Predicting Best Answerers for New Questions: An Approach Leveraging Convolution Neural Networks in Community Question An
Community Question Answering (CQA) websites are becoming increasingly important sources of information where users can share knowledge on various topics. These websites provide many opportunities for users to seek for help and provide answers, but they al
- PDF / 638,989 Bytes
- 13 Pages / 439.37 x 666.14 pts Page_size
- 21 Downloads / 164 Views
)
Dalian University of Technology, No. 2 LingGong Road, GanJingZi District, Dalian, China [email protected]
Abstract. Community Question Answering (CQA) websites are becoming increasingly important sources of information where users can share knowledge on various topics. These websites provide many opportunities for users to seek for help and provide answers, but they also bring new challenges. One of the challenges is that most new questions posted everyday cannot be routed to appro‐ priate users who can answer them. It means that experts are not provided with questions matching their expertise, and therefore new questions cannot be answered in time. Our main goal is to find which user has more potential to be the best answerer for a newly posted question. In this paper, we propose an approach which based on convolutional neural networks (CNN) to predict the best answerer for a new posted question on CQA websites. We have applied our model on the dataset downloaded from StackOverflow, one of the biggest CQA sites. The results show that our approach performs better than Segmented Topic Model. Keywords: Community question answering · Convolutional neural networks · Expert recommendation · Best answerer · Segmented topic model
1
Introduction
Community Question Answering (CQA) websites receive millions of questions and provide high quality answers to users’ questions by cooperating with experts in the community. General sites like Yahoo! Answers, Baidu Knows, Quora and domainspecific CQA websites like Mathematics and StackOverflow have attracted millions of users. In some CQA websites, voting, badges and reputation are provided to measure the quality of answers and answerers. In CQA websites, a user who submits her or his questions need to wait for other users to answer, which may take several days and the answers sometimes tend to be incorrect, useless or offensive, otherwise the user can use the history profiles of CQA websites which need to deal with word-matching between posted questions and archived ques‐ tions. In the second method, those questions may have no relevant answers and in this case the user has to seek for other supports.
© Springer Nature Singapore Pte Ltd. 2016 Y. Li et al. (Eds.): SMP 2016, CCIS 669, pp. 29–41, 2016. DOI: 10.1007/978-981-10-2993-6_3
30
J. Wang et al.
In CQA sites one of the main problem is the low participation. Only a small part of users are responsible for answering a majority of questions. There are two reasons for low participation, one is a majority of users are not willing to answer questions and the other is users are willing to answer questions, but they cannot find the new questions of interest to them. Therefore, it is necessary to route questions to experts who have high probability to be best answerers. The research can reduce user’s waiting time and make valuable contributions to receive high quality answers. The expert-finding task is primarily addressed by modeling the expertise of a user based on his or her answering history. Generally there are two methods for
Data Loading...
