基于协同标注数据的Top-K产品设计
Top-K Product Design Based on Collaborative Tagging Data
摘要:
协作内容站点的广泛使用和普及(例如,IMDB、亚马逊、Yelp等)已经为用户创建了丰富的资源来进行咨询以便对诸如电影、电子商务产品、餐馆等的各种产品做出购买决定。具有期望标签的产品(例如,现代的,可靠的,等等)更有可能被潜在客户选中。这为产品设计师创造了一个机会,可以设计出更好的产品,在发布时可能会吸引人们想要的标签。在本文中,我们研究了如何挖掘协同标签数据,以确定新产品的属性值,并返回的top-k产品,可能会吸引最大数量的理想标签时发布。给定现有产品的训练集及其功能和用户提交的标签,我们首先为每个标签构建一个朴素贝叶斯分类器。我们表明,即使简单的朴素贝叶斯分类器用于标签预测的问题是NP完全的。我们提出了一套解决这个问题的算法:(a)一个确切的两层算法(基于前k查询技术),它的性能比天真的蛮力算法和工程以及中等的问题实例,和(b)一组近似算法较大的问题实例:一个新的多项式时间近似算法与可证明的错误界和一个实用的爬山启发式。我们对从网络上抓取的合成数据和真实数据进行了详细的实验,以评估我们提出的算法的效率和质量,并展示了产品设计师如何通过利用协作标记信息受益。
英文摘要:
The widespread use and popularity of collaborative content sites (e.g., IMDB, Amazon, Yelp, etc.) has created rich resources for users to consult in order to make purchasing decisions on various products such as movies, e-commerce products, restaurants, etc. Products with desirable tags (e.g., modern, reliable, etc.) have higher chances of being selected by prospective customers. This creates an opportunity for product designers to design better products that are likely to attract desirable tags when published. In this paper, we investigate how to mine collaborative tagging data to decide the attribute values of new products and to return the top-k products that are likely to attract the maximum number of desirable tags when published. Given a training set of existing products with their features and user-submitted tags, we first build a Naive Bayes Classifier for each tag. We show that the problem of is NP-complete even if simple Naive Bayes Classifiers are used for tag prediction. We present a suite of algorithms for solving this problem: (a) an exact two tier algorithm(based on top-k querying techniques), which performs much better than the naive brute-force algorithm and works well for moderate problem instances, and (b) a set of approximation algorithms for larger problem instances: a novel polynomial-time approximation algorithm with provable error bound and a practical hill-climbing heuristic. We conduct detailed experiments on synthetic and real data crawled from the web to evaluate the efficiency and quality of our proposed algorithms, as well as show how product designers can benefit by leveraging collaborative tagging information.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。