Adaptive Learning and Personalization Within the Enterprise

Facets of Personalization

Although the context Grout referenced includes a host of different factors that can be as diverse as the number of individuals in an organization, it may help to codify them horizontally. Aasman indicated that there are three dimensions of personalization that one must account for to create tailored experiences rooted in adaptive learning—particularly for the pervasive application of enterprise search. “In order to send the right information to the right person, you have to a) understand the document, b) understand the receiver, and you even have to c) understand the sender, who might have a particular motive to send something to someone,” Aasman observed.

With this approach, organizations are responsible for knowing exactly what information is contained in their data or documents. Doing so may begin with classification and metadata management, but should ideally include relationships between that metadata, both within the same content and between multiple sources. Additionally, organizations must have some means of inputting, stratifying, and expressing the salient characteristics of those transmitting and retrieving information. For example, when a branch of the U.S. Armed Forces was attempting to provide relevant courses for soldiers who had been discharged and were attempting to pursue higher education, “it took the courses and put them into a formal ontology and taxonomy of things you could learn,” Aasman said. “They worked hard at representing all the soldiers in the Army with all the features that person might have, so the soldier could search. Department personnel could also recommend courses.”

Implementation Methods

There are numerous approaches for implementing intelligent systems for personalization and adaptive learning. Nearly all of them are predicated on assembling relevant contextual factors— pertaining to the sender, receiver, and the document or datasets—collocating them, and drawing connections between them. One of the most widely used methods involves vector data stores, in which information about each of those three factors is represented as a numerical vector embedding. With this approach, organizations can vectorize anything relevant to personalization and adaptive learning, including “their intellectual property and best practices,” commented Pega CEO Alan Trefler. This content provides a baseline from which to tailor the results of adaptive learning.

Collocating material for personalization in a vector database is applicable for all forms of personalization and adaptive learning—whether for human or agent users. “Vector databases are what we use to get that content awareness,” revealed Shawn Bradford, Laserfiche’s senior engineering manager. “So, if you’re asking a question of the agent and it requires the agent to pull information from a document, that document is vectorized.”

Knowledge graphs are equally viable for housing and drawing parallels between the data that organizations need to be able to surface personalized experiences, especially for search applications. These frameworks excel in elucidating relationships that are found among nodes, which can easily include content from ERP tools, CRM, and the specifics about individual users and their applications. This way, whether adaptive learning is based on agents, language models, or other forms of statistical or non-statistical AI, the “AI is using that explicit set of relationships and that explicit data to get that learning from, which is key going forward,” Grout noted.

Additionally, intelligent agents are helpful for creating personalized experiences for the enterprise. These language model-powered resources can cull information from document management systems and other repositories. They’re also primed for personalizing results when they function as digital twins for human users, respective databases, or both.