ATopicModelingPerspectiveforTrackingtheEvolutionofSocialEmotions资源-CSDN文库

50 浏览量 2021-02-06 20:51:47 上传评论收藏 2.65MB PDF 举报

### 跟踪社会情绪演化的主题模型视角 #### 概述本文献提出了一种新的方法来跟踪在线新闻中的社会情绪变化，并将其应用于不同类型的在线服务中。研究背景是许多现代在线新闻网站允许用户在阅读新闻后表达自己的情感反应（如愤怒、震惊等）。与传统的用户反馈形式（如评论或评分）相比，这些特定的情感标注更能准确地表达用户的个人情绪。 #### 主题建模视角下的社会情绪演化 ##### 研究问题本文的核心挑战在于如何有效地将时间因素纳入到情感分析中，以跟踪情感随时间的变化趋势。作者通过提出一种基于时间感知的主题建模方法解决了这一问题。 ##### 模型介绍为了达到这一目标，作者首先提出了两种模型：情感-主题随时间变化模型（eToT）和混合情感-主题随时间变化模型（meToT）。这两种模型能够直接揭示新闻、情感和时间之间的潜在关系。具体而言： - **情感-主题随时间变化模型（eToT）**：该模型假设新闻主题随时间呈现为贝塔分布，并且新闻情感呈现为多项式分布。这使得模型能够在一定程度上捕捉到主题和情感的动态变化。 - **混合情感-主题随时间变化模型（meToT）**：在eToT的基础上，该模型进一步考虑了不同情感之间可能存在的复杂关系，通过引入混合情感分布来提高模型的灵活性。然而，这两种模型都未能完全捕捉到主题本身的演变过程。为了解决这一局限性，作者进一步开发了另一种模型——基于情感的动态主题模型（eDTM），该模型利用状态空间模型来追踪主题的动态变化。 ##### 基于情感的动态主题模型（eDTM） eDTM不仅考虑了新闻内容随时间的变化，还关注了主题间的转换关系，这对于理解社会情绪的长期演化至关重要。该模型的关键特点是它能够有效地追踪新闻主题随时间的变化趋势，并在此基础上预测未来可能出现的情感倾向。 #### 应用场景本文还探讨了所提出的模型可以支持的各种潜在应用，包括但不限于： - **情感预测**：通过分析过去的数据来预测未来的用户情感变化趋势。 - **基于情感的新闻推荐**：根据用户的偏好情感推荐相关的新闻内容，从而提高用户体验。 - **情感异常检测**：监测不寻常的情绪波动，帮助识别突发的社会事件或公众情绪变化。 #### 实验验证为了验证所提出的模型的有效性，作者进行了一系列实验。实验数据来自真实世界的数据集，结果表明这些模型在追踪社会情绪演化方面具有很高的准确性，并且能够成功应用于上述提到的各种应用场景中。 #### 结论本文提出了一种新颖的方法来跟踪社会情绪随时间的变化，并通过三种不同的模型展示了如何有效处理这一问题。这些模型不仅有助于深入理解公众情绪的变化趋势，而且还能应用于实际的在线服务中，提高用户满意度并促进更好的社会沟通。

资源推荐

资源详情

资源评论

Knowl Inf Syst

DOI 10.1007/s10115-015-0865-0

REGULAR PAPER

Tracking the evolution of social emotions with topic

models

Chen Zhu

· Hengshu Zhu

· Yong G e

Enhong Chen

· Qi Liu

· Tong X u

· Hui Xiong

Received: 13 December 2014 / Revised: 7 April 2015 / Accepted: 9 July 2015

Abstract Many of today’s online news Web sites have enabled users to specify different

types of emotions (e.g., angry or shocked) they have after reading news. Compared with

traditional user feedbacks such as comments and ratings, these speciﬁc emotion annotations

are more accurate for expressing users’ personal emotions. In this paper, we propose to

exploit these users’ emotion annotations for online news in order to track the evolution of

emotions, which plays an important role in various online services. A critical challenge

is how to model emotions with respect to time spans. To this end, we propose a time-

aware topic modeling perspective for solving this problem. Speciﬁcally, we ﬁrst develop

two models named emotion-Topic over Time (eToT) and mixed emotion-Topic over Time

(meToT), in which the topics of news are represented as a beta distribution over time and a

multinomial distribution over emotions. While they can uncover the latent relationship among

news, emotion and time directly, they cannot capture the evolution of topics. Therefore,

we further develop another model named emotion-based Dynamic Topic Model (eDTM),

where we explore the state space model for tracking the evolution of topics. In addition, we

demonstrate that all of proposed models could enable several potential applications, such as

emotion prediction, emotion-based news recommendations, and emotion anomaly detections.

