A Sentiment Analysis Method to Better Utilize User Profile and - - PowerPoint PPT Presentation

▶
a sentiment analysis method to better utilize user
SMART_READER_LITE
LIVE PREVIEW

A Sentiment Analysis Method to Better Utilize User Profile and - - PowerPoint PPT Presentation

Nov 08, 2023 34 likes •396 views

A Sentiment Analysis Method to Better Utilize User Profile and Product Information Capstone Project Presentation Mingyu MA (Derek) derek.ma@connect.polyu.hk BSc (Hons) Computing, 14110562D Su Supervisor: Prof. Qin n LU Co Co-ex exam

slide-1
SLIDE 1

A Sentiment Analysis Method to Better Utilize User Profile and Product Information

Capstone Project Presentation

Mingyu MA (Derek)

derek.ma@connect.polyu.hk BSc (Hons) Computing, 14110562D Su Supervisor: Prof. Qin n LU Co Co-ex exam aminer er: Dr. Ajay ay Kumar ar PATHAK 2nd

nd As

Assessor: Dr. Richard LUI

slide-2
SLIDE 2

Introduction Related Work Model Design Evaluation and Analysis Conclusion and Future Work Contents

1 / 34

slide-3
SLIDE 3

Businesses would like to know users’ opinions Users can be benefited from others’ opinions

2 / 34

post product reviews post video reviews reviews data users’ opinions to improve services ratings and opinions

  • f other customers

Introduction

slide-4
SLIDE 4

3 / 34

methods of detecting, analyzing, and evaluating people’s state of mind towards events, issues, or any other

  • interest. (Yadollahi et al., 2017)

Sentiment Analysis

Introduction

slide-5
SLIDE 5

4 / 34

user product reviews: main document user profile user’s history user’s preferences ... product information product property

  • ther user’s opinions

…

Background Info Is Available

Introduction provide domain knowledge more facts and possibilities

slide-6
SLIDE 6
  • User’s perspective
  • Mean/lenient user
  • Product’s perspective
  • Type, category
  • Different background information

influences the results in different perspectives

5 / 34

+

Background Information Is Not Unified

Introduction

slide-7
SLIDE 7

A new sentiment analysis model

  • utilize user and product information
  • reflect impacts from user profile and

product information separately

6 / 34

Objectives

Introduction

slide-8
SLIDE 8

7 / 34 (Tang et al., 2014), (Kim, 2014)

NN as classifier for text classification RNN, LSTM

Neural-network- based Approaches

(Wang and Manning, 2012)

Linear model or kernel methods on lexical features

Traditional Way

(Yang et al., 2016) (Long et al., 2017)

Focus more on important text and add more associate data like eye-tracking data

Attention

Machine-Learning-based Sentiment Analysis

Related Work

slide-9
SLIDE 9

8 / 34

  • Memory network

(Tang, Qin and Liu, 2015; Dou, 2017)

  • RNN + external memory
  • Use external info as attention

(Chen et al., 2016)

  • State-of-the-art
  • All consider user profile and

product information as single representation

>

Utilizing User Profile and Product Information in Sentiment Analysis

User and Product Info in Sentiment Analysis

Related Work xt a

slide-10
SLIDE 10

9 / 34

JUPMN Joint User and Product Memory Network Model Design

slide-11
SLIDE 11

10 / 34

Model Overview

Model Design

Hierarchical LSTM with Attention

Document d

UMN

U(d)

^

PMN

Joint Mechanism

P(d)

^

D Document d (text) (numeric vector) Sentiment Prediction

Input & Output

  • Input
  • Document d
  • A writer u
  • A target p
  • Output
  • Discrete

sentiment prediction

slide-12
SLIDE 12

11 / 34

Model Overview

Model Design

Hierarchical LSTM with Attention

Document d

UMN

U(d)

^

PMN

Joint Mechanism

P(d)

^

D Document d (text) (numeric vector) Sentiment Prediction

Structure Part 2: Memory Networks Part 1: Document Embedding

slide-13
SLIDE 13

Hierarchical LSTM with Attention

Hierarchical Long Short-Term Memory Network

Model Design

12 / 34

> Part 1: Document Embedding

slide-14
SLIDE 14

Hierarchical LSTM with Attention

  • Word-sentence-

document level convention (Chen et al., 2016)

  • Add attention in LSTM

layers

  • With user and product

attention

  • With eye-tracking

cognition attention

Hierarchical Long Short-Term Memory Network

Model Design

13 / 34

> Part 1: Document Embedding

slide-15
SLIDE 15

14 / 34

Part 2: Memory Networks

Model Design

Hierarchical LSTM with Attention

Document d

UMN

U(d)

^

PMN

Joint Mechanism

P(d)

^

D Document d (text) (numeric vector) Sentiment Prediction

Part 2: Memory Networks Part 1: Document embedding

slide-16
SLIDE 16

Document d

(embedded by hierarchical LSTM) Softmax Sentiment Prediction

WU wU

U M N

WP wP

U(d)

(embedded by hierarchical LSTM)

...

^

Attention Layer 3

d2

a3

Attention Layer 2

d1

a2

Attention Layer 1

d0

a1

P M N

Attention Layer 3

d2

a3

Attention Layer 2

d1

a2

Attention Layer 1

a1

d0

P(d)

(embedded by hierarchical LSTM)

...

