More Related Content
What's hot
What's hot (19)
Similar to MARC-y MARC and the Coding Bunch
Similar to MARC-y MARC and the Coding Bunch (20)
More from Andrea Payant
More from Andrea Payant (20)
Recently uploaded
Recently uploaded (20)
MARC-y MARC and the Coding Bunch
- 1. MARC-y MARC and the Coding Bunch Anna-Maria Arnljots Metadata Assistant anna-maria.arnljots@usu.edu Paul Daybell Archival Cataloging Librarian paul.daybell@usu.edu Kurt Meyer Government Information and E- Resource Cataloger kurt.meyer@usu.edu Andrea Payant Metadata Librarian andrea.payant@usu.edu Becky Skeen Special Collection Cataloging Librarian becky.skeen@usu.edu Liz Woolcott Cataloging and Metadata Services Unit Head liz.woolcott@usu.edu Utah Library Association Annual Conference May 21, 2021
- 2. 2 Background • Multi-year research into user search behavior for all metadata standards employed by the unit First phase: MARC Next phases: EAD, Dublin Core • Project started just as the library moved everyone to work from home • Whole unit was able to participate in the coding project
- 3. Problem Statement What is the correlation between user search terms, the placement of MARC records in search results lists, and the performance of individual MARC fields in a search process?
- 4. Research Questions • What is the frequency and placement of MARC records in search results list? • Where are Search terms located in MARC Records?
- 5. Methodology
- 6. • Focused on the Discovery Layer (Encore) because it was the primary search portal used by patrons • Pulled list of all URLs accessed on three days • Put into Airtable and coded Web Log Analysis
- 7. • Filtered for URLs that lead to search results pages • Fed URLs into Octoparse, a web-scrapping tool • Scrapped the list of search results, URLs, pagination, and results # • Numbered the results and put into Airtable, linked to originating URL Web Scraping
- 8. • Search Results List and URLs Extracted bib # Created formula to link to MARC view of bib Unit members pulled up Bib record and copy/pasted it into Airtable Assigned codes for : o Creator of record o Material type o MARC fields where term was found o Fields that were not present Automated formula examined wordcount of record Airtable
- 9. • Web Log URLs Coded for basic search features: o Page Types o Advanced Search fields used o Facets used o Page Number Coded the queries (search terms) for: o Search term construction o Search categories (known item, topical) o User Path o Known Item Titles Airtable (continued)
- 10. • Known Items pulled out specifically and coded (most for a separate project looking at the discovery layer) Format/Genre Availability Physical or Electronic Location Steps to access Listed by Final Content Provider Checkouts Discoverability in Google Scholar o Steps to Access Airtable (continued)
- 11. Results Research Question #1 What is the frequency and placement of MARC records in search results lists?
- 12. Analysis 1.1: How frequently are MARC records showing up in search results? Batch 1 Batch 2 Batch 3 Combined MARC-based catalog records 5264 3299 4749 13312 Records from other platforms 20326 17560 16811 54697 Total Records 25603 20859 21560 68022 Percent MARC records 20.56% 15.82% 22.03% 19.57%
- 13. Analysis 1.2: Is there a difference between locally created records and vendor supplied records in the frequency of listing in search results? Record Creator # Records in results list % Total records in results list # Records accessed % Total records accessed Vendor 7,727 58.05% 163 39.00% Cataloging and Metadata Services 5,066 38.06% 239 57.18% Distance Campus Libraries 410 3.08% 5 1.20% Record unavailable at time of coding 52 0.39% 2 0.48% Patron Services, Library Media Collections, or Resource Sharing and Document Delivery 33 0.25% 8 1.91% Acquisitions 16 0.12% 0 0.00% Unknown 5 0.04% 1 0.24% Natural History Library 3 0.02% 0 0.00% Total 13,312 418
- 14. Analysis 1.3: How are MARC records ranked in the search results list? • Most common position for MARC records in a search result set of 25 items, is position 4 • MARC records appear in the top five search results 25.35% of the time
- 15. Analysis 1.4: Where do MARC records for known items rank in the search results list? Percentage of Times Available Whole Object Appeared in Search Results by Position Number Result 1 Result 2 Result 3 Result 4 Result 5 Results 6-10 Results 11-15 Results 16-20 Results 21-25 Total # 125 107 61 49 37 104 67 56 35 % in results 18.7% 16.0% 9.1% 7.3% 5.5% 15.6% 10.0% 8.4% 5.2%
- 16. Results Research Question #2 Where are search terms located in MARC records?