Finally, we validate the proposed models with extensive experiments with a real-world data

set.

Enhong Chen

cheneh@ustc.edu.cn

Chen Zhu

zc3930155@gmail.com

School of Computer Science and Technology, University of Science and Technology of China,

Hefei 230027, China

Big Data Lab, Baidu Research, Beijing 100085, China

Department of Computer Science, The University of North Carolina at Charlotte, University City

Blvd, Charlotte, NC 28223-0001, USA

Management Science and Information Systems Department, Rutgers Business School,

Rutgers University, Newark, NJ 07102, USA

123

C. Zhu et al.

Keywords Social emotions · Topic models · Sentiment analysis

1 Introduction

With the increasing prosperity of Web 2.0, people are encouraged to have various social

interactions on the Web sites. A recent development trend of online news Web sites, such

as Yahoo! and Sina, is to allow readers to specify different types of emotions (e.g., angry

and shocked) after reading news. Compared with traditional users’ feedback (e.g., reviews,

tags), such speciﬁc emotion annotations are more accurate for expressing users’ personal

emotions. For example, Fig. 1a shows an example of users’ aggregated emotions at Sina

News,

which has six different kinds of emotions. Each user can choose one emotion, which

most accurately reﬂects his impression after reading, to annotate a piece of news. If we

collectively look at and analyze all user emotions from the news Web site, we may be able

to get a good picture about the overall emotions of online social media users, namely social

emotions. The social emotions often vary with respect to the topics of news and time and thus

have intrinsic dynamics. For example, Fig. 1b shows the distribution of aggregated social

emotions with respect to different time spans in our data set c ollected from Sina News. We

can observe different distributions of emotions along the time, which indicates that the social

emotions evolve over time. In fact, such evolution of social emotions is inherently driven by

the dynamic topics of news at different time. Capturing such dynamic characteristics of social

emotions is critically important for the successful development of various social services,

such as social opinion monitoring and social event detections. In the literature, there are recent

studies about social emotion-related problems. For example, some works focus on sentiment

analysis [24,35], social emotion analysis [6,13,18] of online documents and user emotion

modeling [3]. However, few of them have paid attention to the dynamic characteristics of

social emotions.

To this end, in this paper, we propose to exploit the users’ emotion annotations from

online news to track the evolution of social emotions. A critical challenge is how to model

emotions with respect to time spans. Along this line, we propose a time-aware topic modeling

perspective for solving this problem. Speciﬁcally, we ﬁrst develop a model named emotion-

Topic over Time (eToT), where we represent each topic of news with a beta distribution over

time and a multinomial distribution over emotions. In this model, the process of modeling

topics of news is affected not only b y the word co-occurrences but also the emotions and time.

Particularly, with some experimental observations, we ﬁnd that not all the topics have distinct

relationships with time, such as the common topics or the topics about news terminology.

Therefore, we further propose an extension of eToT model, named mixed emotion-Topic over

Time (meToT), in which a word can be generated from two different topic sets, i.e., one is time-

independent and the other is time-dependent. Indeed, similar to the popular Latent Dirichlet

Allocation (LDA) [5], news topics in both eToT and meToT are static which do not change

with respect to different time spans. However, while some researchers have revealed that the

latent topics of documents may evolve as time unfolds [4], eToT and meToT cannot capture the

dynamics of topics. Therefore, we propose another model, namely emotion-based Dynamic

Topic Model (eDTM), to capture the dynamics of news topics. In eDTM, we ﬁrst divide all

news into different segmentations with respect to their timestamps and then implement topic

modeling for each segmentation. Models learned from different segmentations are linked

together by the Markov state space model. In fact, all eToT, meToT, and eDTM can model

http://news.sina.com.cn/.

123

Tracking the evolution of social emotions with topic models

Fig. 1 a Examples of different user emotions. b The distribution of social emotions with respect to news in

different time spans

the evolution of social emotion effectively and help us understand the semantic relationships

between social emotion and news topics in different views. In addition, we demonstrate that all

of them are general models and could enable several potential applications of social media,

such as emotion prediction, emotion-based news recommendation and emotion anomaly

detection. Specially, the contributions of this paper can be summarized as follows.

– First, we provide a comprehensive study to track the evolution of social emotions by

exploiting the user emotion annotations from online news. Indeed, this study is vital for

the successful development of various social services.

– Second, we propose three novel topic models, i.e., eToT, meToT, and eDTM, for solv-

ing the problem of tracking the evolution of social emotion. Particularly, the proposed

models can effectively model the emotions, news and time from different views. Also,

we introduce some novel emotion-based applications enabled by the proposed models.

