分层多线索建模：预测异构游客信息下的POI人气

90 浏览量更新于2024-08-26 收藏 2.53MB PDF 举报

"具有异构游客信息的POI人气预测的分层多线索建模" 这篇研究论文探讨了在地点信息预测领域的一个重要问题——如何利用异构游客信息进行点兴趣（Point of Interest, POI）人气预测。POI人气预测在基于位置的服务中扮演着关键角色，比如用于POI推荐系统。当现有的方法由于POI信息稀缺而往往无法取得满意的效果时，这个问题显得尤为突出。这些方法通常将推荐范围限制在热门景点，而忽视了那些可能具有宝贵价值的不热门景点。作者Yang Yang、Yaqian Duan、Xinze Wang、Zi Huang、Ning Xie以及Heng Tao Shen（均为IEEE成员）提出了一种新颖的层次多线索建模方法，旨在解决这一挑战。他们强调了利用不同类型的异构游客信息（如用户行为数据、社交网络互动、地理位置数据等）的潜力，这些信息可以提供更全面、更深入的洞察，帮助提高预测准确性。文章的核心贡献在于开发了一个层次化的框架，该框架能够有效地整合和分析多种线索，包括但不限于： 1. **多层次线索集成**：论文中的“层次化”指的是对不同层面的游客信息进行建模，可能包括时间序列分析（如一天中的不同时间段、一周内的不同天数）、空间分布分析（POI之间的地理关系）和社会网络分析（用户的社交连接）。 2. **异构信息融合**：异构信息指的是不同类型的数据，如用户访问记录、评论文本、照片、评分等。通过有效的融合策略，这些信息可以相互补充，增强预测模型的鲁棒性。 3. **机器学习算法应用**：论文可能会介绍使用何种机器学习或深度学习技术来处理和学习这些多线索数据，如时间序列模型（如LSTM）、图神经网络（GNN）或协同过滤方法。 4. **性能评估与优化**：研究会通过实验验证所提方法的性能，并与其他现有方法进行对比，以证明其优越性。此外，可能会讨论参数调整和模型优化策略。 5. **案例研究**：论文可能会以实际的地理信息数据集为例，展示如何运用此方法进行POI人气预测，并分析预测结果的准确性和实用性。通过这种创新的建模方式，研究期望能更准确地预测POI的人气，从而改善推荐系统的质量和用户体验，尤其是在旅游和本地服务领域。这种技术的进步对于提升用户满意度，促进相关商业活动的发展具有重要意义。

1041-4347 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TKDE.2018.2842190, IEEE

Transactions on Knowledge and Data Engineering

this spot to others. In order to solve the “sparsity problem”,

topic model method (TM) and its improvement are showed

in many works [15], [16], TM is a model that make POI rec-

ommendation by making use of visitors’ preferences. Even

though a user has few POI information, through discovering

the “topics”, we can also recommend proper POI to the user.

We mainly focus on the work by Jiang et al. [17], this

work presents a personalized travel sequence recommenda-

tion by a Topical Package Model. Dataset is built by mining

user travel interest in community-contributed photos and

travelogues, Jiang et al. speciﬁcally represent the structure

of data they crawled from IgoUgo. In the method of ranking

famous travel routes, they consider the popularity of attrac-

tions and visitors’ preferences at the same time. Referring to

these thoughts, we develop our methods based on hierarchi-

cal structure of data. Compared to single data source used

in this work, we collect data from several travel websites.

Because of the different focus of research in [17] and our

work, while they mention the popularity of attractions, the

problem setting and techniques are markedly different.

In the area of popularity prediction, research works

aiming at social inﬂuence and social behaviors are ﬂourish-

ing [18]. Cui et al. [19] focus on measuring item-level social

inﬂuence. To deal with the social information effectively,

they propose a HF-NTF approach to study user-post speciﬁc

social inﬂuence prediction. Jiang et al. [20] consider one-

class collaborative ﬁltering method in retweeting prediction

and achieve the improvement of the OCCF performance

by a novel weighting method that measures retweeting

probability score between user and message. There also exist

work that focus on strategies improving prediction. Song et

al. [21] propose a multiple social network learning model,

which is applied to predicting volunteerism tendency. They

solve the challenge of block-wise missing data as well by

utilizing multiple sources jointly. Wu et al. [22] focus on

time information on social media popularity. They utilize a

Multi-scale Temporal Decomposition method which aims at

factorizing contexts of both user-item and time-sensitive for

predicting photo popularity in social media. Meier et al. [23]

make several design suggestions based on the investigation

of re-ﬁnding strategies and demonstrate the possibility of

predicting if a tweet is likely to be re-found.

