches on weekends.

More unique sessions, total searches, and unique searches were carried out on Mondays than on any other day of the week. Once these totals were placed in a table, an ANOVA test was conducted to check for the variance of means, which are shown in the “Mean number of searches” column in table 3. In terms of the mean number of searches, for example, Monday’s mean is the highest, while Sunday had the lowest mean number of searches.

Table 3.

Total sessions, searches, and the mean number of searches by days of the week

Table 4.

ANOVA and Tukey HSD

Next, an ANOVA and a Tukey HSD test were conducted to find out if the means are statistically significant from each other. Table 4 shows the statistically significant ANOVA among the days of the week (<0.05 = statistically significant).

For Mondays, the mean number of searches is statistically significant with Friday, Saturday, and Sunday. For Tuesdays, the mean number of searches is statistically significant for Saturday and Sunday. While this statistical test demonstrates that the number of searches on Saturday and Sunday especially are statistically significant from other days of the week, the reasons for this are purely speculative. These analyses clearly demonstrate that the general public tends to conduct a larger number of searches during the weekdays and a smaller number of searches on weekends.

The temporal search patterns of users of a public library discovery system during the weekdays and weekends examined are consistent with the previous research on the temporal analysis of Web search queries. For instance, in a study analyzing 3.3 million Web queries, Sanderson and Dumais (2007) found that the frequency of occurrence of search events on the search engine was different for each day of the week, with Monday representing the highest number of search events, followed by Tuesday and Wednesday, and a steady decrease on Wednesday and Thursday.

Hourly analysis of search queries

In addition to the statistical analysis of all of the search queries, six days in June—namely 22 June (Sunday), 23 June (Monday), 24 June (Tuesday), and the weekend of 27 June (Friday), 28 June (Saturday), and 29 June (Sunday)—were specifically examined to see if there were temporal patterns of searching during the day. The peak hours for issuing queries to the EPL discovery system were from 12:00 p.m. to 3:00 p.m., which is followed by a significant decrease until 4:00 p.m. From 4:00 p.m. until 8:00 p.m., there is another, slight increase in searches. The largest increase in searches throughout the day occurs from 11:00 a.m. until 12:00 p.m. Interestingly, there are almost the same number of searches occurring at 6:00 a.m. as there are at 4:00 p.m.

Most searches conducted by users tended to occur during the hours of 12:00 p.m. and 3:00 p.m. with a slight increase in searches from 4:00 p.m. until 8:00 p.m. Even though Sunday, June 29 had over 1,000 more sessions than Sunday, 22 June, the search pattern for both days is very similar.

Similar patterns of searching can be observed on Monday, 23 June and Tuesday, 24 June from around 8:00 a.m. to 4:00 p.m., with a decreasing trend after 4:00 p.m. The data also shows a small number of searches starting from around 6:00 a.m. and then increasing throughout the morning. This temporal pattern of searching is similar to what was found by Beitzel et al. (2004), who reported, in an analysis of Web search queries, that only 0.75% of the day’s total queries appear from 5:00 a.m. to 6:00 a.m., whereas 6.7% of the day’s queries appear from 9:00 p.m. to 10:00 pm. Contrast this search pattern with Sundays. On both Monday and Tuesday, the number of searches steadily increase over the course of the day and remains at a plateau until approximately 6:00 p.m. There are not significant increases and decreases in short periods as was the search pattern on Sundays.

Comparing the number of searches conducted on Fridays and Saturdays provides useful insight into search volume variations when the weekend starts. An interesting trend emerges in the search patterns on Fridays and Saturdays. On both Friday and Saturday, most searches were conducted from 9:00 a.m. until 4:00 p.m. There is a sharp decrease in the number of searches that can be observed on both Friday and Saturday after 4:00 p.m. Looking closely, a difference can be observed in the temporal patterns of searching between Monday and Tuesday and Friday and Saturday. On Monday and Tuesday, there is a steady decrease of the number of searches after 4:00 p.m., whereas on Friday and Saturday, a sharp decline is noticeable. This can be explained by the fact that people tend to be involved in other activities after 4:00 p.m. on Fridays and Saturdays. The searches conducted on Mondays and Tuesdays tend to continue after 4:00 p.m. with a steady drop until 8:00 p.m.

