Genuine Information Needs of Social Scientists Looking for Data
研究针对社会科学者在数据搜索时遇到的信息需求与现有系统能力不匹配的问题，通过在线调查分析了72名研究人员的具体需求，并将其与现有元数据模型及搜索系统功能进行了对比。
0 citationsRead paper
Selected work
Representative Papers
Topic Matters: How Linguistic Properties can Shape Reading Behaviour in Selective Exposure Studies
This study addresses a critical oversight in selective exposure research: the assumption that texts across different controversial topics are comparable once basic variables are controlled, despite inherent linguistic differences that may systematically influence reading behavior. Combining eye-tracking and natural language processing, the authors quantitatively analyzed gaze patterns and textual features among 68 participants reading news articles on climate change and immigration policy. Results reveal significant topic-level differences in both linguistic properties—such as sentiment valence and syntactic complexity—and attentional allocation, demonstrating that topic-specific linguistic characteristics constitute a methodological confound that cannot be ignored in selective exposure studies. These findings provide crucial empirical grounding for designing more precise interventions aimed at mitigating information diet biases.
0 citationsRead paper
Continuous Online Evaluation of Recommendation Strategies in Social Science Academic Search
This study addresses the challenge of recommendation relevance in academic search engines, which arises from disciplinary diversity, heterogeneous resources, and varying user preferences. To tackle this, we integrated three recommendation strategies—term-based similarity, Transformer-derived semantic embeddings, and session-based click-path modeling—into the GESIS Search platform, marking the first real-time, continuous online evaluation of a multi-strategy recommender system in the social sciences. Leveraging the STELLA framework for ongoing tracking of real user interactions, our findings demonstrate that semantic embedding methods consistently outperform both term-based and session-path approaches. Moreover, user interaction patterns vary significantly across resource types—such as publications, datasets, and software—highlighting the critical influence of resource category on recommendation effectiveness.
0 citationsRead paper
Bytes of a Feather: Personality and Opinion Alignment Effects in Human-AI Interaction
This study investigates how AI assistants’ personality traits (e.g., extraversion vs. introversion) and ideological stances (e.g., political orientation, values) jointly shape user preferences and perceptions. A large-scale controlled experiment with 1,000 participants employed machine learning–driven AI personalization simulations to systematically assess the effects of opinion alignment and personality congruence on perceived trustworthiness, competence, likability, and persuasiveness. Results demonstrate that ideological alignment is the dominant driver of user preference and trust—significantly enhancing perceived credibility, competence, and persuasive efficacy—strongly supporting the “similarity-attraction” hypothesis. In contrast, personality matching exerts only marginal, domain-specific effects, with negligible impact across most outcome measures. These findings delineate critical boundaries in AI personalization design: ideological alignment exhibits robust, generalizable effectiveness, whereas personality alignment shows limited utility and potential unintended consequences. The study thus provides empirical grounding and theoretical refinement for designing trustworthy, human-centered AI interactions.
0 citationsRead paper
Computational Reproducibility of R Code Supplements on OSF
This study addresses the widespread lack of computational reproducibility in R supplementary code deposited on the Open Science Framework (OSF). A systematic audit of 296 published R code packages revealed that 98.8% incompletely declare dependencies. To address this, we propose the first automated reproducibility auditing framework tailored to the R ecosystem. It combines static source-code analysis—leveraging regular expressions and abstract syntax trees (ASTs)—to accurately infer dependencies, with Docker-based containerized execution and failure diagnostics (e.g., path errors, OS-specific inconsistencies, missing packages) to enable end-to-end environment reconstruction and validation. Experiments successfully executed 25.87% of scripts, identifying undeclared dependencies, hardcoded file paths, and cross-platform compatibility issues as the three primary barriers to reproducibility. The framework enables large-scale, low-cost, and scalable quantitative assessment of computational reproducibility in scholarly research, providing a practical toolchain to enhance transparency and verifiability.
