Download - EMNLP 2014: Opinion Mining with Deep Recurrent Neural Network

Komachi Lab. M1 Peinan ZHANG

Opinion Mining with Deep Recurrent Neural Networks EMNLP 2014 reading @ Komachi Lab 2014/12/04

All figures and tables in this slide are cited from the paper.


Introduction Fine-grained opinion analysis aims to detect the subjective expressions in

n  a text (e.g. “hate” or “like”)

and to characterize their n  intensity (e.g. “strong” or “weak”) n  sentiment (e.g. “negative” or “positive”)

as well as to identify n  the opinion holder: the entity expressing the opinion n  the target, or topic of the opinion: what the opinion is about

2


Introduction: Tasks Detection of opinion expressions [Wiebe et al., 2005]

DSEs (Direct Subjective Expressions) consist of explicit mentions of private states or speech events expressing private states

ESEs (Expressive Subjective Expressions) consist of expressions that indicate sentiment, emotion, etc., without explicitly conveying them

3


Introduction: Examples

DSE: explicitly express an opinion holder’s attitude ESE: indirectly express the attitude of the writer

4


Introduction: Labeling

Opinion extraction has often been tackled as a sequence labeling problem in previous work.

n  B: the beginning of an opinion-related expression n  I: tokens inside the opinion-related expression n  O: tokens outside any opinion-related class

5


Introduction: Methods (1/3) CRFs (Conditional Random Field)

variants of CRF approaches have been successfully applied to opinion expression extraction using this token-based view.

semiCRF (state-of-the-art) relaxes the Markovian assumption inherent to CRFs and operates at the phrase level rather then the token level, allowing the incorporation of phrase-level features.

But those CRFs hinges critically on access to an appropriate feature set, typically based on

constituent, dependency parse trees, manually crafted opinion lexicons, named entity tagger and other preprocessing

6


Introduction: Methods (2/3) RNN (Recurrent Neural Network)

n  latent features are modeled as distributed dense vectors of hidden layers

n  can operate on sequential data of variable length n  it can also be applied as a sequence labeler

bidirectional RNN n  incorporate information from preceding as well as following

tokens n  allowing a lower dimensional dense input representation n  more compact networks

7


Introduction: Methods (3/3) Deep Recurrent Network

n  lower levels capture short term interactions among words n  higher layers reflect interpretations aggregate over longer spans n  such hierarchies might better model the multi-scale language

effects

Deep Bidirectional RNN (the proposed method) motivated by the recent success of deep architectures in general and deep recurrent networks in particular

8


Agenda Introduction

1.  Tasks2.  Examples3.  Labeling4.  Methods

Methodology 1.  Recurrent Neural Network2.  Bidirectionality3.  Deep in Space

9

Experiments 1.  Data and Metrics2.  Baselines3.  Tuning and Training4.  Results and Discussion

Conclusion


Methodology: Recurrent Neural Network Elman-type network [Elman, 1990]

t: step, h0 = 0 h: hidden layer

x: input layer

y: final output layer

f: nonlinear function (e.g. sigmoid)

g: output nonlinearity (e.g. softmax)

10

W, V: weight matrices between the input and hidden layer

U: output weight matrix

b, c: bias vectors


Methodology: Recurrent Neural Network Problem with this model:

the Elman-style unidirectional RNN

lack the representational power to

model this task.

For example:n  I did not accept his suggestion.n  I did not go to the rodeo.

The first example has a DSE phrase “did not accept”.However, any such RNN will assign the same labels for the word “did” and “not” in both sentences, since the preceding sequences (past) are the same.

11


Methodology: Bidirectionality bidirectional RNN [Schuster et al., 1997]

→ : forward step (representations of the past) ← : backward step (representations of the future) h0→ = hT+1← = 0

Note: the forward and backward parts of the network are independent of each other until the output layer when they are combined.

12


Methodology: Depth in Space (1/2) deep RNN: constructed by stacking Elman-type RNNs on top of each other when i > 1 Intuitively, every layer of the deep RNN treats the memory sequence of the previous layer as the input sequence, and computes its own memory representation.

13


Methodology: Depth in Space (2/2) bidirectional deep RNN

when i > 1

To compute the output layer, we only employ the last memory layers.

14


Experiments: Data and Metrics Data

MPQA 1.2 corpus [Wiebe et al., 2005] n  535 news articles, 11,111 sentences n  manually annotated w/ both DSEs and ESEs at phrase level n  135 development set n  10-fold cross validation over remaining 400 documents

Evaluation Metrics Binary Overlap

count every overlapping match between a predicted and true expression as correct

Proportional Overlap impart a partial correctness, proportional to the overlapping amount, to each match

All statistical comparisons are done using a two-sided paired t-test with a confidence level of α = .05 15


Experiments: Baselines Baselines

n  CRF and semiCRF n  Features: words, POS tag, membership in a manually constructed

opinion lexicon

Word Vectors (+VEC) n  versions of the baselines that have access to pre-trained word

vectors n  CRF+VEC: as continuous features per every token n  semiCRF+VEC: simply take the mean of every word vector for a

phrase-level vector representation n  300-dimensinal n  trained on part of Google News Dataset (～100B words)

16


Experiments: Tuning and Training Regularizer

Dropout: randomly set entries of hidden representations to 0 with a probability called dropout rate

Network Training n  use SGD with fixed learning rate .005 n  update weights after minibatches of 80 sentences n  run 200 epochs n  initialized from small random uniform noise

17


Experiments: Results and Discussion Bidirectional vs. Unidirectional

Shallow biRNN vs. uniRNN n  each network has the same number of total parameters n  65 hidden units for the unidirectional network n  36 hidden units for the bidirectional network DSEs n  Proportional Overlap: 63.83 vs. 60.35 n  Binary Overlap: 69.31 vs. 68.31

ESEs n  Proportional Overlap: 54.22 vs. 51.51 n  Binary Overlap: 65.44 vs. 63.65

Thus, we will not include comparisons to the unidirectional RNNs in the remaining experiments. 18


Experiments: Results and Discussion

19

Adding Depth Bold: Best result Asterisk: statistically indis-tinguishable performance w/ respect to the best

n  for both DSE and ESE, 3-layer RNN provide the best results

n  2, 3 and 4-layer RNNs show equally good performance for certain sizes

n  adding additional layers degrades performance


Experiments: Results and Discussion Compare with other methods

20


Conclusion p Explored an application of deep recurrent neural

networks to the task of sentence-level opinion expression.

p deep RNNs outperformed shallow RNNs. p deep RNNs outperformed pervious (semi)CRF

baselines. p One potential future direction is to explore the

effects of pre-training.

21


Agenda Introduction

1.  Tasks2.  Examples3.  Labeling4.  Methods

Methodology 1.  Recurrent Neural Network2.  Bidirectionality3.  Deep in Space

22

Experiments 1.  Data and Metrics2.  Baselines3.  Tuning and Training4.  Results and Discussion

Conclusion

Download - EMNLP 2014: Opinion Mining with Deep Recurrent Neural Network

Top Related