In a study of temporal analyses of search queries in four Web search engines, Hochstotter and Koch (2009) found that the daily peak lies between 1:00 p.m. and 3:00 p.m. and that the maximum number of queries submitted for all of the four search engines was observed in the afternoon and not in the late evening hours. This is similar to findings from Beitzel et al. (2004) in their study of millions of search queries submitted to a commercial Web search engine in which they found that the highest volume of search queries was conducted in the afternoon, particularly after 1:00 p.m. Hochstotter and Koch (2009) also analyzed the distribution of searches over the days of the week. They found that 14% of all of the searches were conducted on Mondays with a decreasing proportion during the week, and Saturdays represented the lowest number of searches (10%). Their data also showed a slight increase in the number of searches on Sundays, increasing to 12%. Table 3 indicates that there are fewer searches on Sunday than Saturday. In our study, the data showed that there were a larger number of searches conducted on Mondays.

Analysis of October 2014 queries

To compare temporal patterns for weekdays and weekends and for the “summer” season and fall, a small set of search queries from the month of October 2014 were analyzed. Specifically, the data included the searches conducted during the week of 6–9 October (Monday to Thursday) and the weekend of 10–12 October (Friday to Sunday). This data set consisted of 11,931 sampled queries. The rationale for choosing this data set was to conduct an analysis of a small set of search queries issued during weekdays and weekends in the fall of 2014. The first and main observation about this data set is that the number of searches conducted during the week of 6–9 October 2014 was higher than the number of searches carried out on the weekend of 10–12 October 2014. The data shows a steady decline in the number of searches from Monday, 6 October to Sunday, 12 October 2014, which follows the same pattern that was evident during the summer months.

To compare the days of the week of 6–9 October 2014 with the patterns of the summer months discussed earlier in this article, we also examined the temporal patterns of searches for four individual days and observed a steady search pattern from 8:00 a.m. to 4:00 p.m. on Monday, 6 October, which is similar to Monday, 22 June. The temporal pattern of searching on Friday, 10 October shows a steady decrease in the number of searches after 4:00 p.m. This is similar to the temporal pattern of searching observed for the data on Friday, 27 June 2014.

Therefore, it can be argued that, overall, general temporal search patterns during weekdays in the summer and in the month of October are similar and consistent. It would be interesting to explore temporal search patterns in different months of the fall season to see if this pattern is observable. A quick comparison of the temporal patterns of searching on a fall weekend with those observed on the data for 28–29 June 2014 shows similarities in terms of peak and decline times. For Saturday, 11 October, there is a steady temporal pattern of searching from 8:00 a.m. to 4:00 p.m., whereas on Sunday, 12 October, there is a peak at 12:00 p.m. and a steady decline until 4:00 p.m. This comparison of the weekends of June 2014 and October 2014 provide an interesting insight into temporal patterns of searching. Future research might explore other seasons of the year to see if there are specific temporal patterns of searching associated with each month of the year.

Discussion and implications

Wolfram et al. (2001) note that longitudinal studies of Web searching can provide valuable insights into how public Web searching is evolving, changing, and moving in certain directions. Similarly, gaining insight into how the public conducts searches on public library discovery systems can shed light on the ways in which discovery search systems can be designed to provide customized and personalized user experience and recommendations. The temporal aspect of searching is becoming increasingly important as numerous search engines and social network sites make use of time as an instrumental factor for search experience personalization. As Alonso et al. (2011, 6) argue, “temporal information in the form of temporal expressions offer an interesting means to further enhance the functionality of current information retrieval applications” by providing alternative search experience, mechanisms, and filtering techniques. Exploring public searching of public library discovery systems and comparing the observable patterns with those from studies of Web search engines and social networks provides a more holistic and comprehensive perspective of searching. It allows researchers and designers of search systems to gain a better understanding of how people search for information in various contexts and how the public’s mental models of search systems impact their interaction with a wide range of search and information interaction environments. More specifically, the temporal patterns of searching and interaction with various search systems during the day, week, or month allow for more sophisticated personalization and customization functionalities.

