- The paper introduces Personal Knowledge Graphs as a novel method to enhance personalization and semantic features in e-learning.
- It presents a detailed architecture that combines user-generated data with external knowledge graphs via NLP and weighted algorithms.
- The study addresses challenges in privacy, data maintenance, and temporal dependency while improving collaborative search and customized learning.
This paper, "Personal Knowledge Graphs: Use Cases in e-learning Platforms" (2203.08507), explores the application of Personal Knowledge Graphs (PKGs) to enhance personalization and semantic capabilities within e-learning and collaborative search environments.
The core problem addressed is the limitations of current e-learning and collaborative platforms. While large encyclopedic knowledge graphs (KGs) like DBpedia exist, they lack personal user data, and existing e-learning platforms often lack sophisticated semantic features and personalization truly tailored to individual or group needs. Collaborative online work, though increasingly vital, often uses features adopted from single-user settings, creating a gap between the technology offered and the needs of collaborative environments.
The proposed solution is to integrate Personal Knowledge Graphs into the backend of these platforms. PKGs are defined as small, user-centric KGs designed to sit on top of or alongside larger KGs, representing personalized information and interests. Drawing inspiration from PKG use in the medical domain for patient data, the research aims to adapt this concept to the educational domain to represent learners and their activities.
The architecture for creating a PKG, as outlined in the paper, involves several steps:
- Input Stream: Collects user-generated data, including user profiles, individual activity, and user group information.
- Intelligence Computation: Processes the input data. This involves leveraging external Knowledge Graphs, applying Named Entity Recognition (NER) and NLP software to identify concepts and entities within the user data. A weighted algorithm recalculates points of interest over time, filtering entities passed to the PKG.
- Back End Creation: Constructs the PKG, integrating the processed data and entities, while adhering to an ontology and ensuring user/group privacy. The architecture emphasizes interconnecting platform activities with Linked Data and external KGs.
The research focuses on two main use cases:
- Collaborative Search: The goal is to improve platforms like Learnweb by incorporating PKGs. By linking user and group activity to KGs, the system can attain semantic relations between data and identify key entities of collaboration. This enables the development of semantically enhanced personalized and group features. Potential applications include improved user credibility, better understanding of group activity, learning analytics, personalized and group recommendations, and feedback mechanisms. This use case aims to address how PKGs can offer more personalized features (RQ3) and better collaboration (RQ4) in collaborative search learning environments.
- E-learning: Using a platform like eDoer, which connects labor market skills with Open Educational Resources (OERs), the implementation of PKGs would allow for highly personalized recommendations. Based on a user's PKG (reflecting their learning preferences, accessibility needs, past actions, etc.) and the platform's knowledge base, the system can suggest relevant learning content. This use case specifically targets how e-learning platforms can provide better semantically enhanced personalized features like recommendations using PKGs (RQ2). The overarching question of how to syntactically and semantically represent a PKG in the e-learning domain (RQ1) underpins both use cases.
The paper highlights opportunities such as introducing a novel application of PKGs in education, potentially increasing the adoption of knowledge bases and semantic web technologies in e-learning, and exploring features specifically designed for collaborative environments.
However, significant challenges are acknowledged:
- Knowledge Acquisition and Maintenance: The process of identifying, extracting, storing, and maintaining information within PKGs, including linking to external KGs and updating data as user interests evolve.
- Privacy: Ensuring user consent for data collection and storage, managing sharing of data within groups, and implementing the "right to be forgotten" according to regulations like GDPR.
- Temporal Dependency: PKGs are time-dependent, and the frequency and method of computation can impact the accuracy and relevance of the data, requiring careful consideration of processing time and data validity.
The methodology proposed combines qualitative and quantitative analysis. It involves an interdisciplinary literature review, formalizing research questions and hypotheses, implementing the proposed approach, and conducting user studies with human participants. These studies would involve collecting data from platform usage, gathering feedback through interviews and questionnaires, and potentially involving experts for technical and pedagogical evaluation. The paper notes the lack of specific gold standards for evaluating "Searching as Learning" but suggests adapting existing recommendation benchmarks.
Preliminary results mentioned include the publication of the EduCOR ontology [(2203.08507), citing (2109.08163)], which provides a user profiling pattern suitable for e-learning PKGs and has been extended for web search PKGs with accessibility and privacy considerations. Another work, currently under submission [collabgraph], proposes a collaborative search graph summary visualization utilizing backend concepts related to PKGs and showed high user likability in initial evaluations.
In conclusion, the paper posits that integrating PKGs into e-learning and collaborative search platforms connected to knowledge bases offers a path towards enhanced personalization, explainability, and semantic richness. Future work includes deeper investigation into PKG lifecycle management (acquisition, creation, maintenance, updates), addressing privacy concerns in collaboration with legal expertise, potentially integrating author/editor data for improved content credibility, developing advanced personalized recommendation systems, and exploring broader impacts on human factors and knowledge building in educational settings.