利用效用度量提取非冗余关联购买行为
125 浏览量
更新于2024-08-31
收藏 821KB PDF 举报
"通过效用度量提取非冗余关联购买行为"
这篇研究论文探讨了如何利用效用度量来提取非冗余关联的购买行为，旨在从海量的用户点击和购买数据中挖掘有价值的信息。在电子商务和大数据分析的背景下，理解用户的购买行为对于提升销售策略、个性化推荐以及市场预测至关重要。
文章首先指出，用户的在线购物行为数据，如搜索、浏览和购买记录，蕴含着丰富的消费者行为模式。这些模式可以反映用户的兴趣、需求和偏好，对于商家来说具有极大的商业价值。然而，这些数据通常包含大量的冗余信息，直接分析可能会导致模式的重复和不精确性。
论文提出了一个基于效用度量的方法，用于识别和提取那些具有显著相关性但又不冗余的购买行为模式。效用度量在这里起到了关键作用，它能评估每个购买行为模式对用户价值的贡献，从而帮助过滤掉那些低效用或重复的模式。这种方法考虑了购买行为的内在相关性，即某些商品可能因为共同的用途或消费者的特定需求而经常一起被购买。
为了实现这一目标，论文中可能涉及了以下技术：
1. 数据预处理：对原始的用户购买数据进行清洗和转换，以便于分析。这可能包括数据去重、异常值处理、缺失值填充等步骤。
2. 购买模式挖掘：使用聚类、关联规则学习或者序列模式挖掘等方法，找出用户购买行为中的频繁模式。
3. 效用计算：定义并应用效用函数，以量化每个购买模式对用户的价值。这可能涉及到用户评分、商品价格、购买频率等多种因素。
4. 非冗余性检查：通过投影或其他数学手段，检查和消除模式之间的冗余，确保提取出的购买行为模式是独立且有价值的。
5. 结果评估与验证：采用适当的评估指标（如准确率、召回率、F1分数等）来验证所提方法的有效性和效率。
这篇论文为理解和挖掘用户购买行为提供了一个新颖的视角，通过效用度量，不仅可以更准确地捕捉到用户的实际需求，还能有效地减少数据分析的复杂性，对于电商平台和市场研究具有重要的实践意义。
32 W. Gan et al. / Knowledge-Based Systems 143 (2018) 30–41
Recently, a novel list-based algorithm named HUI-Miner [33] is
proposed to efficiently mine HUIs without candidate genera-
tion. It introduces a new data structure named utility-list which
stores the actual utility and remaining utility of an itemset
within each transaction in the processed database. Thus, the one-
phase HUI-Miner algorithm significantly outperforms the previ-
ous Two-Phase [34] and pattern-growth tree-based algorithms
(i.e., IHUP [6] , UP-Growth [40] , and UP-Growth + [41] ). After
that, two enhanced versions of HUI-Miner, FHM [18] and HUP-
Miner [22] , are further developed. Experiments have shown that
both FHM [18] and HUP-Miner [22] significantly outperform
the existing state-of-the-art HUI-Miner algorithm for most static
databases. At the same time, another approach named d
2
HUP is
also proposed to mine HUIs without candidate generation-and-
test [35] . Ryang and Yun introduced an indexed list-based IMHUP
algorithm for mining HUIs [38] . Recently, another faster one-phase
approach called EFIM [45] is designed for mining HUIs and it out-
performs all the previous HUIM algorithms.
Developing efficient high-utility pattern mining algorithms is an
active research. Many algorithms are also extensively developed
for various problems of HUIM, such as mining HUIs from dynamic
environment databases [27] , mining on-shelf high-utility itemsets
that the products having different on-shelf behavior [23] , top- k
high-utility itemset mining without setting the minimum utility
threshold [42] , mining the up-to-date HUIs which can show the
recent trend [26] , HUIM from big data [32] . Different from the tra-
ditional definition of HUIM, Hong et al. introduced the concept
of high average utility itemset [16] , and some approaches have
been developed in recent years, such as MAU-Growth [21] . Re-
cently, some interesting issues of mining high-utility itemsets un-
der various constrains has been extensively studied, such as HUIM
with multiple minimum utility thresholds [30] , HUIM with consid-
eration of various discount strategies [29] , utility-based association
rule mining [36] , mining HUIs over uncertain databases [25] , min-
ing HUIs with both positive and negative unit profits [31] , and so
on.
2.2. Affinity/correlation pattern mining
Traditional algorithm of frequent pattern mining (FPM) [14] and
association rule mining (ARM) [2,3] uses the support (frequency)
of patterns as a constraint to prune the search space. However,
this support-based approach has major drawbacks [39,43] . To ad-
dress this issue, the problem of strong affinity pattern mining
has been extensively studied in recent year [7,20,37,39,43] . To find
strong affinity patterns containing low-support items, the con-
cept of hyperclique patterns [20] was proposed. It defines a new
measure named hyperclique (h)-confidence, which is equivalent to
the all-confidence, to discover hyperclique patterns. With the h-
confidence measure, a cross-support property is used to effectively
eliminate spurious patterns. Besides, three interesting measures
called any-confidence, all-confidence, and bond [37] were proposed
to find frequency-based or support-based correlated patterns.
