没有合适的资源?快使用搜索试试~ 我知道了~
细粒度IP定位参文27:Identifying user geolocation
0 下载量 168 浏览量
2024-03-09
20:11:01
上传
评论
收藏 2.32MB PDF 举报
温馨提示
试读
13页
细粒度IP定位参文27:Identifying user geolocation
资源推荐
资源详情
资源评论
Information Fusion 81 (2022) 1–13
Available online 12 November 2021
1566-2535/© 2021 Elsevier B.V. All rights reserved.
Contents lists available at ScienceDirect
Information Fusion
journal homepage: www.elsevier.com/locate/inffus
Identifying user geolocation with Hierarchical Graph Neural Networks and
explainable fusion
Fan Zhou
a
, Tianliang Wang
a
, Ting Zhong
a,
∗
, Goce Trajcevski
b
a
University of Electronic Science and Technology of China, China
b
Iowa State University, USA
A R T I C L E I N F O
Keywords:
User geolocation
Social information fusion
Graph Neural Networks
Interpretable fusion
Influence function
A B S T R A C T
Determining user geolocation from social media data is essential in various location-based applications —
from improved transportation/supply management, through providing personalized services and targeted
marketing, to better overall user experiences. Previous methods rely on the similarity of user posting content
and neighboring nodes for user geolocation, which suffer the problems of: (1) position-agnostic of network
representation learning, which impedes the performance of their prediction accuracy; and (2) noisy and
unstable user relation fusion due to the flat graph embedding methods employed. This work presents
Hierarchical Graph Neural Networks (HGNN) – a novel methodology for location-aware collaborative user-
aspect data fusion and location prediction. It incorporates geographical location information of users and
clustering effect of regions and can capture topological relations while preserving their relative positions. By
encoding the structure and features of regions with hierarchical graph learning, HGNN can primarily alleviate
the problem of noisy and unstable signal fusion. We further design a relation mechanism to bridge connections
between individual users and clusters, which not only leverages the information of isolated nodes that are
useless in previous methods but also captures the relations between unlabeled nodes and labeled subgraphs.
Furthermore, we introduce a robust statistics method to interpret the behavior of our model by identifying the
importance of data samples when predicting the locations of the users. It provides meaningful explanations on
the model behaviors and outputs, overcoming the drawbacks of previous approaches that treat user geolocation
as ‘‘black-box’’ modeling and lacking interpretability. Comprehensive evaluations on real-world Twitter datasets
verify the proposed model’s superior performance and its ability to interpret the user geolocation results.
1. Introduction
The plethora of Online Social Networks (OSN) have enabled novel
interactions in daily activities – e.g., sharing notifications about events
related to product descriptions and traffic jams; sharing personal expe-
riences on Instagram and Facebook; reading news and popular topics
on Twitter; building academic connections on ResearchGate, etc. These
have not only changed our way of communication, reading, and social
activities but also enabled a generation of an unprecedented volume
of heterogeneous data, which, in turn, fosters business innovations
and emerging industrial opportunities [1]. Among various applications,
identifying the geographic locations of users receives lasting interest
from both academia and industry and has become an essential Internet
service for many industrial services, such as location-based targeted
advertising, emergency location identification, political elections, sub-
stance use surveillance, local event/place recommendation and natural
disaster response [2–4].
∗
Corresponding author.
E-mail addresses: [email protected] (F. Zhou), [email protected] (T. Wang), [email protected] (T. Zhong), [email protected]
(G. Trajcevski).
Fine-grained localization, such as various sensor-based tracking of
assets and processes, have already been exploited in multiple industrial
applications. However, in more extensive geographical settings, there
is the issue of inaccuracy due to, e.g., cellular access restrictions,
high measurement overhead, and unreliable client response times [5].