- 17. Analysis 2.1: What fields are used most in retrieving records? 9100 4998 4806 3700 1328 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 245 505 650 520 600 Number of Records MARC Fields MARC Fields Where Search Terms Were Located (Top 5)
- 18. Analysis 2.2: For records accessed by the patron, is there a difference in where search terms are located? • The 245 Title statement remained highest, appearing 64% more often than the next most utilized field • Instead of the 505 Formatted Contents Note being in second place, the 650 Subject Added Entry is the next most used field • The 505 Formatted Contents Note and 520 Summary fields retained a spot in the top four fields
- 19. Analysis 2.3: For locally created records and vendor-supplied records, is there a difference in where search terms are located? Percentage of fields used in record retrieval (top 5 most frequent) Field Field Description CMS Records Vendor Records 245 Title Statement 43.80% 51.64% 505 Formatted Contents Note 28.13% 69.65% 650 Subject Added Entry - Topical 40.89% 56.58% 520 Summary, etc. 23.41% 76.03% 600 Subject Added Entry – Personal Name 59.94% 32.68%
- 20. Analysis 2.4: What fields are not present in the records? CMS Vendor Not Present Present Not Present Present Author (both 1xx and 7xx) 0.75% 99.25% 1.18% 98.82% Subject (any authorized) 4.46% 95.54% 6.73% 93.27% 505 Formatted Contents Note 63.96% 36.04% 45.54% 54.46% 520 Summary Note 75.60% 24.40% 50.45% 49.55% All Categories Present 14.86% 33.26%
- 21. Analysis 2.5: Which fields would make the greatest impact if not included in the record? • The top four fields with the greatest impact on retrieval, if not found in a record: 505, 245, 520, and 650 • Without the 505 or 520, 16.86% of all records appearing in results would not have shown up • In contrast, without 650 and 600 fields, only 0.66% of records would not have appeared in the search results
- 22. Analysis
- 23. 23 • Non-MARC records have advantage over MARC Of all records in search results are Non-MARC Analysis • MARC vendor records appear more often than locally created MARC records Of MARC records place in the top 5 search results. Occur more frequently in vendor records Occur at the same rate in Vendor and Locally created records
- 24. 24 Analysis Title fields are most important over all, but… • Ranked higher than 245 for records where search terms matched only one field • Consistently in the top 4 fields that retrieved a record (along with 520) • If missing, 12% of all MARC results would not have been displayed
- 25. 25 Analysis Subject fields are important But… Most important field for matching search terms Most important field for records viewed by patrons Would not have been displayed if field were missing Instance of subject fields being “clicked on” 1xx fields were much more likely to be “clicked on”
- 26. ▫ Cataloger will retain ability to make best judgment for each record, but will be asked to consider the following guidelines: - More emphasis on creating 505 and 520 notes in local records - Less emphasis on 6xx fields as an entry point - More emphasis on 1xx fields as an entry point 26 Take-Aways
- 27. MARC-y MARC's Coding Bunch • Anna-Maria Arnljots • Josee Butler • Ryan Bushman (Stats) • Paul Daybell • Barbara Fleming • Maddie Gardner • Alisha Grant • Bryn Larsen • Sabrina Leatham • Rachel Olsen • Andrea Payant • Kurt Meyer • Jessica Mills • Abby Rodabough • MaKayla Roundy • Melanie Shaw • Becky Skeen • Sara Skindelien • Seth Westenburg • Liz Woolcott
- 28. Resources
- 29. Full Procedures: https://usulibrary.atlassian.net/l/c/8H7jgU98 Article with final results: Liz Woolcott, Andrea Payant, Becky Skeen & Paul Daybell (2021) Missing the MARC: Utilization of MARC Fields in the Search Process, Cataloging & Classification Quarterly, 59:1, 28-52, DOI: 10.1080/01639374.2021.1881010 Related articles Robert Heaton & Liz Woolcott. Unraveling the (Search) String: Assessing Library Discovery Layers Using Patron Queries. Library Assessment Conference, January 2021 • Presentation: https://www.libraryassessment.org/program/2020- schedule/#jan21 • Paper: https://www.libraryassessment.org/2020-proceedings/
- 30. Questions? Anna-Maria Arnljots Metadata Assistant anna-maria.arnljots@usu.edu Paul Daybell Archival Cataloging Librarian paul.daybell@usu.edu Kurt Meyer Government Information and E- Resource Cataloger kurt.meyer@usu.edu Andrea Payant Metadata Librarian andrea.payant@usu.edu Becky Skeen Special Collection Cataloging Librarian becky.skeen@usu.edu Liz Woolcott Cataloging and Metadata Services Unit Head liz.woolcott@usu.edu
- 31. Thank You!
Editor's Notes
- I will now give you a quick overview of our methodology for our project
- In order to determine how MARC records interacted with the user search process, the research team examined the logs of URLs that were generated by Encore, our library’s discovery layer. Each search session in Encore generates a combination of static and dynamic URLs. Dynamic URLs capture a user’s search terms and any facets selected, advanced search categories used, additional search result pages accessed, and bibliographic record numbers for MARC record pages. Google Analytics was used to gather reports of time-stamped, URL logs generated over the course of multiple days. Resulting data was put into Airtable, a relational database for further analysis
- The Google analytics report of URL logs was downloaded, and dynamic URLs that led to a search results page were isolated from the main report and fed into Octoparse, a web scraping tool. Each resulting page from the dynamic URL was scraped by Octoparse to gather data for the search terms used, the number of results on the page, the total number of results available to the user, and the title and link of each item in the list of results presented to the user on that page. The results were numbered and added to our Airtable database and then linked to the originating URL.
