User-based Question Recommendation for Question Answering System

An approach to automatically recommending question based on user-word model is proposed. We first employ language modeling approach to map the relationship between a user and a question into the relationship between the user and words in the question. A use-word model is then designed to reveal and quantify the affinity relationship between users and words in the corpus. In the recommendation model, a new question is assigned to users based on the evaluation of question-user relationship. The user who has the strongest relationship with the question is recommended to answer the question. We also implement an incremental update model which can dynamically maintain the user-word model. 216,563 questions (spreading into 30 categories) from Yahoo! Answers are collected as dataset and preliminary experiments show our approach achieves question recommendation accuracy by 85.2%, which exceeds baseline methods.