FuEPRe: a Fusing Embedding Method with Attention for Post Recommendation

Post recommendations refer to finding solutions related to a user’s problem on QA websites to help them solve their problems. However, finding the most relevant post from a large number of posts related to a problem is a challenging task. This paper proposes a novel recommendation model called FuEPRe, which based on a multi-headed self-attention network integrates semantic information, structural information of code and description information. It accurately recommends relevant Stack Overflow posts based on users’ queries, thereby helping them solve problems quickly and solving the problem of inaccurate post recommendations in the past. Each pair of codes and descriptions is represented as two vectors, and then, the three different types of information are fused into these two vectors through an attention mechanism. At this point, each vector contains the above three types of information and then recommends posts by comparing the similarity between the vectors. The proposed approach is evaluated on the Stack Overflow Posts dataset, and the results demonstrate that it outperforms some state-of-the-art methods in the post recommendation task. Specifically, the approach improves the recall, MRR, and NDCG of recommendations, enabling programmers to solve problems faster.