基于用户兴趣的文献个性化推荐研究

针对当前文献推荐中个性化程度不高等问题，提出一种对用户行为重新分配权重的度量算法。运用用户行为数据按照时间顺序重新分配权重，突出近期用户兴趣构建用户兴趣模型。通过LDA主题分布、关键词分布等方法构建学术资源模型，实现两模型间匹配，完成推荐。通过实验验证，该算法准确性达到80%，比传统等权重算法提高近20%，召回率与F值分别提升了7%和5%。研究表明，基于时间因素的用户兴趣度量算法相较于传统等权重算法具有更高的准确性，未来可进一步优化用户兴趣度量以实现精准推荐服务。

Personalized Recommendation of Literature Based on User Interest

To solve the problem of low degree of personalization in current literature recommendation， a measurement algorithm for reallocating weights to user behaviors is proposed. The paper uses the user behavior data to redistribute the weight according to the time sequence， and highlights the recent user interest to build the user interest model. Through LDA topic distribution， keyword distribution and other methods to build academic resource model， the two models were matched and the recommendation was completed. The experimental results show that the accuracy of the algorithm reaches 80%， which is nearly 20% higher than the traditional equal weight algorithm. The recall rate and F value are increased by 7% and 5% respectively. The user interest measurement algorithm based on time factor has higher accuracy than the traditional equal weight algorithm. In the future， the user interest measurement can be further optimized to provide ideas for optimizing accurate recommendation services.