Based on our analysis of data from the summer and fall of 2014, it is evident that there are specific patterns of searching across weekdays and weekends. EPL users conduct searches most frequently on Mondays with a steady decline until Thursday and a significant drop in the number of total searches on the weekend. Our analysis shows that there are specific temporal patterns of searching on individual weekdays as well. For instance, there is a sharp decline in the number of searches after 3:00 p.m. on Sundays, whereas the decrease in the number of searches on weekdays is steady and continues until 8:00 p.m. Our analysis of a small set of data from October 2014 also shows similar patterns of hourly searching, indicating that the number of searches decreases in the later days of the week and that Friday, Saturday, and Sunday are associated with a smaller number of searches. A comparison of a week in the summer of 2014 and a week in October 2014 shows that hourly patterns of searching tend to be similar—that is, both in June and October 2014, users tended to start conducting searches on the discovery system from 6:00 a.m. with a steady increase in the number of searches until between 2:00 p.m. and 3:00 p.m. and then a steady decline until roughly 4:00 p.m. and then a slight rebound until around 8:00 p.m.

Another key observation of this study substantiates that searches conducted on the EPL discovery system exhibit similar temporal patterns in the summer and fall, meaning that a greater number of searches are conducted on weekdays than on weekends in both summer and fall months. This finding has implications for what materials or programs might be promoted or featured on the library’s website at different times of the week. At the same time, these data do not provide any explanation as to why fewer people search the EPL discovery system on the weekend. This is a limitation in analyzing large data sets without understanding the user’s context. Understanding temporal search patterns and volume in public library discovery systems is particularly useful as it allows researchers to identify the hours, days, and times when the public is engaged. This insight will provide opportunities for libraries to maintain user engagement through dynamic and time-sensitive recommendations of relevant content, services, and programs. In addition, combined with popular topics and search terms, temporal information can provide event-based or seasonally related recommendations.

Comparing our findings with the previous research on the temporal aspects of users interacting with social networks allows for a more comparative and inclusive picture of the ways in which people interact and engage with a wide range of online environments. In a study of mood and happiness analysis, based on tweets and their temporal information, for example, it was found that people tend to be happier on Friday, Saturday, and Sunday and that their daily tweeting behaviour was influenced by different times of the day (Dodds et al. 2011). For instance, the daily high usage of the term “coffee” in the analyzed tweets occurs between 8:00 a.m. and 9:00 a.m. This finding may partly explain the fact that tweeting behaviour and the searching of a discovery system, as was found in our study, represent distinguishing daily and weekly temporal patterns. Dodds et al. (2011) concluded that the seven-day week cycle is a historical and cultural artefact.

Several studies have explored how time and temporal patterns are correlated with daily activities. Research on location-based social networks has found that people tend to show certain temporal patterns of interaction related to searching for places to eat. A study found that most search activities happen between 9:00 a.m. and 11:00 p.m., with two peaks at around 1:00 p.m. and 7:00 p.m. (Ye et al. 2011). In another study that focused on predicting flu trends using Twitter data, Achrekar et al. (2011) analyzed hourly activity patterns at different hours and found high traffic volumes from late morning to early afternoon and fewer tweets posted from midnight to early morning, reflecting people’s work and rest hours within a day. Their weekly analysis of tweets showed that the average daily usage pattern within a week showed more people discussing flu on week-days than on weekends. This study found that a larger number of searches were associated with weekdays and that the temporal patterns of searching during the day shows that there are more searches conducted between 8:00 a.m. and 4:00 p.m. than at any other times. Our findings in the context of a large public library discovery system are consistent with this study in that most searches conducted on discovery search systems tend to happen between 8:00 a.m. and 4:00 p.m.

Conclusion

This article has addressed the temporal aspects of searching in a public library discovery system. Two sample data sets from the summer and fall of 2014 were examined to understand if there are specific temporal patterns in public searching and, in particular, if there are temporal patterns of days of the week and hours of the day. The goal of this study was to understand temporal information-seeking behaviours in order to help formulate recommendations to improve search-and-discovery systems, collections, and services used in public libraries. This study is also an example of collaboration with the EPL to engage in research that bridges the research/practice gap often found in LIS. Further research should investigate the temporal aspects of hourly, daily, and weekly patterns of searching across the 12 months of the year to examine if the patterns are similar or different from the ones reported in this article. Understanding temporal aspects of searching in the context of public library discovery systems contributes to the design of better exploratory search interfaces that encourage users to navigate, explore, and discover new content in the system. Future research might also be conducted to study the nature of search topics, queries, and terms across various hours, days, and times in public library discovery systems as well as in public and open digital libraries to understand the temporal aspects of searching in various contexts. This kind of research will provide new ways of designing user experiences and customization in Web search engines, open digital libraries, and archives.

For the supplementary appendix to this article click here.