Abstract: Recommendation system has played an important role in developer community Stack Overflow, Zhihu, Baidu Know and other popular question and answer communities, and will become the key technology for Haichuan Chemical Forum to improve the efficiency of question and answer. As the largest chemical question-and-answer community in China, Haichuan Chemical Forum is unable to get timely and effective answers due to two major difficulties: sparsity and cold start. This paper presents a hybrid recommendation method combining DeepFM and matrix decomposition. The algorithm takes DeepFM as the auxiliary algorithm and matrix decomposition as the main algorithm. By combining the user’s personal characteristics and the problem’s own characteristics, it recommends suitable respondents for new problems in the forum, effectively solving the problem redundancy in the community. By calculating the root-mean-square error and mean absolute error of the test set, the validity and feasibility of the proposed method in Haichuan Chemical Forum are further verified.

Key words: recommendation system, mixed recommendation, matrix decomposition, respondent recommendation