– Third, we carry out extensive experiments on a real-world data set which was collected

from Sina News to evaluate the proposed models. As shown in our experiments, the

proposed models are all effective for modeling social emotions and other enabled appli-

cations.

The remainder of this paper is organized as follows. In Sect. 2, we brieﬂy introduce some

related works of this paper. Section 3 introduces the details of proposed topic models. In

Sect. 4, we make some discussions about our models and introduce their potential applica-

tions. Section 5 reports the experimental results. Finally, we conclude this paper in Sect. 6.

2 Related work

The related works can be grouped into two categories: sentiment analysis and topic model.

We will give brief discussions on these related works and the relationships between them and

the proposed models in this paper.

2.1 Sentiment analysis

In the ﬁeld of sentiment analysis, existing researches mainly focus on two aspects, namely

sentiment c lassiﬁcation and sentiment information extraction. Among them, since sentiment

extraction aims to extract units that are related to emotion at sentence level or paragraph level,

it is less relevant to our scenario. Therefore, in the following, we only discuss the works on

sentiment classiﬁcation [13,16,18,26,31,37].

Classiﬁcation is a very fundamental problem in machine learning [26,39,40]. Tradition-

ally, sentiment classiﬁcation is o ften addressed by classic classiﬁcation approaches directly.

123

C. Zhu et al.

Recently, researchers have proposed a lot of new approaches for sentiment classiﬁcation and

begun to study the readers’ emotions. For instance, in [17], two ranking methods were pre-

sented to rank readers’ emotions. Lin et al. [18] also studied the readers’ emotions that are

triggered by articles, and the authors mainly focus on feature selection. Kozareva et a l. [13]

designed an approach for classifying headline emotion based on the information collected

from the World Wide Web. Besides, emoticon, as a kind of emotion label, has also been used

for sentiment classiﬁcation. Zhao et al. [38] built a system called MoodLens, in which 95

emoticons are mapped into four categories o f sentiments for the sentiment classiﬁcation of

Chinese tweets in Weibo. Liu et al. [20] treated the emoticons as labeled data and integrated

it and labeled data into a uniform sentiment classiﬁcation framework.

We can ﬁnd that most of existing works aim to study the sentiment directly from the

documents (words). Since there is a consensus that a document is composed of topics, it

is also reasonable to treat sentiment of a document as a combination of sentiment on these

topics. Therefore, in this paper, we focus on analyzing sentiment from the perspective of

the sentiment (emotion) of the topics. There have been some works about the emotion with

respect to topics. For instance, Mei et al. [22] modeled texts through a mixture of topic model

and sentiment model. Lin and He [16] also extended LDA to detect sentiment and topics.

However, none of them considers the temporal effect.

2.2 Topic models

Topic model is a tool for analyzing texts or other types of discrete data. Many different kinds

of topic models [4,5,21,33] have been proposed, among which Latent Dirichlet Alloca-

tion (LDA) [5] is very popular. Actually, the models proposed in this paper can be also traced

back to LDA. The basic assumption of LDA is that each document can be treated as a mixture

of topics and the generation of words in the document is attributable to one of those topics.

However, LDA is a static model that does not consider the inﬂuence of time, so it is inap-

propriate to be adopted in many applications, including social emotion analysis. To this end,

several topic models have examined topics and their changes with time. For instance, Topics

over Time (ToT) [33] model associates each topic with a beta distribution over time to directly

capture the topic changes over time. The relative simplicity of ToT makes it easy to inject its

ideas to other topic models. Different from ToT, Dynamic Topic Model (DTM) [4] uses state

space model on the natural parameters to model the evolution of topic and the variational

approximations depending on Kalman ﬁlters and wavelet regression. DTM requires that time

must be discretized, and thus how to determine the length of time span is an important prob-

lem. Wang et al. [32] solved this problem by replacing state space model with Brownian

motion which is actually an acontinuous generalization of state space model. Another way to

solve this problem was proposed in [11], which considers multi-scale time span to track the

topic changes. Topic Tracking Model (TTM) [9] is another Dynamic Topic Model for track-

ing consumer purchase behaviors. Compared with DTM, it simpliﬁes the model by replacing

Gaussian distribution with Dirichlet distribution so that the parameters could be estimated by

Gibbs sampling which is simpler compared with variational methods. In summary, there are

two main approaches to model the evolution of topics over time: treating time as an attribute

of topics, e.g., ToT; using state space model or Brownian motion on natural parameters, e.g.,

DTM and TTM.

