23 minute read

The Editorial Board of The Alabama Librarian

The Editorial Board of The Alabama Librarian

Jessica E. Platt, Editor-in-Chief Alabama State University

Advertisement

M. Delores Carlito, Editor University of Alabama at Birmingham

Tim Dodge, Editor Auburn University

Charlotte E. Ford, Editor University of Montevallo

Jessica Hayes, Editor Auburn University Montgomery

Peggy Kain, Editor University of Alabama at Birmingham

Kathy Wheeler, Editor University of South Alabama

Laura Pitts, Editor Scottsboro Public Library

Michelle Hamrick, Proofreader Irondale Public Library

Lori Northrup, Proofreader Samford University

Jodi Poe, Proofreader Jacksonville State University

Kelly Marie Wilson, Proofreader Independent Researcher

What Machine Learning Means for Libraries

The COVID-19 pandemic threw the role of technology in our lives into sharp relief. While videocon-

ferencing programs, delivery apps, and the availability (or not) of broadband internet were some of the most

visible examples, the pervasive use of artificial intelligence, particularly in the form of machine learning, has

received comparatively less attention. Machine learning allows “smart assistants” like Alexa, Cortana, and Siri

to comprehend our commands, cloud photo services to organize and tag friends in our photos, and our pre-

ferred streaming app to recommend what to watch or listen to next.

Although we encounter these technologies every day, many people still do not know what machine

learning is or how it works. Machine learning is a subset of artificial intelligence. Rather than operating in gen-

eral ways like human intelligence often does, machine learning is intended to accomplish a specific task, such

as making a decision in a narrow domain. For example, a machine learning tool could be built to determine

whether a given picture contains a cat. While this may seem like a trivial task, the advantage of the technology

becomes clear when the goal is to identify the presence of a cat in thousands, or even millions, of pictures. Ma-

chine learning is indispensable in the era of big data.

The basis of all machine learning is pattern recognition. As a machine learning algorithm operates, it

builds mathematical models of the subject of interest. Importantly, these mathematical models are not static,

and that is where the “learning” aspect of the technology comes in. To return to the cat example, instead of a

programmer trying to input all the possible properties of a cat picture ahead of time, the algorithm draws on

data from a collection of examples that do and do not contain cats to create mathematical models representing

the degree of “catness” in a picture on its own terms. Then, as it is exposed to more data over time, it updates

and refines these models. This not only improves the accuracy of the models, but also allows them to adapt.

The flexibility inherent in the bottom-up identification of properties is a key advantage of machine learning.

Medical researchers made extensive use of machine learning in order to develop prognoses and treat-

ments for COVID-19. In the early days of the pandemic, when the disease was not well understood, scientists

the risk factors for poor outcomes, as well as to gauge the effectiveness of various medical interventions. Ma-

chine learning excels in situations where little is known in advance about the system of interest, such as when

human bodies react to a novel virus. Arguably, automated analysis of electronic health records at this scale al-

so helped protect patient privacy since researchers never saw or handled data that could be traced to individu-

als.

When it comes to libraries, many familiar tools already leverage machine learning. For example, com-

mercial discovery systems designed to let users search the library more like they search the web are built on

machine learning, particularly in how they determine relevance rankings. With user profiles, those relevance

rankings could be personalized based on the preferences of users the system has judged to be similar. Another

application of machine learning is text analysis, which is changing the way scholars use libraries to do re-

search. With natural language processing and a corpus of text, someone could glean information from a source

that they’ve never read, or even directly accessed.

One area in which machine learning allows for rapid advances is knowledge organization. The systems

libraries use to classify information have traditionally been top-down, for example with controlled vocabular-

ies and subject headings, and do not easily coexist with alternative classifications or terminologies. Machine

learning can be employed to create knowledge graphs that link subjects, sources, authors, and more based on

connections that emerge from people’s actual engagement with these works. While not a replacement for cata-

loging, these bottom-up webs of interrelated information can provide a useful complement to it. They are par-

ticularly beneficial in areas of rapid development, where decisions about organization and labeling are neces-

sarily contingent and more easily permit shifts between research contexts or frameworks of understanding.

