Speaking on the record: Combining interviews with search log analysis in user research.

Editor's Notes

1. Lynn Jansen, Bernard J. 2006. “Search Log Analysis: What It Is, What’s Been Done, How to Do It.” Library and Information Science Research 28, no. 3: 407–432.      Connaway, Lynn Silipigni, and Marie L. Radford. 2017. Research Methods in Library and Information Science, 6th ed. Santa Barbara, CA: Libraries Unlimited.    Hunter, Rhonda N. 1991. “Successes and Failures of Patrons Searching the Online Catalog at a Large Academic Library: A Transaction Log Analysis.” RQ 30, no. 3: 395–402.  https://www.jstor.org/stable/25828813.     Jamali, Hamid R., David Nicholas, and Paul Huntington. 2005. “The Use and Users of Scholarly E-Journals: A Review of Log Analysis Studies.” Aslib Proceedings 57, no. 5: 554–571. https://doi.org/10.1108/00012530510634271.     Lamkhede, Sudarshan, and Sudeep Das. 2019. “Challenges in Search on Streaming Services: Netflix Case Study.” In the Proceedings of SIGIR ’19, July 21–25, 2019, Paris, France. https://arxiv.org/pdf/1903.04638.pdf.     Nouvellet, Adrien, Florence D’Alché-Buc, Valérie Baudouin, Christophe Prieur, and François Roueff. 2019. “A Quantitative Analysis of Digital Library User Behaviour Based on Access Logs.” Qualitative and Quantitative Methods in Libraries 7, no.  1: 1–13.  
2. Lynn Connaway, Lynn Silipigni, John M. Budd, and Thomas R. Kochtanek. 1995. “An Investigation of the Use of an Online Catalog: User Characteristics and Transaction Log Analysis.” Library Resources and Technical Services 39, no. 2: 142–152.  
3. Lynn
4. Brittany We started by analyzing the data that was already available to us—search logs from WorldCat Discovery. Benefit: “One gets real behaviors from real users using real systems interacting with real information. Log analysis is everything that a lab study is not.” (Jansen 2017, 349) Caveat: Log data is trace data, meaning it is a trace left behind of what a user did, and it must be interpreted. Jansen, Bernard J. 2017. “Log Analysis.” In Research Methods for Library and Information Science, 6th ed., edited by Lynn Silipigni Connaway and Marie L. Radford, 348-349. Santa Barbara, CA: Libraries Unlimited. 
5. Brittany One of the first things that we had to establish was what the search logs could tell us about user behavior. So we looked manually at some logs to determine what type of information we could glean. The logs could tell us: Type of search (index) Search terms/string Elements of the system interacted with, in this case the autosuggest feature Evolution of searches in a search session Part of this process was working with libraries to recruit participants so we could connect logs with the users who created them. Logs in the system are anonymous—they can’t be tied to the user who generated them. The participants we recruited gave us the ID from their search, which we then used to pull all of the log events associated with that search. On the left-hand side of this slide you can (maybe) see the search log for this participant, CBU09. What we found in interpreting the logs was that this participant started with a mistyped keyword search that returned 0 results. They corrected their spelling and returned over 700,000 results. They then began typing an additional word into their keyword search and selected one of the autosuggested keyword phrases, returning just under 2,000 results. Once we had interpreted the participant logs, we used that information to build an interview protocol. Peggy will talk about the interview process a little later. CBU09 - Female, 19-25, undergraduate, humanities, Christian Brothers University
6. Brittany One of the things that CBU09’s search session demonstrates are the different ways in which users can go about evolving a search during a single session. In order to perform bulk analysis of the search logs, which we wanted to do to be able to provide context for our interview data, we needed to be able to distinguish those different techniques quantitatively. Based on quantitative analysis that was manually reviewed, we established the three categories you see here. A corrected search was defined as being more than 90% similar to a previous search string. That primarily includes misspellings, as seen in step 2 of CBU09’s search. A refined search was defined at 80-90% similarity, but in manually reviewing, we also determined the need to include any instance in which the first search string was contained in the second search string or the search string remained the same but the index was changed. You saw this in steps 3 and 4 of CBU09’s search. The final category we established was a new search, which showed less than 80% similarity with the previous search string. We want to emphasize that this does not necessarily mean that they began searching for a different topic or type of material (although it certainly could). Instead, it represents a search where they changed tactics and began searching using a new or meaningfully different search string.
7. Brittany Once we had established those categories, we could begin looking at the overall patterns in our search logs. For the bulk log analysis, we looked at all of the logs in the Discovery system for April of 2018. We found that on average, a search session lasted 5 minutes and contained just over two searches, with an average of about 5 words per search. Twelve percent of sessions showed search refinements and 33% of sessions showed multiple searches.