However, there are not many researches in popularity

prediction of scenic spots. In [7], Cho et al. conduct a

research on POI prediction in Location-Based Social Net-

works, where the prediction of POI to be visited next by

a user is evaluated based on a novel model of human

movement and mobility. While the method has an inﬂuence

on researches in POI prediction, the problem of concern in

this work is different from our work which concentrates on

popularity change of an attraction. Additionally, the data

used in this work is sparse, which leads to unsatisfactory

datasets to the requirements of their method.

Hierarchical method is applied effectively in various

ﬁelds [24], especially in dealing with UGC content. In [25],

Zhu et al. present an automatically generated and updated

topic hierarchy to organize information from multiple UGC

sources. The hierarchy considers topic items as well as sub-

topic relations by which UGC content can be organized.

Ahmed et.al provide a hierarchical method named nCRF.

This hierarchy is of three distributions over locations, topics,

Grades of Unpopular Spots Grades of Popular Spots

0--1 0.11 0--1 0

1--2 1.12 1--2 0.13

2--3 25.08 2--3 7.4

3--4 66.47 3--4 75.4

4--5 7.22 4--5 17.07

0--1

1--2

2--3

25%

3--4

67%

4--5

Grades of Unpopular Spots

0--1 1--2 2--3 3--4 4--5

0--1

1--2

2--3

3--4

76%

4--5

17%

Grades of Popular Spots

0--1 1--2 2--3 3--4 4--5

Fig. 3: Grade Distribution.

and user characteristics, which performance well in location

estimation [26]. In [27], a feature learning model based on

hierarchy is proposed in event forecasting. The Nth-order

strong hierarchy has interaction with group Lasso across

multiple data sources. In the problem of prediction, Zhang

et.al [28] propose a hierarchical model combined with bi-

modal deep belief network, which considers both vision-

speciﬁc and text-speciﬁc DBN in the multinomial level.

Existing hierarchical models seldom refer to aspects of

POI, especially POI prediction. Besides, these hierarchies

overlook the combination of nature language models with

multi-feature models, which is especially signiﬁcant to en-

sure the completeness of real-world information.

As compared to existing studies above-mentioned, our

work focus on POI popularity prediction by integrating

multiple sources with new hierarchical POI modelling

method, which is unexplored previously. The information

of POIs can be completed relatively even though very few

UGC content is in regard to certain POI.

3 DATA COLLECTION

Recently, as social networks keep emerging, people incline

to comment on POIs and share their experiences in on-

line tourism websites, where they also search for tourist

destination referring to other visitors’ comments. Multiple

media types are explored widely in existing researches [29].

Taking Dianping as an example, it has over 250 million

users and over 20 billion monthly page views, which pro-

vide comprehensive descriptions such as textual comments,

images, grades, for each individual POI. Considering the

fact of data scarcity problem and the incompleteness of

information from a single website, it is necessary to establish

our POI dataset across multiple sources. The illustration of

multi-source description of POIs is shown in Figure 2. Take

Chengdu Folk Custom Park as an example, while informa-

tion is complete in Dianping, there is still missing data on

other websites. This phenomenon is common among the

majority of POIs, even worse, a large number of POIs have

few information in every single platform.

We build a multi-source POI dataset by integrating

content of several speciﬁc provinces in China from four

main-stream tourism platforms, i.e., Dianping

, Mafengwo

2. https://www.dianping.com/

3. http://www.mafengwo.cn/

剩余11页未读，继续阅读

weixin_38747815

粉丝: 54
资源: 889

分层多线索建模：预测异构游客信息下的POI人气

上海菜鸟驿站位置信息POI数据全览

高效地图POI信息采集工具与多平台支持

社交网络POI信息展示技术在网络游戏中应用

POI解析EXCEl分层

利用城市POI数据提取分层地标.pdf

poi

基于多源异构POI融合方法的省级地名地址数据库建设及应用.docx

利用POI建模分析城市国土开发密度.docx

桂林市POI信息

POI导出 POI导出 POI导出

最新资源