this is a placeholder image
An example showing the difference between the ECE task and the ECPE task.

Introduction

Emotion cause extraction (ECE) aims at extracting potential causes that lead to emotion expressions in text.

In the above figure, taken from the corpus by (Gui et al., 2016), there are five clauses in a document. The emotion happy is contained in the fourth clause. The authors denote this clause as emotion clause, which refers to a clause that contains emotions. It has two corresponding causes: a policeman visited the old man with the lost money in the second clause, and told him that the thief was caught in the third clause. The authors denote them as cause clause, which refers to a clause that contains causes.

Problem

  • Emotion must be annotated before cause extraction in ECE, which greatly limits its applications in real-world scenarios.
  • The way to first annotate emotion and then extract the cause ignores the fact that they are mutually indicative.

How it Solves

The authors propose a new task: Emotion Cause Pair Extraction (ECPE), which aims to extract all potential pairs of emotions and corresponding causes in a document.

For example, given the annotation of emotion: happy, the goal of ECE is to track the two corresponding cause clauses: a policeman visited the old man with the lost money and and told him that the thief was caught. While in the ECPE task, the goal is to directly extract all pairs of emotion clause and cause clause, including (The old man was very happy, a policeman visited the old man with the lost money) and (The old man was very happy, and told him that the thief was caught), without providing the emotion annotation happy.

Approach

The authors propose a two-step approach to address this new ECPE task.

1. Individual Emotion and Cause Extraction

Convert the emotion-cause pair extraction task into emotion extraction and cause extraction respectively. Two kinds of multi-task learning networks are proposed to model the two sub-tasks in a unified framework, with the goal to extract a set of emotion clauses \(E = \{c_1^e, ... , c_m^e\}\) and a set of cause clauses \(C = \{c_1^c, ... , c_n^c\}\) for each document.

a) Independent Multi-task Learning

A document contains multiple clauses: \(d = [c_1, c_2, ..., c_{|d|}]\), and each $c_i$ also contains multiple words $c_i = [w_{i,1}, w_{i,2}, …, w_{i,|ci|}]$.

To capture such a “word-clause-document” structure, the authors employ a Hierarchical Bi-LSTM network which contains two layers. The lower layer consists of a set of word-level Bi-LSTM modules, each of which corresponds to one clause, and accumulate the context information for each word of the clause. The upper layer consists of two components: one for emotion extraction and another for cause extraction. Each component is a clause-level BiLSTM which receives the independent clause representations $[s_1, s_2, …, s_{|d|}]$ obtained at the lower layer as inputs.

this is a placeholder image
The Model for Independent Multi-task Learning (Indep).

b) Interactive Multi-task Learning

To capture the correlation between emotion and cause, Interactive Multi-Task Learning network was proposed. Compared with Independent Multi-task Learning, the lower layer of Inter-EC is unchanged, and the upper layer consists of two components, which are used to make predictions for emotion extraction task and cause extraction task in an interactive manner. Each component is a clause-level BiLSTM followed by a softmax layer.

this is a placeholder image
Two Models for Interactive Multi-task Learning: (a) Inter-EC, which uses emotion extraction to improve cause extraction (b) Inter-CE, which uses cause extraction to enhance emotion extraction.

2. Emotion-Cause Pairing and Filtering

Pair the emotion set $E$ and the cause set $C$ by applying a Cartesian product to them. This yields a set of candidate emotion-cause pairs. The authors finally trained a filter to eliminate the pairs that do not contain a causal relationship between emotion and cause.

Metrics

The authors stochastically selected $90\%$ of the data for training and the remaining $10\%$ for testing. In order to obtain statistically credible results, they repeated the experiments $20$ times and reported the average result. Precision, recall, and F1 score were used as the metrics for evaluation.

Results

F1 scores of all models on the ECPE task are significantly improved by adopting the pair filter. These results demonstrate the effectiveness of the pair filter. Specifically, by introducing the pair filter, some of the candidate emotion-cause pairs in $P_{all}$ are filtered out, which may result in a decrease in the recall rate and an increase in precision.

The precision scores of almost all models are greatly improved (more than \(7\%\)), in contrast, the recall rates drop very little (less than \(1\%\)), which lead to the significant improvement in F1 score.

keep_rate indicates the proportion of emotion-cause pairs in $P_{all}$ that are finally retained after pair filtering.

this is a placeholder image
Experimental results of all proposed models and variants using precision, recall, and F1-measure as metrics on the ECPE task with or without the pair filter.

Limitations and Future Work

  • Goal of ECPE is not direct.
  • Mistakes made in the first step will affect the results of the second step.

In the future work, a one-step model that directly extracts the emotion-cause pairs would be more beneficial.

Takeaways

  • ECPE solves the shortcomings of the traditional ECE task that depends on the annotation of emotion before extracting cause, and allows emotion cause analysis to be applied to real-world scenarios.
  • Approach achieves comparable cause extraction performance to traditional ECE methods and removes the emotion annotation dependence.

References

  1. Xia, R., & Ding, Z. (2019). Emotion-Cause Pair Extraction: A New Task to Emotion Analysis in Texts. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 1003–1012. https://doi.org/10.18653/v1/P19-1096


Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

Leave a comment