社会系统中多样性和新颖性推荐的改进方法

需积分: 5 20 浏览量更新于2024-07-10 收藏 805KB PDF 举报

"基于民俗分类的社会系统中总体推荐多样性和新颖性的改进" 在当前的信息时代，随着物联网技术的进步，各种基于用户生成标签（folksonomy）的社交媒体应用日益普及。folksonomy是一种用户自定义标签的方式来组织和分类网络内容的体系，它在提供信息共享和发现的便利性的同时，也带来了信息过载的问题。由于folksonomy系统规模庞大且复杂，用户在寻找感兴趣的内容时可能感到困惑，因此推荐系统的需求变得至关重要。推荐系统的准确性一直是研究的重点，但除了准确性之外，推荐的多样性和新颖性也是提升用户体验的重要因素。多样性指的是推荐结果的丰富性，能够满足用户的多样化需求；新颖性则是指推荐未被用户接触过的、新鲜的资源，增加用户的探索体验。然而，在实际操作中，提高多样性往往会导致准确性的下降，这是一个亟待解决的矛盾。本文的研究论文提出了一个名为Diversity-aware Personalized PageRank (DaPPR)的方法，旨在从整体视角解决这个问题。DaPPR方法结合了个性化PageRank的思路，这是一种基于随机游走的算法，用于评估网络中节点的重要性。在推荐系统中，PageRank可以理解为衡量资源对用户兴趣的影响力。通过引入多样性意识，DaPPR能够在保持推荐准确性的同时，提升推荐列表的多样性和新颖性。 DaPPR的创新之处在于它考虑了用户兴趣的多维度，以及资源之间的关联性。在计算推荐分数时，不仅考虑用户过去的偏好，还考虑了推荐资源的多样性和新颖性。通过调整算法参数，可以在多样性和准确性之间找到一个平衡点，从而提供既准确又具有吸引力的推荐列表。此外，论文可能会详细探讨DaPPR的实现过程，包括如何构建用户-资源网络，如何定义多样性和新颖性的度量，以及如何优化算法以提高推荐性能。可能会通过实验对比分析，展示DaPPR与其他推荐策略的性能差异，例如基于协同过滤或基于内容的推荐方法，并用真实数据集进行验证。这篇研究论文致力于解决基于folksonomy的社会系统中的推荐问题，通过DaPPR算法在保证推荐质量的同时，增强了推荐的多样性和新颖性，为推荐系统的设计提供了新的思路和实用工具。这一贡献对于改善用户在大规模信息环境下的信息获取体验具有重要意义。

(k jLj) places of L, the precision at the ranking position k

is deﬁned as following (P@k is short for Precision@k),

Precision@k ¼ N

=k ð1Þ

2.2 Diversity and novelty of recommendations

As mentioned above, the diversity of recommendations

can be measured in two ways: individual and aggregate.

The metrics of individual diversity usually characterize an

accumulated dissimilarity between all pairs of items

within a recommendation list, e.g., intra-list similarity [4,

34] and item novelty [15]. In contrast to individual

diversity, there have been few studies that explore

aggregate diversity in recommender systems. Herlocker

et al. [13] experimented with the coverage deﬁned as the

percentage of items for which the recommender system is

able to make recommendations. Adomavicius and Kwon

[1] deﬁned diversity-in-top-N as an aggregate diversity

measure, to count the total number of distinct items rec-

ommended to all users. Since we intend to evaluate rec-

ommender systems based on the top-k recommended lists

of items that the system provides to its users, in this

paper, we take hamming distance as our aggregate

diversity measure. Given two candidate lists L

and L

for

user u and v, the difference in the top-k

(k  m axðjL

j; jL

jÞ) places can be measured using

hamming distance as following [33],

HD@k ¼ 1  overlap@k=k ð2Þ

where overlap@k is the number of shared resources in the

top-k places of the two lists. Averaging over all pairs of

users, we can obtain the aggregate diversity of the system.