Regardless of whether or not librarians encounter or employ machine learning tools in their libraries, it

is crucial to understand how these technologies work. More than ever, algorithms that rely on machine learn-

ing shape the information landscape of our users. In order to help users be information literate (and to remain

information literate ourselves), we need to know how this happens. Whether it’s searching with a search en-

gine or browsing a social media feed, machine learning drives what we’ll see, and just as importantly, what we

won’t see.

example, librarians could help users set up profiles on search tools that refine results using machine learning in

order to find information that is highly specific to their research topic or area of interest. The algorithm will

take advantage of available data based not just on the user’s own history, but that of all the other users making

similar requests, perhaps taking into account their judgments about source quality, to decide on the best results.

It may fall to librarians to explain how these services differ from more traditional keyword searches or alerts

so that users can get the most out of them.

On the other hand, since machine learning is designed to recognize patterns and reproduce them accu-

rately, tools of this kind can be limiting in cases where someone is researching a different topic or studying

within an emerging area. These scenarios involve a deliberate break from the user’s past habits and, in the lat-

ter case, a divergence from the large datasets the search tools rely on for learning. Thus, it may be a mistake to

rely on “smart” services in these instances. In addition, since machine learning methods are designed based on

presumed similarity between examples, they are not suited to take into account unique situations or highly spe-

cialized information needs.

It is particularly important to educate users about these issues with respect to familiar commercial

search engines. Companies like Google have engineered their algorithms as precisely as possible to return re-

sults that conform to user expectations of what they are requesting. This is especially apparent in their efforts

to make general internet search tools responsible to natural language questions. That may not be a problem

when there is a simple, direct, and singular answer, for instance if someone were to ask “What was the score of

the last Iron Bowl?” But as librarians are well aware, this way of finding information is less than ideal for most

substantive queries. It might help users understand the limitations of these tools if librarians can explain that

machine learning reduces the likelihood of seeing highly relevant results if they are less popular.

There is also the problem of information siloing that arises when people rely on social media feeds to

discover news and other content. The algorithms that govern users’ timelines are designed to learn what to dis-

play in order to maximize user engagement—the time people spend interacting with the app. This means there

is no guarantee the material presented to users is truthful or represented accurately, because the goal of these

programs is not to inform or educate, but to inspire a response. The app collects data on what kinds of posts

sonalized, people are presented with very different information environments, leading them to wonder whether

people who have another perspective or express opinions contrary to their own are living in a different world.

In a sense, with respect to what they experience online, they are. Librarians who teach information literacy

need to raise these issues, help learners confront their implications, and offer alternatives that represent multi-

ple points of view.

Finally, as libraries provide services and support to help their communities recover from the pandemic,

making users aware of how decisions affecting their lives are being made by algorithms based on machine

learning could reduce barriers to accessing resources. For instance, many people come to the library to get as-

sistance searching for jobs. Since the volume of applicants for certain jobs has dramatically increased, employ-

ment websites often use algorithms to filter for keywords or other indicators that a resumé matches the em-

ployer’s criteria. A job seeker frustrated that they are not getting any responses to their applications might ad-

just their strategy if they knew their resume was being screened out before it reached any human review. Li-

brarians do not have to become experts on the inner workings of these systems to be able to give basic advice

to such users and help them connect with opportunities for which they are qualified.

Like any technology, machine learning has its pros and cons, but in the age of big data it is here to

stay. Librarians will need to familiarize themselves with developments in this area to provide the best possible

education and service to members of our communities, modeling lifelong human learning in the process.

Ali Krzton

Research Data Management Librarian

Auburn University Libraries

Migration to Koha: Selection and Implementation at a Small Academic Library

Amanda Melcher

Head of Technical Service and Associate Professor

University of Montevallo

melcheras@montevallo.edu

Kaycee Ledbetter

Head of Circulation

University of Montevallo

kledbetter@montevallo.edu

Abstract

Carmichael Library at the University of Montevallo decided to migrate away from the integrated library sys-

tem that was in place for approximately 25 years during the summer and fall of 2018. This paper describes the

rationale behind the change, along with the factors that went into selecting Koha, an open-source integrated

library system (ILS). Specifically, the authors describe the implementation, including the selection rubric,

training process, migration, and advice to achieve the best outcome. The entire process took a mere six months

to complete. This practical paper will serve as a roadmap with tips for libraries that are considering an ILS

switch.

