The Artificial Intelligence application of machine learning predictive analytics and recommender systems is pervasive in a number of companies today like Amazon, Spotify, and Pinterest, but what about libraries? It’s unlikely a person will see or have heard of such systems being widely used in many Libraries, even though they can offer users a unique and seamless way to access and receive recommendations for services they might have otherwise overlooked.
I believe the adoption of these systems in libraries can elevate library services in many ways and transform the way students discover resources in and outside of their discipline. It can also help librarians rethink traditional approaches to outreach. An illustration of this is Amazon’s predictive and recommendation system; a library version might look like a user selects a workshop and based on their current and past preferences they will receive a list of personalized services.
The 2022 IDEA Institute on Artificial Intelligence has made me aware of the advantages, implications, and scope of AI in libraries and provided me with the building blocks I need to develop and design a project of my own using machine learning. The beauty of AI is that there is often more than one method of approaching a project. The course and complexity are dependent on several factors but ultimately determined based on the goal or desired outcome of the project.
The question then becomes, how do libraries leverage these innovative technologies? Fortunately, there are many approaches to initiating a straightforward project. I learned there are simple machine learning systems that exist like linear and logistic regression models, which take binary classifications of data and predict outcomes. These models help evaluate the class an item belongs to and predicts the classification of new items based on the data fed into the machine. More advanced models like neural networks and deep learning take data and combine different ways to capture features through different layers (neural networks). PyTorch, Keras, and TensorFlow are three of the most commonly used libraries for more complex models.
An AI project in particular that I am excited to implement is an idea from one of my colleagues. The project is to build a predictive model that takes user-profiles and analyzes which groups of users are using which services. This model will help us learn which services are used and by which users so that we can target users with similar profiles. To begin this project, I need to consider what user data I need and how I plan to collect that data. The collection of user data will help me create features that will be important for the predictive model. There are a few machine learning systems that do similar tasks, but one, in particular, I want to explore further is a supervised machine-learning decision tree algorithm. The decision tree algorithm takes data and continuously splits the data according to parameters. In the future, I would like to expand on this project to design and develop a recommender system that curates services according to users’ preferences (i.e., Pinterest).
Many ethical considerations arise with predictive models and recommender systems; a few that come to mind are the privacy, autonomy and personal identity of user data, fairness, transparency, and social effects. These are elements that I am keeping in the forefront of my mind as I begin and continue to design my project.
Finally, as with any other project, this is an iterative process. There will be times when I have to go back and rework areas of my project. Additionally, depending on the scope and complexity of my project, I might consider pulling in subject experts to assist with the implementation of the project.