Complementary to this, the increased popularity of social media ser-
vices (e.g., Twitter, Facebook, and Instagram) provide rich and timely
metadata, e.g., published message contents, mention tags, and fol-
low/followee relations. This information could be efficiently leveraged
to promptly geolocate OSN users — which has recently spurred re-
search interest in the, so calls, User Geolocation (UG) problem in
OSN [6–9]. For example, the CDC (centers for disease control and
prevention) has been utilizing social media to help the epidemiological
investigation in responding to the virus that causes COVID-19 [10].
https://doi.org/10.1016/j.inffus.2021.11.004
Received 9 December 2020; Received in revised form 6 July 2021; Accepted 6 November 2021
Information Fusion 81 (2022) 1–13
2
F. Zhou et al.
Online user geolocation is a passive crowd-sensing problem that
requires hybrid information fusion and insights from many user activ-
ities and sensing data to distill the knowledge and refine the predicted
results. Early efforts [6,11] mainly focused on mining indicative infor-
mation from users’ posting content relying on indicative words that can
link users to their home locations, based on various natural language
processing techniques (e.g., topic models and statistic models). For
example, Term Frequency–Inverse Document Frequency (TF-IDF [12])
is a commonly used method to measure the distribution of location
words [6]. More recent efforts fuse users interactions for collaborative
sensing and boosting the geolocation accuracy – e.g., node2vec [13]
is used to learn representation of users [7], combined with text repre-
sentation via doc2vec [14] to predict user locations in an end-to-end
manner. Recurrent Neural Networks (RNNs) with attention mechanism
to model user tweet content are also used in [8], further combining
the metadata such as timezone and self-declared profiles to predict
user locations. A more recent work [9] employs GCNs [15] for learning
network structures with graph convolution and pooling operations.
Broadly speaking, the existing state-of-the-art methods employ deep
learning techniques for learning user interaction and content represen-
tation — without fully exploiting the specific constraints in the user
geolocation task. When learning user interactions, graph representation
methods (e.g., GCN [15], GAT [16], node2vec [13], GraphSAGE [17])
are commonly used — however, the approaches are general, unweighted
and location-agnostic graph learning methods, without considering the
geographical position/location of nodes (users). Since the graph em-
bedding methods are not specifically tailored for user geolocation
task, existing approaches ignore the strong geolocation dependencies
among nodes and thus cannot capture the relative distance between
any pair of nodes. In addition, existing graph-based UG methods are
inherently flat graph learning models, which cannot capture the region-
level features and thus are very sensitive to local network structure.
For example, the homphily assumption, i.e., online interactions imply
a higher probability of geographical proximity, is not held in many
cases [2,18].
Our main motivation is based on the observation that the method-
ologies in the existing literature do not exploit the benefits of joint
consideration of identifying the topological structure of users along
with the influence of crowds from different regions. While the former is
usually noisy and unstable, the latter may provide a more robust signal
for geolocating. In addition, existing models, especially those based on
deep neural networks, often lack transparency and cannot interpret
model behavior and localization results. Thus, their applicability in
safety-critical areas is restricted. For example, when locating area with
specific emergencies (for example, the spread of COVID-19), it would
be more significant to explain why and how such a prediction was made
instead of just presenting the predicted results [19–21].
To address the aforementioned limitations of previous works, we
propose a novel multi-view user geolocation framework, called Hi-
erarchical Graph Neural Networks (HGNN), to fuse user-generated
content and network information for collaborative user geolocation.
It enhances user geolocation performance from the following aspects.
First, it incorporates the relative distances of each node to other nodes
(clusters) in the network, which enables the model to discriminate the
nodes having similar topological structures but residing in different
regions. Second, the hierarchical feature fusion method that we propose
provides both coarse- and fine-grained graph representation by learning
and distinguishing the crowd effects from different geographic regions.
Third, our model naturally exploits unlabeled and isolated nodes for
context information aggregation, which are absent in previous UG
models. Fourth, the interpretability of information fusion allows us
to understand the trained geolocation model’s behavior and how it
is affected by the information aggregated from the training samples
(i.e., all in-network users and their associated features). The main
contributions of this work in terms of the novelty of the proposed
approach are four-fold. Specifically, we present:
• A new location-aware node relation learning model that takes the
geographical location and relative distance into account when per-
forming non-linear transformation and feature aggregation, which not
only preserves network topology but also encodes node position with
respect to the other nodes and/or clusters.
• A new hierarchical GNN framework that learns both region- and
node-level features for robust feature aggregation and propagation,
which can be combined with any graph learning approaches in an
end-to-end manner. Compared to flat node-level embedding in exist-
ing UG approaches, we are able to alleviate the influence of noisy
interactions and the impact of outlier nodes.
• A new general framework to explain the behavior of user geolocation
models and the prediction results. We take the initiatives to use
influence function [22] to quantify the impact of in-network users
and corresponding features on the predicted outcomes.