0 citationsRead paper
Recent publications
Latest Papers
Genuine Information Needs of Social Scientists Looking for Data
研究针对社会科学者在数据搜索时遇到的信息需求与现有系统能力不匹配的问题，通过在线调查分析了72名研究人员的具体需求，并将其与现有元数据模型及搜索系统功能进行了对比。
0 citationsRead paper
Topic Matters: How Linguistic Properties can Shape Reading Behaviour in Selective Exposure Studies
This study addresses a critical oversight in selective exposure research: the assumption that texts across different controversial topics are comparable once basic variables are controlled, despite inherent linguistic differences that may systematically influence reading behavior. Combining eye-tracking and natural language processing, the authors quantitatively analyzed gaze patterns and textual features among 68 participants reading news articles on climate change and immigration policy. Results reveal significant topic-level differences in both linguistic properties—such as sentiment valence and syntactic complexity—and attentional allocation, demonstrating that topic-specific linguistic characteristics constitute a methodological confound that cannot be ignored in selective exposure studies. These findings provide crucial empirical grounding for designing more precise interventions aimed at mitigating information diet biases.
0 citationsRead paper
Continuous Online Evaluation of Recommendation Strategies in Social Science Academic Search
This study addresses the challenge of recommendation relevance in academic search engines, which arises from disciplinary diversity, heterogeneous resources, and varying user preferences. To tackle this, we integrated three recommendation strategies—term-based similarity, Transformer-derived semantic embeddings, and session-based click-path modeling—into the GESIS Search platform, marking the first real-time, continuous online evaluation of a multi-strategy recommender system in the social sciences. Leveraging the STELLA framework for ongoing tracking of real user interactions, our findings demonstrate that semantic embedding methods consistently outperform both term-based and session-path approaches. Moreover, user interaction patterns vary significantly across resource types—such as publications, datasets, and software—highlighting the critical influence of resource category on recommendation effectiveness.
0 citationsRead paper
Bytes of a Feather: Personality and Opinion Alignment Effects in Human-AI Interaction
This study investigates how AI assistants’ personality traits (e.g., extraversion vs. introversion) and ideological stances (e.g., political orientation, values) jointly shape user preferences and perceptions. A large-scale controlled experiment with 1,000 participants employed machine learning–driven AI personalization simulations to systematically assess the effects of opinion alignment and personality congruence on perceived trustworthiness, competence, likability, and persuasiveness. Results demonstrate that ideological alignment is the dominant driver of user preference and trust—significantly enhancing perceived credibility, competence, and persuasive efficacy—strongly supporting the “similarity-attraction” hypothesis. In contrast, personality matching exerts only marginal, domain-specific effects, with negligible impact across most outcome measures. These findings delineate critical boundaries in AI personalization design: ideological alignment exhibits robust, generalizable effectiveness, whereas personality alignment shows limited utility and potential unintended consequences. The study thus provides empirical grounding and theoretical refinement for designing trustworthy, human-centered AI interactions.
0 citationsRead paper
Computational Reproducibility of R Code Supplements on OSF
This study addresses the widespread lack of computational reproducibility in R supplementary code deposited on the Open Science Framework (OSF). A systematic audit of 296 published R code packages revealed that 98.8% incompletely declare dependencies. To address this, we propose the first automated reproducibility auditing framework tailored to the R ecosystem. It combines static source-code analysis—leveraging regular expressions and abstract syntax trees (ASTs)—to accurately infer dependencies, with Docker-based containerized execution and failure diagnostics (e.g., path errors, OS-specific inconsistencies, missing packages) to enable end-to-end environment reconstruction and validation. Experiments successfully executed 25.87% of scripts, identifying undeclared dependencies, hardcoded file paths, and cross-platform compatibility issues as the three primary barriers to reproducibility. The framework enables large-scale, low-cost, and scalable quantitative assessment of computational reproducibility in scholarly research, providing a practical toolchain to enhance transparency and verifiability.
0 citationsRead paper