Keywords: Koha, OPAC, ILS, integrated library system, online public access catalog, academic library, imple-

mentation

Background

The University of Montevallo, a small liberal arts college located in Montevallo, Alabama, is home to

the Carmichael Library. Full time enrollment for the University is about 2,500 students and there are approxi-

mately 500 faculty and staff employed. The Library has 12 staff members, seven librarians and five paralibrar-

ians. The Library offers over 200,000 physical items for use across different collections such as general circu-

lating, reference, youth, multimedia (e.g., DVD, CD), reserves, and archives/special collections.

Rationale for Migrating

For many reasons which will be described, there was a sense of urgency to migrate away from the inte-

grated library system (ILS) that had been in place for a number of years. First and foremost, there was a con-

sensus among library staff that the system no longer met our needs. Secondly, the Library was on the sixth

year of a flat budget and needed to put cost saving methods in place. Thirdly, the University’s Information

Systems and Technology (IS&T) Department asserted that the Library needed to either switch to a cloud-based

system or update the server that housed the ILS. Finally, the Library was able to hire a systems librarian.

An ILS facilitates access to and management of the resources available in a library. It organizes and

tracks materials and keeps a record of library patrons. A library’s ILS is the backbone of operations, making it

an essential tool. The previous ILS had been in place for twenty-five years and prior to that, the Library used a

manual card catalog. The system was becoming increasingly outdated and lacked upgrades and innovation. To

successfully work with other systems in place on campus, we needed an ILS that would allow processes to be

modified for efficiency.

The budget was a critical factor in deciding the right time to transition. Along with the previously stat-

ed flat budget, database costs increase 3-5% annually, so cancellations would be required. The high cost of the

previous ILS felt outsized compared to its strengths and weaknesses. We had no indication that our budget cir-

cumstance would change in the near the future, but we knew the ILS cost would continue to rise. The library

director decided we could no longer justify paying for the expensive ILS, and we needed to see what other op-

tions were available.

In addition to the annual cost increases, the lack of innovation of the ILS was holding us back. We

were paying a large subscription and maintenance fee every year, but only performing software upgrades when

absolutely necessary. The reason we were reluctant to update was two-fold. One, we did not have a systems

librarian for a number of years. Without that knowledge and expertise, we didn't feel confident enough to

make such a big change. Secondly, when an update was needed, every individual computer had to be updated

manually. With 12 staff computers and eight student worker computers, this was very time-consuming. When

we did upgrade the software, there was very little innovation or added functionality. The changes were minor

and usually security-related. The system had simply stopped meeting our needs.

A third factor that necessitated the change was a serious warning from the University’s IS&T depart-

ment. A 25+ year old server hosted the previous ILS; a disaster waiting to happen. IS&T cautioned us that if

the subscription and maintenance renewal time (Fall 2018) for the previous ILS. IS&T recommended purchas-

ing a new server as soon as possible to safeguard our data. We had not budgeted for this expense and it did not

address the other issues.

Finally, for the first time in a number of years, the Library had a dedicated systems librarian. Many of

our routine updates and upgrades had been put on hold until this position was filled. The IS&T department

helped us with critical systems processes, but this type of change required a knowledgeable library employee

to be very involved in, not only the migration process, but also maintenance of the new system after the imple-

mentation.

Selecting a new system

While most everyone in the Library agreed that the previous ILS was outdated, but work had to be

done to build consensus for the change. Support across the library departments was essential for the transition

to be successful. When we started the migration in earnest, we had six months until the subscription and

maintenance contract renewed on the previous ILS. We could not afford to renew the contract and pay for a

new system out of the same fiscal year. The only other option was to be without an ILS until we were ready to

go live. The stakes were high, and consequently, the timeline was very intimidating.

One of the first steps taken by the library director was to name a migration leadership team. While

changing systems would affect everyone in the Library, it would have the biggest impact on the three employ-

ees that used the system most often. The Technology Integration and Web Services Librarian (systems), the

Head of Technical Services Librarian, and the Circulation Manager became the driving force behind the deci-

sion to move forward with finding a new system. This team was integral in taking the project from conception

to completion.

The library director and the systems librarian selected three ILS contenders for migration based on

systems’ integration with our Discovery system. The three selected for consideration were SirsiDynix’s Blue-

Cloud, WorldCat’s WMS, and Koha. The leadership team was tasked with doing the legwork to determine

which option was best for the Library. The library director took a supportive but hands-off approach. The team

briefed the director throughout the process.

