• community question&answering
• expert finding
• fairness
• gender bias
• social search

This thesis advances the Social Search field by developing novel approaches for leveraging implicit social information in community Q&A (CQA) platforms, with a particular focus on the Expert Finding task and gender bias. We first present a comprehensive taxonomy of Social Search systems, identifying key challenges and opportunities in modeling social dimensions of user behavior. We then introduce TUEF, a novel framework that represents CQA platforms as multi-layer graphs which combines content-based analysis with social network exploration.
Given the documented gender disparities in technical CQA platforms, we conduct two complementary analyses focused on Stack Overflow. First, we evaluate how Expert Finding algorithms impact gender representation in their outcomes. Our findings reveal that while content-based components tend to favor male users who are more active, social network components show a counterbalancing effect by capturing women's stronger relationship-building tendencies.
Second, we conduct a qualitative assessment of answer quality by gender to investigate potential biases in the selection of best answers. We first perform human evaluations to assess the accuracy of the gender inference tool and to measure the alignment of large language models (LLMs) with human judgments. Finally, we leverage LLM-based evaluation on a large-scale dataset, revealing that observed disparities in user recognition are primarily driven by differences in participation patterns.