Clearly, HD@k can characterize the uniqueness of different

user’s recommendation lists, higher diversity means higher

personalization of users’ recommendation lists, HD@k ¼ 1

points to the fact that every user receives his/her own

unique top-k items [33].

Novelty and diversity are different though related

notions. The novelty of a piece of information generally

refers to how different it is with respect to ‘‘what has been

previously seen’’ by a speciﬁc user, or a community as a

whole [29]. To evaluate aggregate novelty of recommen-

dations, we use popularity-based item novelty proposed by

Vargas & Castells [29],

Nov@k ¼logpðkjuÞð3Þ

where k indicates the k-th resource in the recommendation

list L of u. Here, the posterior pðkjuÞ cannot be estimated

directly, since it assumes no observation of u accessing k

before recommendations are made. Instead, Vargas &

Castells suggested estimating pðkjuÞ based on other items

the user has accessed. Let us assume the observed data

consist of a set S of user-based assignment of tags to

resources, reﬂecting item access by users; then, pðkjuÞ can

be inferred as followings,

pðkj uÞ/

i2I

pðkjiÞpðijuÞ

pðkjiÞ/

\ U

pðijuÞ/

jðu; t; iÞ2Sj

jðu; t; i

Þ2Sj

ð4Þ

where I

denotes the set of items collected by u, and U

denotes the set of users who have accessed k. We average

Nov@k across all users, to measure the capability of a

recommender system provides novel items to its users.

Since the novelty measure is not normalized and universal

for different systems, it is considered as a complementary

measure to the diversity. Note that there still exists an

apparent dilemma between recommendation accuracy and

diversity (novelty). Thus, in this paper, we aim to ﬁnd new

techniques to improve aggregate diversity and novelty

while maintaining accuracy of recommendations, espe-

cially for folksonomy-based systems.

3 Methodology

3.1 Folksonomy schema

A folksonomy describes the users, the resources, the tags

and the user-based assignment of tags to resources. It is a

tuple ðU; T; R; SÞ where U; T and R are ﬁnite sets whose

elements are called users, tags and resources, respectively,

and S is a ternary relation between them, i.e.,

S ¼ U  T  R, whose elements are called tag assignments

[17]. Users are typically described by their user ID, and

tags may be arbitrary strings. What is perceived as a

resource depends on the type of social application system.

For instance, in Delicious, the resources are URLs, and in

Lastfm, the resources are artists and tracks. There are also

other homogenous links among users or resources, such as

friendship among users, hyperlinks among Web pages and

citation links among papers. In addition, resources may

have various attributes information. A typical folksonomy

schema can be described as Fig. 1, which can be also

formulated as an association link network [21].

3.2 Personalized PageRank

PageRank [5] is an ordering nodes technique by a random

surfer model in a directed graph G ¼ðV; EÞ, where

VðjVj¼nÞ is the set of nodes and E is the set of edges. The

random surfer performs a Markovian walk on G. The surfer

jumps from node to node following a link with uniform

probability d (called as damping factor) or gets bored and

Pers Ubiquit Comput (2014) 18:1855–1869 1857

123

剩余14页未读，继续阅读

weixin_38501826

粉丝: 9
资源: 893

社会系统中多样性和新颖性推荐的改进方法

基于SSM的民俗文化保护系统设计与实现

标签组合用户建模：民俗分类中的个性化搜索新方法

基于SpringBoot+Vue的陕西民俗网站系统源码及毕业论文

（正文）综合材料在传统民俗装饰造型中的运用与表现.zip

论大型民俗歌舞剧《蜀红》音乐的独特性.docx

brizzle-ipsum:bri lorem ipsum受民俗方言的启发

浅谈音乐教学中多样化的应用与创新能力的培养

旅游资源的分类与特性PPT学习教案.pptx

浅析农业嘉年华主题场馆中的展陈设计.pdf

中华传统文化在校园文化建设中的实践应用.pdf

最新资源