Meanwhile, some work extended LDA from another perspective. The assumption behind

them, which is similar to that of meToT, is each word could come from several different

topic sets. Beyond bag of words, Wang et al. [34] introduced phrases and proposed topical

n-grams, in which a word may come from a topic-speciﬁc unigram distribution or a bigram

123

Tracking the evolution of social emotions with topic models

distribution. Following similar idea, Pual et al. [27] tried to detect cultural differences from

online blogs and forums. They assumed that an article from a special cultural group is a

mixture of some general topics and culture-special topics. Besides, some researchers have

modeled sentiment by this way. For instance, Titov and McDonald [31] proposed a Multi-

Aspect Sentiment model consisting of an extension of LDA, Multi-Grain LDA, and sentiment

predictors for sentiment summarization.

As introduced above, there have been some works aiming to analyze sentiment by topic

model, but none of them focus on analyzing social emotions with respect to temporal infor-

mation.

3 Modeling social emotion

In this section, ﬁrstly, some preliminaries and backgrounds will be introduced. Then, we

explain the details of our time-aware topic modeling solutions for analyzing social emotion.

3.1 Preliminaries and backgrounds

Recently, many of todays online news Web sites have enabled users to specify their emotions

after reading news. Therefore, for each piece of news, we can get not only the textual con-

tent and timestamp but also the emotion that was annotated by users. For example, Fig. 1a

illustrates the emotion annotations from 75 users, and these emotions are classiﬁed into six

categories, i.e., “Funny,” “Touched,” “Angry,” “Sad,” “Novel,” and “Shocked.” As shown in

Fig. 1a, all emotion annotations of a piece of news from different users are usually aggregated

together.

For social emotion modeling, intuitively, users’ emotion about a document (e.g., news) is

mainly decided by their emotion about the topics of the document, which could be extracted

by topic model. Indeed, topic model is an unsupervised method to effectively extract topics

from the texts based on the co-occurrence of words. Considering that different topics lead

to different user emotions, e.g., some topics tend to depress people but some other topics

may be very affecting, it is reasonable to associate each topic with a distribution with respect

to emotion. In this way, we can model the evolution of social emotions by introducing an

emotion and time generation layer into the topic models, in which the discovery of topics is

inﬂuenced by social emotions, co-occurring of words, and also temporal information.

Formally, we assume that a document d is a bag of words w

, size of which is N

,and

is from a vocabulary containing V words. Each document also has a timestamp t and

an observed multinomial distribution over emotions e. Thus, we can use a set of triples

D ={(w

, t

, e

),...,(w

, t

, e

)} to represent a collection of D documents. Next, we

will explain our solutions to model the evolution of social emotion from the corpus D.

3.2 Emotion-topic over time

In this subsection, we would propose a novel topic model named emotion-Topic over

Time (eToT) to directly uncover the latent relationship among documents, emotion, and time.

Compared with LDA, each topic discovered by eToT also has a d istribution with respect to

time and a distribution with respect to emotion. That is, besides understanding the content

of a topic, we can know the people’s impression of this topic (i.e., social emotion) and when

this topic emerges.

123

剩余27页未读，继续阅读

评论收藏

内容反馈

weixin_38742124

粉丝: 3
资源: 897

A Topic Modeling Perspective for Tracking the Evolution of Socia...

最新资源

A Topic Modeling Perspective for Tracking the Evolution of Socia...

topicmodeling

Topic Model

topic_modelling

Topic Medels主题模型

TOPSIS模型

GRM: Generalized Relation Modeling for Transformer Tracking结构图

argeted Topic Modeling for Focused Analysis

ViSAGE-A Virtual Laboratory for Simulation and Analysis of Social Group Evolution

The Neural Simulation Language A System for Brain Modeling

Modeling Purposeful Adaptive Behavior with the Principle of Maximum Causal Entro

Network Modeling and Simulation A Practical Perspective

The TV-APM interface: a web service for collaborative modeling

A tutorial on modeling and analysis of dynamic social networks

A Handbook on the Modeling of Internal Structures and Networks 2014

occlusion reasoning for multiple object tracking

Modeling of RF absorber for application in the design of anechoic chamber

Modeling, Identification and Control of a Quadrotor Aircraft

The unified modeling language reference manual

Cyber-Physical Systems A Computational Perspective

Modeling and Design of Electromagnetic Compatibility for High-Speed Printed epub

MISRA C_2012 Guidelines for the use of the C language in critical systems.pdf

Introduction to the Modeling and Analysis of Complex Systems

System Dynamics_ Modeling, Simulation, and Control of Mechatronic Systems

最新资源