Before actually looking at any of these systems in depth, it was decided that we needed an impartial

way to grade the systems based on our needs. The team decided to create a rubric to evaluate them in a system-

atic and fair way. This provided an opportunity to include all the library staff. Each department was asked to

provide a list of attributes and functions needed to efficiently do their job and a “wish” list of items that they

did not currently have, but would be ideal. Within the areas of circulation, reporting, cataloging, acquisitions,

serials, public interface, system, and pricing, the staff identified approximately 40 items for the rubric that

were considered essential, important, or just nice to have. These 40 features were scored on a Likert scale of 1-

5, 1 being poorly supported and 5 being fully supported. The rubric not only provided an organized way to

evaluate the system options, but also provided a clear vision of what the group needs were. The rubric illustrat-

ed what the previous system lacked and reinforced the need for the change. The leadership team divided up the

rubric areas and, with the help of staff, worked through the evaluation. As we progressed through the rubric,

we would stumble on a new feature in one ILS and would need to re-visit the other systems to check for avail-

ability. Overall, this was tedious and time-consuming, but a beneficial endeavor. (See Appendix for the ILS

Selection Rubric.)

While all of the systems had merit, two clear frontrunners emerged, Koha and WMS. The migration

team had the opportunity to visit two Alabama libraries, one that was using Koha and one using WMS. It was

extremely beneficial to interact with a live system as opposed to a demo. We learned things from the employ-

ees using the systems that were not obvious when using a demo version. For example, creating reports in one

of the systems seemed very simple in demo mode. However, our visit revealed that the process was complicat-

ed and oftentimes the employee had to call customer support for help. We are deeply appreciative of this op-

portunity to see the systems in action. These visits were an invaluable tool in helping us come to the unani-

Implementation

In June, the implementation team reached out to ByWater Solutions to gain reassurance that the transi-

tion timeline would be possible. The vendor responded with a systematic implementation checklist and time-

line, giving us confidence that the transition could be completed within our time parameters. The first step in

the checklist was to set up a testing website, which provided a demonstration of how the system would operate

and allowed staff to begin engaging with the process. Each department was able to log into the website and

begin their set-up. This step gave everyone an opportunity to evaluate outdated practices and policies.

There is a plethora of system settings for circulation, acquisitions, cataloging, and reporting. Engaging

everyone in the department with the setup process provided the group with momentum and enthusiasm about

the decision. Staff made changes and reevaluated their processes to better suit the abilities of the new system.

There were many instances where a minor change in the workflow was made due to a lack of, or abundance of,

system preferences available. The group was cohesive, making it easy to collaborate and agree on new proce-

dures.

A crucial part of the implementation checklist was data extraction from the previous system, which

needed to begin as soon as possible. Exporting our data in the correct format would ensure compatibility with

the importation to Koha. The previous ILS vendor charged a fee to export our data, which we were not willing

to pay. ByWater helped the systems librarian format and export our data from the old system. Not all infor-

mation contained in the previous system was able to be migrated to Koha, while other data did not export

smoothly. For instance, extracting patron data was laborious. The previous system did not have an easy way to

export all the data fields needed in one report, instead multiple extractions had to take place, and the data had

to be pieced together by mapping matching data points. This required hours of detailed work to construct the

file in the necessary format with the required information. When the data was suitable to import, it allowed for

further testing of parameters to solidify the policies before the new system was implemented. Seeing our own

data in the test site was exciting and made the transition timeline seem feasible. It allowed employees to famil-

vided a regimented schedule that covered each area (e.g., circulation, cataloging, reporting, etc.) While all staff

were encouraged to attend any session they desired, the schedule was devised so that staff could attend the ses-

sion most relevant to their department and job responsibilities. Prior to the ByWater visit, the implementation

team asked staff to attempt all their typical tasks in the Koha test site and document any questions or issues

that arose. We wanted to get the most out of the visit by making the sessions targeted and relevant. Training

was important to the transition momentum. As the staff prepared for the new system, they were able to provide

input on ways to improve previous procedures and workflows based on the new system. Another important

part of the transition was training the student workers. It was important that they felt invested in the change

and equipped to use the new system. The students adapted very quickly to the system with little to no major

issues.