• Extensive evaluations on three benchmark Twitter datasets. The
results demonstrate that our method significantly outperforms the
state-of-the-art baselines while providing explanations on both model
behavior and detection results.
In the rest of this paper, Section 2 reviews the related work,
followed by Section 3 that formalizes the problem and presents the
necessary backgrounds. In Section 4, we give the details of the method-
ology, as well as the approach for explaining the user-aspect data
fusion and location prediction. Experimental evaluations quantifying
the benefits of our approach are performed in Section 5. We conclude
this work and outline directions for future work in Section 6.
2. Related work
In the body of previous works on geolocating online social networks,
the models can be broadly categorized into three groups according to
the type of data used to make the prediction. We now review relevant
works and position our paper in the context of the existing literature.
2.1. Content-based approaches
User-generated content (UGC) such as textual posts and photos
may be casually attached with real-time locations facilitated by the
increasing popularity of GPS-equipped devices. However, these geo-
tagged tweets are extremely sparse, e.g., no more than 1% of published
tweets are labeled with geographical locations [23]. A plethora of
works [6,11,24–26] have studied the possibility of leveraging UGC
for locating users. These methods address the geolocation problem by
inferring locations from the location-relevant words with various clas-
sification models. Therefore, identifying meaningful indicative words
is an important step towards accurate user geolocation, where TF-
IDF [12] is a widely adopted textual content representation method in
the literature [6,9,27–29]. For example, inverse location/city frequency
has been used to measure the location words in the content [6,27]. In
contrast, probabilistic models are usually used to characterize the users’
location distributions w.r.t. their published UGC, which, however, re-
quires extensive manually labeled location-related words to achieve
satisfactory results.
Inspired by recent advances in applying deep learning in natural
language processing, a few studies turn to model users’ textual con-
tents with various neural networks based models in order to learn
the tweet representation in an end-to-end manner [7,8,30,31]. Among
these methods, doc2vec [14] and recurrent neural networks (RNNs)
are simple yet effective choices for learning vector representation of
textural contents. For example, in [7], combining TF-IDF and doc2vec
representations of textual information is proposed to enhance the pre-
diction performance. GRU [32] with attention mechanism [33] was
used in [8] to model user tweet content and obtain a timeline rep-
resentations. Though doc2vec and RNN-based methods can learn the
Information Fusion 81 (2022) 1–13
3
F. Zhou et al.
language characteristics efficiently without manual location feature en-
gineering, a recent study [34] finds that TF-IDF is consistently superior
to doc2vec due to the location-indicative words captured in TF-IDF.
Our present work enables better location-awareness than the exist-
ing literature and, in particular, HGNN distinguishes the crowd effects
from different geographic regions.
2.2. Network-based methods
Online social relationships are also important indicators for user
geolocation under the homophily assumption [35–38], i.e., people
prefer to interact with others in nearby areas. Backstrom et al. [35]
examine the relationship between users’ geographical proximity and
online friendships on Facebook, and find that the likelihood of relations
between any user pair drops monotonically as a function of distance.
Rather than solely relying on friendships, more and more works uti-
lize various types of connections, such as the co-mention tags and
mentions between non-friends, to construct closer social interactions
beyond friendships [2,31]. In this way, similar interests among users
can be retrieved from such implicit networks to improve geolocation
accuracy [30,39,40]. Moreover, researchers also identify some noisy
interaction factors that may degrade the prediction performance. For
example, social influence of celebrities is a distracting factor that
may confuse the prediction and thus is removed from the built user
network [30,41].
Although the existing approaches have tackled the aspect of ex-
plicitly modeling location dependency between social connected users,
some challenges have not been properly addressed — namely, the
sparsity of geo-tagged users and the inaccurate label propagation. More
importantly, friends’ locations are usually contradicting each other,
which hinders the practical applicability of these works. In contrast, our
HGNN learns both region-level and node-level features and aggregates
them in a manner that provides better intepretability.
2.3. Multi-information fusion based models
Recent efforts have leveraged deep graph learning methods to
model user interaction networks by fusing user-generated contents
and various meta-data, such as user profiles, tweeting time, and user
timezone. For example, MENET [7] exploits node2vec [13] to learn
user representations, combined with text representation learned by
doc2vec, for predicting users’ locations. Another work [9] employs
GCNs [15] for learning network structures with the graph convolution
and pooling operations, which has achieved state-of-the-art geolocation
performance. A recent work [34] investigate several graph embedding
methods and found that NetMF [42] performs better than node2vec and
GraphSAGE [17] on user geolocation task, but does not show superior
performance than GCN-based models [9,34].
