Authors:
Yiming Luo、Patrick Cheong-Iao、Shanton Chang
Paper:
https://arxiv.org/abs/2408.08894
Introduction
In the digital age, the sheer volume of information available to learners has grown exponentially, thanks to advancements in information and communications technology (ICT). Digital libraries, Massive Open Online Courses (MOOCs), and other online resources have made it easier for students to access a wealth of information. However, this information explosion has also introduced challenges in how students find, evaluate, and effectively use this information. Exploratory search, a type of information-seeking activity, has emerged as a solution to these challenges. Unlike targeted search, which involves looking for specific goals and expecting specific results, exploratory search involves initially undefined and ever-changing information needs.
This study aims to address the complexity of information retrieval in the context of large language models (LLMs) by combining exploratory search strategies with exploratory learning theories. By adapting Kolb’s learning model to incorporate high-frequency exploration and feedback loops, the study seeks to promote deep cognitive and higher-order cognitive skill development in students. Additionally, the paper discusses how advanced LLMs can support students in their exploratory searches, thereby enhancing student-computer interaction and supporting their learning journeys in the new era of LLMs.
Related Work
Exploratory Learning
Exploratory learning is an educational theory that emphasizes active exploration, discovery, and reflection to acquire new knowledge and solve problems. It involves iterative learning processes, deep cognitive engagement, and the integration of real-world contexts and social interactions. Classic instructional design theories of exploratory learning include constructivist teaching and Inquiry-Based Learning (IBL). Constructivism posits that learning is a process based on learners’ prior knowledge and experiences, involving the interaction and reconstruction of new information to build new knowledge and understanding. Inquiry-Based Learning (IBL) is a student-centered learning method that builds a knowledge system through posing questions, conducting investigations, collecting and analyzing data, communicating and sharing, and reflecting and evaluating.
Kolb’s experiential learning cycle, which involves experiencing, reflecting and observing, forming abstract concepts, and testing, is a foundational model in exploratory learning. However, traditional exploratory learning often faced skepticism due to its reliance on external interaction and real-world experiences, which were limited in the early stages of educational digitalization.
Development of Large Language Models in Information Retrieval
The rapid development of deep learning technology in artificial intelligence, especially the emergence of large language models like ChatGPT, has brought new advancements to the application of exploratory search strategies in the learning process. Early models based on natural language processing (NLP) had limited abilities to understand complex language structures. However, the transition from statistical language models to neural language models offered a pathway to overcoming these challenges. The publication of ‘Attention is All You Need’ in 2017 marked a significant turning point in the field of language modeling, introducing the Transformer architecture, which includes a self-attention mechanism.
Large language models (LLMs) are state-of-the-art artificial intelligence systems designed to process and generate text, with a focus on coherent communication. Trained in a self-supervised environment on extensive text corpora, they learn complex language patterns and structures, enabling a qualitative leap in performance in translation, summarization, information retrieval, and natural dialogue interactions.
Students’ Struggles in Information Search
Numerous studies indicate that students struggle with using search to learn and solve problems. They often exhibit blind confidence in search results, cognitive biases in evaluating information, and difficulties in appropriately applying search strategies in specific scenarios. Although the concept of “search as a learning process to support and improve human learning” was proposed long ago, there is still a lack of research linking the two fields. Additionally, the inconsistent evaluation standards for exploratory search systems may lead to students feeling lost and unsure when trying to internalize exploratory search strategies as learning strategies.
Research Methodology
To investigate the relationship between exploratory learning and exploratory search strategies, the study performed a literature analysis using CiteSpace and LDAvis to visualize the literature keywords and abstracts from the past ten years. The aim was to explore the connection between these two concepts to further enrich and develop school theories, information retrieval, and educational research in the era of large models.
Relevant literature was retrieved from the Web of Science database using “Exploratory Search” and “Exploratory Learning” as keywords. The top 200 relevant articles from 2014 to 2024 were searched, and after removing duplicate articles, a total of 284 relevant papers were retrieved. These documents were imported into CiteSpace to explore the research trends and interrelationships in these two fields. The specific settings included time slices from 2004 to 2024, one slice per year; node type as keywords; selection criteria as the Top 50 high-frequency keywords; visualization settings using Cosine similarity measurement, and the link retention factor as 2.5 times the number of nodes.
To find deeper internal links in the field of education, especially in the sub-field of learning theory, the WOS search strategy was readjusted, initially using “Exploratory Search” and “education” as keywords to collect the top 100 most relevant papers, and then using “Inquiry-Based Learning” and “education” to repeat the same collection operation. LDAvis was used to perform topic modeling on the abstracts of the articles to measure the connections between the subfields within the field of education in a more quantitative way.
Experimental Design
Planning and Designing the Experiment
The experiment involved a comprehensive literature analysis using CiteSpace and LDAvis to visualize the literature keywords and abstracts from the past ten years. The goal was to explore the connection between exploratory learning and exploratory search strategies to further enrich and develop school theories, information retrieval, and educational research in the era of large models.
