Research On Answer Recommendation Method For Qusetion And Answering Community

A social platform is provided by the Question and Answer community for users to solve difficult problems and share knowledge,experience,and insights.Users don't have to worry about how to extract keywords to express their information needs.A huge knowledge base is formed in the Question and Answer community because it has a wealth of content and wide topics.And the problems in the Question and Answer community are more about seeking opinions,suggestions,and discussing questions in order to collect opinions.Valuable information is contained in each answer of the problem.So,this makes the Question and Answer community need intelligent and accurate answer recommendation services to solve the “information overload” problem in the current Question and Answer community.Recently,research of answer recommendation for the Question and Answer community was mainly based on the quality of the answers,user interest bias was not fully considered in the research.At the same time,some feature of answer,such as the author's authority,the relativity of question and answer,was considered in the research.Research is less involved in the semantic dimension of the answer's text.Therefore,an answer recommendation method for the Question andAnswer community is proposed in this paper.First,in order to reduce calculations of the recommended method,the improved UserCF is proposed to obtain a candidate set for target user.Second,the improved LDA topic model is proposed,is used to model user's interest.The improvement of the method is aimed at the existence of meaningless topics in the original LDA topic model,considering the low similarity between the words in the meaningless topic distribution.The model automatic filtering of meaningless topics,eliminating the influence of meaningless topics on building user interest vectors.Third,a method for representing user interest vector and answer text in multi-dimensional semantic space is proposed.The Euclidean distance is used to measure the semantic similarity between them.Forth,the answer feature is processed,and the answer score calculation method is proposed to provide a basis for sorting the answer recommendation results.Fifth,the above intermediate results are applied to the answer recommendation method.The method consider user interest migration,the semantic similarity between the answer text and the user interest vector and the answer score,which make the result of the answer recommendation is more in line with the user's interest bias.Finally,with using the real dataset from Zhihu,the answer recommendation method,which is proposed in this paper,is compared with the LDA-based recommendation method and the Skip-gram-based recommendation method.The results show that the method proposed in thispaper can satisfies the interest bias of the user.The method achieves a good effect.