Semantic-Based Expert Identification in Community Question Answering Platforms
Community Question Answering (CQA) are the platforms where users can ask and answer questions on variety of domains. Identifying the expert for a new question is becoming difficult as the number of users and content are increasing in large number. Semantic approaches play a crucial role by capturing contextual meaning between questions which enables more accurate matching of queries to suitable experts beyond simple keyword overlap. Traditional approaches like TF-IDF and BM25 techniques provide lexical keyword matching only and fail to show the semantic meaning of technical queries. This paper explores the effects of different textual representation strategies on expert identification by comparing five different retrieval-based techniques on a Stack Overflow dataset within a similar retrieval and selection framework to maintain uniformity and fairness. Two lexical keyword-based (TF-IDF and BM25) and three dense embedding-based models that are Sentence BERT (all-mpnet-base-v2), BGE-large (BAAI/bge-large-en), and CodeBERT have been compared. To evaluate the performance various metrics such as Hit Rate, Mean Reciprocal Rank (MRR) and Normalised Discounted Cumulative Gain (nDCG) has been used. The results indicate that dense embedding models have performed better than traditional keyword methods. Out of all, BGE-large has achieved the highest performance and showed strong semantic discrimination in expert ranking. CodeBERT have shown lower performance and indicates that models pre-trained on code do not provide an advantage for question-question matching in CQA. These results indicate the significance of dense semantic embeddings for expert identification in CQA systems.
Authors
- Iram Fatima (ORCID: https://orcid.org/0000-0002-5104-8958)
- Seema
Institutions
- University of Delhi (IN)
Publication Details
- Journal
- International Journal of Innovation and Multidisciplinary Research
- Published
- 2026-11-15
- DOI
- https://doi.org/10.67505/ijiamr.v6.i2.62007
- Primary Topic
- Expert finding and Q&A systems
- Type
- article
- Field-Weighted Citation Impact
- 0.00