基于改进协同过滤算法的个性化新闻推荐技术

Personalized NewsRecommendation Technology Based on Improved Collaborative Filtering Algorithm

摘要:

针对传统的基于内容协同过滤算法只是依据用户历史访问矩阵向用户做出推荐，存在数据稀疏以及不能及时反映用户兴趣变化等问题，个性化新闻推荐技术在传统的协同过滤算法基础上提出了新闻文本内容相似度的计算方式和时间窗的概念，新闻内容相似度计算中还考虑了特征词的词性和在新闻中的位置的影响，时间窗用来建立适应用户兴趣随时间变化的模型；实验结果表明，改进后的算法有效地改善了新闻用户历史访问数据的稀疏问题，及时捕获用户兴趣，F-measure值相比传统的算法最大提高了10%，平均绝对误差值最高下降了7%，显著提高了推荐质量。

Abstract:

The traditional collaborative filtering algorithm only based on matrix produced by user access history to make recommendation and sparse data,and also cannot reflect the user’s interests timely, contrary to these problems, the personalized recommendation technology news in the traditional collaborative filtering algorithm proposes the calculation of news text content similarity and the concept of the time window , the calculation of news content similarity also takes into account the part of speech and positions of the feature words in the news, the time window is used to create user interest model which will change over time; The experimental results show that the improved algorithm effectively improves the sparse problem of data which user has accessed and captures user interest timely, F-measure value improves the maximum 10% compared to the traditional algorithm, the highest value of mean absoulte error fell by 7%, greatly improving the quality of recommendation.