^

d3

U

d3

P

Model Design

15 / 34

> Part 2: Memory Networks

Attention Layer 3

slide-17
SLIDE 17

Attention Layer 3

Input Output

ATT W External Memory

ak dk-1

Attention Layer k

16 / 34

Structure of Attention Layers

Model Design > Part 2: Memory Networks

  • Attention weight
  • Output of attention layer
slide-18
SLIDE 18

17 / 34

  • IMDB
  • Diao et al., 2014
  • Yelp 13, Yelp 14
  • Tang et al., 2015a

Three Benchmark Datasets

Benchmark Datasets and Performance Metrics

Evaluation and Analysis

slide-19
SLIDE 19

18 / 34

Three Benchmark Datasets

Benchmark Datasets and Performance Metrics

Evaluation and Analysis

slide-20
SLIDE 20

19 / 34

Performance Metrics

Benchmark Datasets and Performance Metrics

Evaluation and Analysis

slide-21
SLIDE 21

20 / 34

Experimental Results

JUPMN and Comparison Models

Evaluation and Analysis

Group 1: simple methods based

  • n language features

Group 2: models using machine learning Group 3: models with user profile and product information in machine learning

slide-22
SLIDE 22

21 / 34

Experimental Results Findings

  • JUPMN outperforms

the state-of-the-art model

  • Generally Group 2

performs better than Group 1, Group 3 performs better than Group 2

  • Exceptions exist
  • TextFeature
  • LSTM+CBA

JUPMN and Comparison Models

Evaluation and Analysis

slide-23
SLIDE 23

22 / 34

Four aspects of configurations

JUPMN with Different Configurations

Evaluation and Analysis

Number of Hops Memory Size

Hierarchical LSTM with Attention

Document d

UMN

U(d)

^

PMN

Joint Mechanism

P(d)

^

Document d (text) (numeric vector) Sentiment Prediction

D

Joint Weights Importance of User vs Product Memory Network

slide-24
SLIDE 24

23 / 34

  • User profile influences

sentiments of movie reviews more

  • Product information

influences sentiments of restaurants reviews more

  • JUPMN-U
  • With only User Memory Network
  • JUPMN-P
  • With only Product Memory Network

Importance of User vs Product Memory Network

Evaluation and Analysis > JUPMN with Different Configurations

Experimental Results Observations

slide-25
SLIDE 25

24 / 34

Investigating by Checking Joint Weights

  • Verified the hypothesis

Average joint weight for three datasets

Importance of User vs Product Memory Network

Evaluation and Analysis > JUPMN with Different Configurations

Joint weights for three datasets

slide-26
SLIDE 26

25 / 34

10 users give average highest/lowest rating score

Importance of User vs Product Memory Network

Evaluation and Analysis > JUPMN with Different Configurations

Investigating by Word Frequency Plotting

For IMDB dataset 10 movies have average highest/lowest rating score

slide-27
SLIDE 27

26 / 34

Importance of User vs Product Memory Network

Evaluation and Analysis > JUPMN with Different Configurations

Investigating by Word Frequency Plotting

For IMDB dataset

For movies reviews

  • Users’ words are very different
  • Products’ words are very objective
slide-28
SLIDE 28

27 / 34

Importance of User vs Product Memory Network

Evaluation and Analysis > JUPMN with Different Configurations

Investigating by Word Frequency Plotting

For Yelp dataset

For restaurants reviews

  • Users’ words are not distinguishable
  • Products’ words shows the sentiments
slide-29
SLIDE 29

28 / 34

  • Smaller hop works

better

  • Possible explanations
  • Data distortion
  • Over-fitting

Number of Hops (Computational Layers)

Evaluation and Analysis > JUPMN with Different Configurations

Experimental Results Observations

slide-30
SLIDE 30

29 / 34

  • Larger memory helps
  • When memory size reaches 75,

no longer improve

  • There is not enough

documents

Memory Size

Evaluation and Analysis > JUPMN with Different Configurations

slide-31
SLIDE 31

30 / 34

  • Weighted version works better
  • Weight help to balance the influences of UMN and PMN

Joint Weights

Evaluation and Analysis > JUPMN with Different Configurations

JUPMN (not weighted) JUPMN

slide-32
SLIDE 32

31 / 34

  • What is this user’s opinion?
  • Cite negative reviews to praise
  • JUPMN can learn the features
  • f this user
  • This user is a science fiction

movie

  • JUPMN can learn the features
  • f this movie(product)
  • This movie is relative great

according to other reviews

Case Study

Evaluation and Analysis

slide-33
SLIDE 33

32 / 34

  • More knowledge in memory network
  • Application of JUPMN in more languages datasets

Future Work Conclusion

  • Proposed JUPMN
  • JUPMN outperforms the state-of-the-art sentiment analysis model
  • Analysis on different configuration is employed
  • Research paper

Yunfei Long*, Mingyu Ma*, Rong Xiang, Qin Lu, Chu-Ren Huang. Fusing User Memory and Product Memory for Sentiment Classification. (*: Equal contribution)

Conclusion and Future Work

slide-34
SLIDE 34

33 / 34

References

slide-35
SLIDE 35

34 / 34

References

slide-36
SLIDE 36

Thanks!

Mi Mingyu MA MA (Dere rek) supervised by Pr Prof.

  • f. Qin LU

A Sentiment Analysis Method To Better Utilize User Profile and Product Information