8. Christopher Log events for last requests and for the full sessions were categorized in the following way: Online Access Attempt: The user clicked an item or made a request to digitally access the full text of the item.  Physical Access Attempt: The user clicked an item or made a request to place a hold on a physical copy of the item.  Physical Access Option: Some users left the system after looking at holdings. In these cases, users were in a place in the system where they were able to identify the physical item call number and/or location. The interviews revealed that at least some of the participants looked for the item on the library shelf at this point. Since logs cannot reveal what users did after leaving the system, these users were categorized as having the option to physically access the item.  Attempt to Save: The user made an attempt to export or otherwise save the citation.  Search Results: The user made a request for search results. This could include a new search, refinement of an existing search, or the addition of limiters.  Other: These requests and click events did not fit cleanly into any category, and there were no overarching themes to them.
9. Christopher While 54% of the total click events were to generate search results, only 39% of users left on a search results page. More than 50% of users left the system with a request to access an item online, an attempt to access an item physically, or an option to find its physical location. This discrepancy between all request categories and last request categories suggests that users were more likely to leave the system on a successful note. However, it cannot be determined whether the access attempts were successful, so conclusions must be made cautiously.  
10. Christopher
11. Peggy As mentioned earlier, our methodology consisted of both bulk log analysis and user interviews. We conducted semi-structured interviews, using the critical incident technique. How many of you are familiar with the critical incident technique? For those of you who aren't familiar with it, the critical incident technique is a methodology in which a subject is asked to reflect on a specific incident and discuss the cause, description and outcome of the incident as well as his/her feelings and perceptions of the situtation. Connaway, Lynn Silipigni, Chris Cyr, Brittany Brannon, Peggy Gallagher, and Erin Hood. 2019. “Speaking on the Record: Combining Interviews with Search Log Analysis in User Research.” ALISE/ProQuest Methodology Paper Competition Award Winner.    Go over methodology (semi-structured interviews with critical incident technique) Connaway, Lynn Silipigni, and Marie L. Radford. 2018. Survey Research. Webinar presented by ASIS&T, January 23. https://www.slideshare.net/LynnConnaway/survey-research-methods-with-lynn-silipigni-connaway and https://www.youtube.com/watch?v=4dlpAT7MXh0 .
12. Peggy We developed specific protocols for each of our interviewees based on what we learned from their search logs. The protocols included some standard questions such as what they were looking for and why, whether they were successful – for example, if they were doing a known-item search, did the item appear in their search results, and whether they were delighted and/or frustrated with their search experience.
13. Peggy Looking again at the same interviewee that Brittany spoke of – CBU09 – we knew from the search log what she did: she misspelled her search term, corrected the spelling, had over half a million results, refined her search by additional another search phrase, had nearly 2,000 results and then that was it. She appears to have abandoned the search. CBU09 - Female, 19-25, undergraduate, humanities, Christian Brothers University
14. Peggy What we learned from her interview, however, was that she was just getting started searching for resources for a paper and hadn't yet narrowed down her topic. She was overwhelmed by how many search results she got, and so she did indeed abandon the library search. However, she came back to the library search tool at a later time, after she had done some Googling and honed in on her topic. She told us how she found the library search tool very helpful once she knew what she wanted to write about, and that she appreciated the help she received from some student workers. She also mentioned that because of the library instruction she had received in her first year, she felt confident in using the library.  CBU09 - Female, 19-25, undergraduate, humanities, Christian Brothers University
15. Peggy So, once we had conducted all of our interviews, we reviewed the interview transcripts, developed a code book and then had two different team members code each of the interviews. The themes that emerged from the interviews tended to fall into seven different buckets. Interviewees talked about the various search strategies they used and the factors that went into their decision-making process on which resources they wanted to access. As we've learned in several other studies over the years, convenience and ease of access are major factors. They talked about the formats they were most interested in – PDFs being the most mentioned. They talked about how they evaluate resources such as considering the type of study involved from a particular article and whether the article was peer-reviewed. They mentioned features they like about the current library search tool as well as features they wished it had. Many of them talked about being satisfied with their search experience but wouldn't go as far as saying they were delighted by it and spoke of some of their frustrations with the top frustration being too many search results. And many of them spoke favorably about the impact of working with a librarian for receiving library instruction on their confidence in their information-seeking abilities.  So, that's a quick review of what we did and how we did it. I'm going to turn things back over to Lynn who's going to talk about some of the challenges and benefits of the methodology we used for this study.
16. Connaway, Lynn Silipigni, Chris Cyr, Brittany Brannon, Peggy Gallagher, and Erin Hood. 2019. “Speaking on the Record: Combining Interviews with Search Log Analysis in User Research.” ALISE/ProQuest Methodology Paper Competition Award Winner.   
17. Lynn
18. Lynn
19. Lynn It’s changing the way we think, the way we work together. Research goes beyond the academic… influencing product development We earn the right to share our insights by listening.