Post-Implementation

Implementation day came and went and the new system proved to be the much-needed upgrade we had

hoped for. Employees quickly adapted to Koha; creating new procedures and workflows. Predictably, there

were a number issues that had to be resolved after launch day. The patron data we originally imported into Ko-

ha came from data extracted from the previous system. When the spring semester started, it was time to update

the patron data. It took a few months and lots of troubleshooting to securely connect and download the newest

batch of patrons from the University’s administrative software. Another snag we encountered was an error in

the patron data that was extracted from the previous system. This error caused the patron data to not map cor-

rectly with the late fine and lost book data. Fines were assigned to the wrong patron accounts. Discovering this

issue and correcting it took almost six months.

Before the implementation, we had been reassured that Koha would work with our primary book ven-

dor to enable EDI ordering. We later learned that Koha will only work with EDIFACT, not X12 (the vendor

preference). To enable this feature, our book vendor charged a fee (higher than some of our discipline alloca-

tions). Another glitch we had not anticipated was the need for Structured Query Language (SQL) knowledge.

Scripts had to be written to export our data from the previous ILS and to import it into Koha. We needed it to

run anything other than the simplest of reports in Koha. The systems librarian had to devote some time to gain

While many processes in the new system were improved and simplified after the migration, serials pre-

diction patterns were equally as frustrating. There was also more clean-up than we anticipated, particularly

with our catalog item types and collection codes. Another issue we encountered was, unlike the previous sys-

tem, borrowers in Koha have to be set for each individual item type, i.e., there is no sweeping “faculty privi-

lege”. Other minor issues were discovered, but most were quickly resolved.

This is a timeline of our implementation and migration:

Conclusion

Hopefully, this practical paper will assist libraries that are contemplating an ILS migration. Most

changes of this magnitude occur over the span of a year or more, but we did the bulk of the migration in six

months. The process was made manageable with appropriate preparation, effort, and determination to see it

through to completion. A few of the most compelling reasons for our decision were: cost, usability, cloud host-

ing, positive user reviews, on-site training opportunities, and the fact that Koha is open source. After the initial

year’s implementation cost, Koha cut our ILS cost in half! The system is innovative and has many dynamic

aspects. We have relied on ByWater support and they have not let us down. Overall, we are very glad we made

the switch to Koha.

Following are some suggestions that we hope will assist colleagues in a similar situation:

Involve individuals in implementation that will most often use the system;

Start early on the rubric, it will take more time than you anticipate;

Determine if any additional skills are needed for implementation (e.g., SQL knowledge);

Have all staff test the system before going live;

RUN REPORTS in the previous system before it is turned off, just in case they are unavailable in the new

system; and

If implementation spans two fiscal years, have an acquisitions plan in place detailing how to wrap-up open

orders.

References

Breeding, M. (2017). Library systems report 2017: competing visions for technology, openness,

and workflow. A m erican Libraries (5), 22.

Appendix ILS Selection Rubric: Criteria scored by Likert scale (1=Not Supported, 5=Fully Supported)

Circulation

Course reserves Ease of viewing and editing patron details (at a glance) Check in/out process (fewer steps than Horizon) View list of patrons w/ fines (unpaid) In house circulation capabilities iPad/Tablet inventory and shelf reading capabilities Integration for payments (credit cards, etc) Lost item fees & tracking based off status of book

Reporting

All checked in/out items per day Collection sorting by Call Number Items by status Item Notes Local notes in MARC records

Cataloging

Integration with Connexion Integration with EDS (RTAC and regular item updates) Copy Cataloging Integration with labeling program Ability to see the cover of the book (helpful in finding a book) RSS feature to help in population of new materials (New books, DVD's etc for lists) Local Call numbers will migrate Can gift information be migrated?

Acquisitions

EDI capabilities Integration/flow through for YBP and/or Amazon Import order information into ILS Statement information import Compare YBP and current collection for duplicates (notification?) Capability of searching Stat Class (professor's names when creating PO)

System/General

Cloud based

Externally hosted Intuitive language Intuitive menu design Strong community support No need for high level dedicated technical support inhouse OAI harvesting

Serials

Select multiple titles to check in at once/Batch check in Batch claim edit Prediction patterns

Public Interface

Ease of finding items Intuitive interface Browse by call number

Vendor Considerations

Relationship with vendor/rep Community Support Vendor/Software age Reputation (librarytechnology.org)

Pricing

Total cost of implementation and maintenance

4

This article is from: