Hierarchical Expert Recommendation on Community Question Answering Platforms
| dc.contributor.advisor | Lee, Roy Ka-Wei | |
| dc.contributor.advisor | Vassileva, Julita | |
| dc.contributor.committeeMember | Deters, Ralph | |
| dc.contributor.committeeMember | Horsch, Michael | |
| dc.contributor.committeeMember | Li, Zhi | |
| dc.creator | Jalali, Amirabbas | |
| dc.creator.orcid | 0000-0001-9956-2481 | |
| dc.date.accessioned | 2022-03-29T20:17:12Z | |
| dc.date.available | 2022-03-29T20:17:12Z | |
| dc.date.created | 2022-03 | |
| dc.date.issued | 2022-03-29 | |
| dc.date.submitted | March 2022 | |
| dc.date.updated | 2022-03-29T20:17:12Z | |
| dc.description.abstract | The community question answering (CQA) platforms, such as Stack Overflow, have become the primary source of answers to most questions in various topics. CQA platforms offer an opportunity for sharing and acquiring knowledge at a low cost, where users, many of whom are experts in a specific topic, can potentially provide high-quality solutions to a given question. Many recommendation methods have been proposed to match questions to potential good answerers. However, most existing methods have focused on modelling the user-question interaction — a user might answer multiple questions and a question might be answered by multiple users — using simple collaborative filtering approaches, overlooking the rich information in the question’s title and body when modelling the users’ expertise. This project fills the research gap by thoroughly examining machine learning and deep learning approaches that can be applied to the expert recommendation problem. It proposes a Hierarchical Expert Recommendation (HER) model, a deep learning recommender system that recommends experts to answer a given question in the CQA platform. Although choosing a deep learning over a machine learning solution for this problem can be justified considering the degree of complexity of the available datasets, we assess performance of each family of methods and evaluate the trade-off between them to pick the perfect fit for our problem. We analyzed various machine learning algorithms to determine their performances in the expert recommendation problem, which narrows down the potential ways for tackling this problem using traditional recommendation methods. Furthermore, we investigate the recommendation models based on matrix factorization to establish the baselines for our proposed model and shed light on the weaknesses and strengths of matrix- based solutions, which shape our final deep learning model. In the last section, we introduce the Hierarchical Expert Recommendation System (HER) that utilizes hierarchical attention-based neural networks to rep- resent the questions better and ultimately model the users’ expertise through user-question interactions. We conducted extensive experiments on a large real-world Stack Overflow dataset and benchmarked HER against the state-of-the-art baselines. The results from our extensive experiments show that HER outperforms the state-of-the-art baselines in recommending experts to answer questions in Stack Overflow. | |
| dc.format.mimetype | application/pdf | |
| dc.identifier.uri | https://hdl.handle.net/10388/13855 | |
| dc.subject | Expert Recommendation | |
| dc.subject | Hierarchical Networks | |
| dc.subject | Deep Learning | |
| dc.title | Hierarchical Expert Recommendation on Community Question Answering Platforms | |
| dc.type | Thesis | |
| dc.type.material | text | |
| thesis.degree.department | Computer Science | |
| thesis.degree.discipline | Computer Science | |
| thesis.degree.grantor | University of Saskatchewan | |
| thesis.degree.level | Masters | |
| thesis.degree.name | Master of Science (M.Sc.) |