It is worth noting that some works make use of various meta-data
(e.g., self-declared location in profile and timezone information) for
improving the prediction performance. For example, user timezone,
as well as UTC offset and country noun, have been used for user
geolocation [7,8,28,31,43]. While such auxiliary information is a strong
indicator for regularizing the locations the model predicted, a majority
of users are not willing to open this privacy information, which is
sometimes camouflaged or posted casually. We further note that there
is another line of efforts [36,44–47] studying the Twitter message
geolocation problem which tries to identify the tweeting locations
rather than the Twitter user location discussed in this work.
Despite the promising results on improving geolocation perfor-
mance, existing state-of-the-art methods fail to identify the importance
of individual users that we addressed in this work. Arguably, while
various graph embedding techniques can be utilized for network rep-
resentation in user geolocation, understanding the influence of user
connections is more important for interpreting the behavior of the
geolocation models and therefore benefits downstream decision mak-
ings. In this spirit, we initiate the attempt to analyze theoretically
and experimentally how the properties of graph structures influence
the geolocation performance. This not only demystifies and interprets
the predictions made by the model but outlines the underlying con-
straints of existing approaches, which, in turn, should be taken into
consideration in modeling and predicting user geolocation.
2.4. Graph neural networks
Graph neural networks are effective methods models for analyzing
and learning from data on graphs, and have been successfully applied
to a variety of domains including image processing [48], social net-
works [49], transportation systems [50], etc. Existing GNN models vary
from each others on message passing mechanisms, while most of them
rely on flat information aggregations [15]. There are several hierar-
chical GNN frameworks that gradually coarsen the original graph with
pooling operation for graph classification [51,52] and image recogni-
tion [48,53]. The main difference with our work is how HGNN model
defines the graph hierarchy for clusters and exploits the geographic
information. Directly applying GraphPool [51] or HGP-SL [52] for UG
task is problematic since both of them fail to consider the relative
location of nodes w.r.t. other nodes/clusters and cannot cluster the
unlabeled nodes. Another related work is PGNN [54], recently proposed
to learn the relative position of nodes. However, it does not leverage
nodes’ geographic information that is critical for UG. More importantly,
all these methods are suitable for fully connected graph learning, while
our HGNN model is capable of incorporating unlabeled and isolated
nodes and thus is more suitable for UG task.
Despite the promising performance gains on many graph tasks,
most GNNs are still black-box models without human-understandable
model behaviors and explanations. Although GAT [16] can learn the
importance of edges and thus, to some extent, explain the node
aggregation behaviors via attention mechanism, it is limited to spe-
cific architectures and fails to provide single-instance explanations.
To adaptively adjust the influence of each node, a learnt exploitation
of information from neighborhoods of differing locality and selective
combining of different aggregations was proposed in [55]. Though
their method can automatically discover the importance of each node
in a GNN, it is not specifically designed for explaining model predic-
tions. GNNExplainer [56] was proposed to explain the predictions of
model-agnostic GNNs. It interprets the GNN models by maximizing the
mutual information between a subgraph (or a subset of node features)
and the predictions for the original graph. Another work [57] uses
image interpretation methods, such as sensitivity analysis, guided back-
propagation, and layer-wise relevance propagation (LRP), to explain
the node-level predictions. GraphLIME [58] is a local interpretable
method that captures the nonlinear dependency between features and
predictions. It then considers the perturbation near a node and uses
a linear explanation model to find features as explanations for GNNs.
X-GNN [59] proposes to find the graph patterns that maximize a
particular prediction through graph generation, which is formulated as
a reinforcement learning problem and trained with a policy gradient
method. GNN-LRP [60] is a theoretically founded XAI method for inter-
preting GNN predictions, which is derived from the higher-order Taylor
expansions based on LRP. A recent work [61] systematically reviews
existing explainable GNN methods, and proposes to enable information
fusion for multi-modal causability using interpretable GNNs.
What separates our work from the existing GNN-based approaches is
that we propose a learning model which incorporates the geolocations
and distances and we provide a greater extent of explainability.
剩余12页未读,继续阅读
资源评论
路由跳变
- 粉丝: 2755
- 资源: 19
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功