The degree of the expectation-based correlation is highly in-
fluenced by the number of null transactions [43] , i.e., transac-
tions which do not contain items whose correlation has been mea-
sured. Hence, such measures are not suitable for the study of
correlations in large datasets, where the number of null transac-
tions could be large and unstable. For the problem of contrast-
ing positive and negative correlations, it is crucial to adopt a re-
liable correlation measure that is unconcerned with the number
of null-transactions present in the database. These measures are
called null (transaction) - invariant [43] . The main property of a
null-invariant measure is its independence of the total number of
transactions in a database. According to the study in [43] , all five
known null-invariant measures, including: all-confidence [37] , Co-
herence [37] , Cosine [43] , Kulczynsky [9] , can be viewed as a gen-
eralized mean of conditional probabilities. Kulczynsky [9] has the
null (transaction)-invariant property, which implies that the corre-
lation measure is independent of the dataset size [43] .
2.3. Comparative analysis with related work
Up to now, the problem of HUIM has been extensively stud-
ied. Summary of the developed HUIM algorithms are shown in
Table 1 . Although the above traditional models of HUIM can reflect
the utility of the itemsets which beyond a minimum utility value,
the inherent correlation of items inside the patterns has not been
considered. In real-world situations, the discovered HUIs may be
meaningless, invalid even misleading (have happened by chance)
if they are weakly correlated. Identifying correlation and utility
information in databases can provide valuable information. Thus,
it is a critical issue and a challenge to discover non-redundant
and more correlated HUIs from transaction databases. In the past
studies, the HUIPM algorithm [7] and the faster FDHUP algorithm
[28] were respectively proposed to simultaneously consider both
the frequency affinity and utility of patterns. FDHUP introduces
two compact structures named EI-table and FU-tree, and utilizes
three pruning strategies to reduce the search space, thus performs
better than the HUIPM algorithm [7] . Both of them only consider,
however, the co-occur frequency the transactions as the correla-
tion factor, which cannot discover the real inherent correlation of
among items inside the desired patterns.
Thus, it is an important and challenging task to design an ef-
ficient algorithm for extracting non-redundant correlated purchase
behaviors by utility measure. In this paper, the proposed method
aims at discovering non-redundant correlated high-utility itemsets
from transactional quantitative databases and has the following
main differences compared to the state-of-the-art algorithms for
mining high-utility itemsets.
• The mining goal w.r.t. derived patterns is different. All the
above algorithms except HUIPM are designed to discover high-
utility itemsets, while the CoHUIM algorithm aims at mining
the high positive correlated but interesting patterns which hav-
ing high utility values.
• The correlation measure used in this paper is quite different
from the support affinity measure used in HUIPM. The sup-
port affinity was designed only with the co-occurrence consid-
eration. It used the utility relationship instead of the presence
information to measure the correlation of the items inside an
itemset. Utility factor of objects is, however, the user-specified
important value, the subjective criteria may be wrong to mea-
sure the objective correlation relationship. Moreover, they are
out of our purposes since the designed approach is used to
avoid mining the misleading HUIs from transactional databases.
• Furthermore, we develop some techniques to improve mining
performance. Not only the projection technology is adopted, but
also a global sorted downward closure ( SDC ) property is used in
the proposed approach.
3. Preliminaries and problem statement
Let I = { i
1
, i
2
, . . . , i
m
} be a finite set of m distinct items in a
temporal transactional database D = { T
1
, T
2
, . . . , T
n
}, where each
transaction T
q
= { q (i
1
, T
q
) , q (i
2
, T
q
) , . . . , q (i
j
, T
q
) } is a subset of I ,
and has an unique identifier ( TID ). Notice that the q ( i
j
, T
q
) in each
T
q
is the different purchase quantity of each item. An unique profit
pr ( i
j
) is assigned to each item i
j
∈ I , which represents its impor-
tance (e.g., profit, interest, risk), and they are stored in a profit-
table ptable = { pr ( i
1
), pr ( i
2
), . . . , pr ( i
m
)}. An itemset X ∈ I with k
distinct items { i
1
, i
2
, . . . , i
k
} is of length k and is referred to as a
剩余11页未读，继续阅读
2024-10-12 上传
2023-08-11 上传
2021-03-09 上传
2023-08-11 上传
2021-03-09 上传
2021-06-08 上传
2021-03-05 上传
2021-05-16 上传
2021-03-09 上传
weixin_38708461
- 粉丝: 5
- 资源: 993
我的内容管理
展开
最新资源
- Angular实现MarcHayek简历展示应用教程
- Crossbow Spot最新更新 - 获取Chrome扩展新闻
- 量子管道网络优化与Python实现
- Debian系统中APT缓存维护工具的使用方法与实践
- Python模块AccessControl的Windows64位安装文件介绍
- 掌握最新*** Fisher资讯，使用Google Chrome扩展
- Ember应用程序开发流程与环境配置指南
- EZPCOpenSDK_v5.1.2_build***版本更新详情
- Postcode-Finder：利用JavaScript和Google Geocode API实现
- AWS商业交易监控器：航线行为分析与营销策略制定
- AccessControl-4.0b6压缩包详细使用教程
- Python编程实践与技巧汇总
- 使用Sikuli和Python打造颜色求解器项目
- .Net基础视频教程：掌握GDI绘图技术
- 深入理解数据结构与JavaScript实践项目
- 双子座在线裁判系统：提高编程竞赛效率