Preparing the Data and Conditions
Relevant literature was retrieved from the Web of Science database using “Exploratory Search” and “Exploratory Learning” as keywords. The top 200 relevant articles from 2014 to 2024 were searched, and after removing duplicate articles, a total of 284 relevant papers were retrieved. These documents were imported into CiteSpace to explore the research trends and interrelationships in these two fields. The specific settings included time slices from 2004 to 2024, one slice per year; node type as keywords; selection criteria as the Top 50 high-frequency keywords; visualization settings using Cosine similarity measurement, and the link retention factor as 2.5 times the number of nodes.
To find deeper internal links in the field of education, especially in the sub-field of learning theory, the WOS search strategy was readjusted, initially using “Exploratory Search” and “education” as keywords to collect the top 100 most relevant papers, and then using “Inquiry-Based Learning” and “education” to repeat the same collection operation. LDAvis was used to perform topic modeling on the abstracts of the articles to measure the connections between the subfields within the field of education in a more quantitative way.
Results and Analysis
Keyword Clustering and Mutation Detection
The results of the keyword clustering analysis indicated that “Exploratory Search” and “Exploratory Learning” formed independent clusters, suggesting that these two concepts have few connections in the existing literature. Additionally, other main clusters, such as Knowledge Base, Exploratory Product Search, and Exploratory Motivation, also showed their independence. The link found from the keyword clustering analysis is the Knowledge Base cluster, which mainly covers knowledge acquisition, storage, and management research, especially in information systems and knowledge discovery. This may indicate that both the concepts of exploratory learning and exploratory search involve knowledge acquisition, information processing, and iterative feedback.
Burst word detection revealed 16 keywords with the strongest bursts. These keywords displayed burst characteristics within specific periods. For example, “Exploratory Learning” showed a strong burst during 2007-2016, while “Exploratory Search” displayed a burst during 2017-2018. Although both exhibited bursts in certain years, their burst periods mostly did not overlap, further supporting the independence of the two concepts in the literature.
Subject Clustering
Both the concepts of exploratory search and exploratory learning pointed mostly to the fields of information systems and computing in the clustering, while content related to the field of education was minimal. The keywords in the obtained nodes also support this point. The results of the subject clustering of the literature’s keywords clearly show that there is no overlap between research in the education field and research in the computing and information retrieval fields. Additionally, using LDAvis as a tool to visualize topic modeling on the abstracts of all the articles again revealed that when re-analyzing the abstracts’ topic modeling, none of the topics overlapped, and the topic emphasizing students’ exploratory information retrieval and search behavior was the farthest from other topics. All these pieces of evidence suggest that research related to learning theory in the field of education has not been closely linked with knowledge in the fields of computing and information retrieval.
Improved Exploratory Learning Process
Exploratory learning enhanced by exploratory search strategies has two characteristics that are not present in traditional exploratory learning processes. One of the core characteristics of exploratory search is the high frequency of internal and external interactions. The traditional exploratory learning process is usually goal-oriented or divergent, involving fewer non-linear iterations and adjustments. However, in the learning process guided by the exploratory search strategy, learners need to conduct a series of complex, non-goal-oriented searches between information retrieval systems and provide feedback, re-entering the internal exploration loop. Through these iterative exploration and feedback cycles, learners can conduct deeper analysis and evaluation of information.
Another core characteristic is the deep exploration and feedback loop. Although traditional inquiry-based learning does not usually rely on highly structured evaluation models to measure school effectiveness, it still focuses on phased conceptual results. Compared to the traditional exploratory learning process, exploratory search allows learners to move to the next exploration step after each feedback without having to form new concepts immediately. It provides learners with more space to explore deep information and helps them to continuously adjust and optimize their cognitive framework during the exploration process.
Overall Conclusion
With the advent of the information age, effectively finding, evaluating, and utilizing the overwhelming amount of information for learning and reflection have become challenging issues for learners. Despite the immense potential of exploratory search strategies in education, there is a scarcity of research that combines the advanced concepts of exploratory search with learning theories in the educational field. Meanwhile, with the introduction of generative artificial intelligence technologies such as LLMs, how to properly train students to be more exploratory has also become a thorny issue.
This study attempts to address this gap by integrating exploratory search strategies from information retrieval into the learning process from the perspective of student learning. This approach aims to form a novel exploratory learning model that combines exploratory search with learning theories. The model adapts Kolb’s learning model to the information age, focusing on high-frequency exploration and feedback loops, which help students explore new information under uncertainty and promote deep cognitive and higher-order cognitive skill development.
The paper also discusses the opportunities and challenges of exploratory learning theory in the era of LLMs. With the deepening application of LLM in the field of education, future research will further strengthen the relationship between exploratory search and the learning process through more examples to verify the accuracy of the exploratory learning model proposed. Through interdisciplinary research and practice, the goal is to promote the common development of educational theory and information retrieval theory, ultimately improving students’ problem-solving abilities in the information age.
