Evolution of Research Method Usage Across the Academic Careers of Library and Information Science Scholars

1. Problem Statement

The study addresses a gap in understanding how research method selection evolves throughout a scholar's career. While previous literature has examined research methods in isolation or analyzed academic age regarding productivity and collaboration, few studies have integrated academic age with methodological practices.

• The Gap: Junior scholars often lack guidance on method selection, sometimes blindly following trends without aligning methods to research questions. Senior scholars may rely on outdated concepts or established habits.
• The Objective: To systematically investigate the longitudinal patterns of research method adoption, diversity, and combination among Library and Information Science (LIS) scholars across different career stages (early, mid, and senior).

2. Methodology

The study employs a mixed-method approach combining large-scale bibliometric data, deep learning, and statistical analysis.

A. Data Collection and Preprocessing

• Corpus: 26,677 full-text journal articles published between 1990 and 2023 from 14 representative LIS journals (e.g., JASIST, Scientometrics, IPM).
• Author Disambiguation: Utilized OpenAlex to uniquely identify authors via DOIs.
• Academic Age Calculation: Defined as the time elapsed since a scholar's first publication ($AAS = PYA - EPY + 1$).
• Sample Selection:
  • Filtered for scholars who are first or corresponding authors.
  • Removed outliers using the Interquartile Range (IQR) method.
  • Final Cohort: 435 senior scholars (Academic Age > 14 years) with consistent publication records at five-year intervals, totaling 6,116 articles.
• Classification: Scholars were categorized into three groups: Young (<7 years), Middle-aged (7–14 years), and Senior (>14 years).

B. Automated Method Classification

• Model: Employed the CogFT (Cognize Full Text) model, a deep learning architecture adapted from the CogLTX model.
• Architecture:
  • Full texts are segmented into chunks.
  • SciBERT embeds text chunks into 768-dimensional vectors.
  • Self-Attention mechanisms capture relationships between chunks to predict method summaries.
  • A standard BERT classifier assigns one of 16 predefined method labels.
• Classification System: Based on Chu & Ke (2017), covering 16 categories (e.g., Bibliometrics, Content Analysis, Experiment, Questionnaire, Webometrics, etc.).
• Performance: The model achieved an F1 score > 0.85.

C. Topic Modeling

• Tool: Top2Vec (unsupervised deep learning topic modeling).
• Output: Identified 20 distinct research topics (e.g., Scientometrics, Information Retrieval, Social Media Analysis) based on abstracts.

D. Statistical Analysis

• Chi-Square Test: Used to determine significant differences in method selection frequencies across the three academic age groups.
• Visualization: Heatmaps, 3D scatter plots (Publication Count vs. Method Diversity vs. Topic Diversity), and bubble charts were used to map method-topic relationships.

3. Key Contributions

1. Integration of Metrics: Successfully linked academic age with research method usage using a large-scale, automated corpus, moving beyond manual coding limitations.
2. Advanced NLP Application: Demonstrated the efficacy of the CogFT model for document-level research method identification in full-text academic papers, overcoming the noise and length challenges of traditional deep learning models.
3. Longitudinal Career Mapping: Provided a granular view of how method diversity and specific method preferences shift from early career exploration to mid-career diversification and senior career stabilization.
4. Method-Topic Correlation: Mapped specific research methods to specific research topics across different career stages, revealing how methodological choices are context-dependent.

4. Key Results

A. Method Diversity Trends

• Inverted U-Shape: The diversity of research methods used by scholars follows an inverted U-curve over their careers.
  • Early Career: Moderate diversity as scholars experiment.
  • Mid-Career (7–14 years): Peak diversity. Scholars utilize the widest variety of methods and combine them most frequently.
  • Senior Career (>14 years): Diversity declines as scholars stabilize around core, high-impact methods.

B. Dominant Methods by Career Stage

• Bibliometrics: The most frequently used method across all stages, but its dominance increases with seniority (19.61% for early-career vs. 31.81% for senior scholars).
• Webometrics: Highly favored by young scholars (early career), driven by technological affinity.
• Qualitative/Mixed Methods: Senior scholars show a propensity for combining methods (e.g., pairing conventional with unconventional approaches).
• Shifts: Traditional methods like "Theoretical Approach" and "Interview" have seen declining usage proportions in recent years, while data-driven methods (Bibliometrics, Webometrics) have risen.

C. Method-Topic Interactions

• Young Scholars: Focus on algorithmic and retrieval topics (Information Retrieval, Search Behavior) using Webometrics and Experiments.
• Mid-Career Scholars: Act as a nexus, bridging classical topics (Scientometrics) with emerging ones (Text Mining) using diverse method combinations.
• Senior Scholars: Focus on theoretical refinement and practical consolidation (Information Theory, Library Policy) using established methods like Bibliometrics and Content Analysis.

D. Individual Case Studies

Analysis of four senior scholars (e.g., Mike Thelwall, Amanda Spink) confirmed that:

• Mid-career is a period of methodological expansion and experimentation.
• Senior scholars often return to a core set of methods but apply them with greater sophistication and topic breadth.

5. Significance and Implications

Theoretical Implications

• Social Cognitive Theory: The findings support Bandura's theory, suggesting that method selection is a dynamic interaction between cognition, behavior, and environment. Early career involves observational learning and experimentation; seniority leads to self-regulated habit formation and specialization.
• Disciplinary Evolution: The study highlights the shift in LIS from conceptual to empirical and data-driven research strategies over the last three decades.

Practical Implications

• For Junior Scholars: Provides a roadmap for methodological development. Instead of blindly imitating senior scholars' current outputs, young researchers should understand the evolutionary path of methods, allowing them to plan targeted learning (e.g., mastering Webometrics early, transitioning to mixed methods mid-career).
• For Institutions: Suggests the need for stage-specific training programs.
  • Early Stage: Focus on foundational and emerging tech methods.
  • Mid-Stage: Encourage method combination and interdisciplinary expansion.
  • Senior Stage: Focus on theoretical synthesis and mentoring.
• Collaboration: Encourages inter-generational collaboration where senior scholars provide theoretical framing and junior scholars introduce new technical methods.

Limitations

• Sample Bias: Restricted to 14 journals and senior scholars with long publication histories; excludes conference proceedings and monographs.
• Classification Rigidity: The Chu & Ke (2017) framework may not fully capture emerging AI/ML-driven methods or complex interdisciplinary hybrids.
• Causality: The study identifies correlations but does not deeply probe underlying causal factors like funding changes or specific journal review policies.

Conclusion

The paper concludes that research method usage in LIS is not static but evolves dynamically with academic age. The trajectory moves from exploration (young) to diversification and integration (mid-career) and finally to specialized synergy (senior). Understanding these patterns is crucial for optimizing researcher training, funding allocation, and the strategic development of the LIS discipline.