- Search results list and urls were coded to identify the bibliographic record number. A formula was created within the system to link out to the MARC view which was used to access and copy the full text of the MARC record into Airtable. Codes were assigned for record creator (whether generated by library personnel or vendor supplied) and material type. Codes also identified where the search terms appeared in the MARC record and they also related prominent categories of fields that were not present in the record. For every instance where the search term appeared in the field, that field was copied into a separate column for further analysis. Also, an automated formula examined the word count of each record.
- Web logs URLs were also coded for basic search features, including page types, advanced search fields, facets used, and search result page numbers Queries, or search terms, were coded as well to parse out how search terms were constructed, search categories (either known item or topical), user paths, and known item titles.
- Finally, known item searches were pulled out and coded. The search terms entered by the user were analyzed through a multi-step process that reran the same terms in a browser to ascertain if the search terms reasonably matched the title or identifier of a known item. When found, the corresponding URLs were tagged as Known Items and coded for format, availability, medium, location, keywords used etc. Following this coding, each known item was double checked by a research team member to determine if the library provided access to it, either physically or in electronic format. Paul will now go over the results of our data and coding
- So, just to summarize what Paul said. Non-MARC records have clear advantage over MARC in our discovery layer. 80% of all results came from non-MARC sources, despite non-MARC records making up 60% of the database. AND MARC records only place in top 5 results a quarter of the time. If we just look at MARC records by themselves, though, we see that Vendor records appear more often than locally created records and are more likely to include the 505 and 520 fields. They have the same frequency of author and subject fields as records cataloged locally, though, so 1xx and 6xx fields are not making a difference between the two types of records. We suspect that full text search in non-MARC records and the greater presence of 505 and 520 fields in Vendor records provide more words and phrases for the index to search against. And that our own work is less visible because we aren’t putting our emphasis in these places.
- In fact, if we look further into how the 505 functions, we find that while title fields were the most important field overall, the 505 ranked higher than 245 for records where search terms matched only one field (meaning those search terms weren’t found anywhere else in the record.) The 505 and 520 Summary Notes were consistently in the top 4 fields that retrieved a record Most telling of all was that in 12% of all records, if 505 had not been present, the record would not have been displayed in the search results list AT ALL. The only other field more significant that this was the Title field
- Let’s take a look now at how authorized fields like the subject and author field interact with search terms. Subject fields are important, but results on how they interact with search terms are mixed, It is the 3rd most important field for matching search terms and the 2nd most important field for records viewed by patrons, but only .55% of records would not have been displayed if the Subject field had been missing. So, while the data demonstrated that search terms matched subject headings frequently, it also demonstrated that those same terms were frequently available elsewhere in the record already. Additionally, it was very obvious that subject headings were rarely ever used as a means for finding other materials (for instance, when we envision a patron "clicking on" a subject link to find like materials.") There was only one instance of subject fields being “clicked on” to bring up related records. This is, in large part, due to the visibility of subject headings on the main search page,. You can only access the terms through the record itself (if the patron clicks on it) or on occasion in a “tag” field at the bottom of the facet column. Whether due to interface design or to the utility of the field itself, we cannot definitely say. However, 1xx creator fields were the most likely authorized heading fields to be used and the data displayed evidence of them being used to find related records and materials. They are also the more visible of the authorized headings fields – not only showing up in the search results list, but also being actionable from that list without having to enter the record.
- In reviewing all the data, the unit developed a few "take-aways" that we could incorporate in our day-to-day work. These included taking more time to add 505 Formatted Content Notes or 520 Summary fields to locally created records. We felt the data demonstrated that additional 505 and 520 fields would likely make our records more visible to the search algorithms. Additionally, we will place less emphasis on the subject fields as part of our workflow. This doesn't mean eliminating subject work from what we do – but rather just not spending as much time developing subject headings as before. We will also continue our authority work on the 1xx creator fields, as they are the most visible of the controlled headings fields and also highly visible in the search results page. These aren't hard and fast rules, but rather guidelines to follow. Our catalogers will continue to be able to exercise their own judgment when creating records. But having this understanding of how the records are used will be imperative in that judgment making process.
- We would like to thank the following people for all of their help in making this research process possible. The whole Cataloging unit at USU Libraries, including catalogers, cataloging assistants, and student technicians participated in this project. We would also like to thank Ryan Bushman, the assistant to our Assessment Librarian for all his help with the statistics for this project. We are so appreciative to this whole coding bunch!
- If you would like to try out this process yourself – we have put our step by step instructions online at the URL you see above. This will include all of the procedures we used to pull the data from Google Analytics, scrape the data with Octoparse, and our codebooks that all of the project contributors used. We will also put this link into the chat for you. You can also read about this process and the results in our recently published article in Cataloging and Classification Quarterly. It is titled "Missing the MARC: Utilization of MARC Fields in the Search Process." and the link DOI above is a link to the article. We will also put that into the chat for you. Note that both of these links are available on the handout for this session, too. The data from this project was also used in a recent publication and presentation at the Library Assessment Conference which examined how patrons used the Library Discovery Layer Encore. The links are available on this slide and we will put them into chat as well. Just note that the proceedings are quite up yet, but should be soon.
- Thank you for your time! Does anyone have any questions?