情报科学 ›› 2021, Vol. 39 ›› Issue (3): 19-24.

• 理论研究 • 上一篇    下一篇

融合微观行为特性的用户画像增强研究

  

  • 出版日期:2021-03-01 发布日期:2021-03-15

#br#

  • Online:2021-03-01 Published:2021-03-15

摘要:

【目的/意义】目前在线阅读平台上显性数据不足,导致了用户画像较为单一,使得传统推荐算法难以达到
理想的推荐效果。事实上,读者序列化的行为数据多而丰富,不应被忽视。因此,如何更加精细地构建动态用户画
像,以提升推荐结果的精准度、可接受度是当前个性化推荐研究中的热点问题。【方法/过程】本文以在线阅读平台
为例,提出了一种融合微观行为的可解释性数字图书推荐算法IBS,通过捕捉读者的在线微观行为序列信息进行实
时的用户画像动态更新,通过引入认知偏差和个体差异校正项进行评分矩阵校正,最后根据不同的数据来源赋予
不同的推荐理由,在增加了用户画像精准度的同时辅以透明化的可解释性推荐算法,以此来优化推荐结果。【结果/
结论】仿真结果表明,IBS算法能提高在线阅读平台上数字图书推荐的召回率,有效提升数字图书推荐的精准度和
可接受度。【创新/局限】本文创新之处在于利用读者的微观行为序列动态增强了用户画像,并引入偏差项和可解释
性理由来优化推荐结果。在后续研究中,可通过扩展数字资源类型以及融入读者对推荐结果的反馈信息来进一步
优化推荐算法。

Abstract:

【Purpose/significance】At present, the lack of explicit data on the online reading platform leads to the traditional recom⁃
mendation algorithm difficult to understand users' interests. In fact, readers' serialized behavior data is abundant and it should not be
ignored. Therefore, how to build dynamic user profiles more precisely to improve the accuracy and acceptability of recommendation re⁃
sults is a hot issue.【Method/process】Taking online reading platform as an example, this paper proposes an interpretability algorithm
of implicit behavior sequence (IBS). This article updates the user profile by capturing the reader's online implicit behavior sequence.
In addition, the scoring matrix is adjusted by introducing cognitive bias and individual difference correction terms. Finally, different
recommendation reasons are given according to different data sources.【Result/conclusion】Simulation results show that IBS algorithm
can improve the recall rate of the recommendation on online reading platform, and effectively improve the accuracy and acceptability
of digital book recommendation.【Innovation/limitation】The innovation of this paper is to use serialized behavior data of readers to dy⁃
namically enhance the user profile, and introduce deviation term and interpretability reason to optimize the recommendation results. In
future practice, the recommendation algorithm can be further optimized by extending the types of digital resources and incorporating
the feedback information of readers.