Rationale-based Opinion Summarization

Haoyuan Li,Snigdha Chaturvedi

2024-03-30

Abstract:Opinion summarization aims to generate concise summaries that present popular opinions of a large group of reviews. However, these summaries can be too generic and lack supporting details. To address these issues, we propose a new paradigm for summarizing reviews, rationale-based opinion summarization. Rationale-based opinion summaries output the representative opinions as well as one or more corresponding rationales. To extract good rationales, we define four desirable properties: relatedness, specificity, popularity, and diversity and present a Gibbs-sampling-based method to extract rationales. Overall, we propose RATION, an unsupervised extractive system that has two components: an Opinion Extractor (to extract representative opinions) and Rationales Extractor (to extract corresponding rationales). We conduct automatic and human evaluations to show that rationales extracted by RATION have the proposed properties and its summaries are more useful than conventional summaries. The implementation of our work is available at

Computation and Language

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Traditional opinion summarization is too generalized and lacks supporting details**. Specifically, the summaries generated by existing opinion summarization methods are usually very general and cannot provide sufficient supporting evidence to help users make decisions.

To solve these problems, the author proposes a new paradigm - **rationale - based opinion summarization**. This new method not only generates representative opinions, but also provides one or more rationales for each representative opinion to enhance the informativeness and practicality of the summary.

### Main problems and solutions

1. **Problems with traditional opinion summarization**:
- **Over - generalization**: The summaries generated by traditional methods are often too general and lack specific details.
- **Lack of supporting evidence**: Traditional summaries do not provide enough reasons to support their conclusions, making it difficult for users to make decisions based on the summaries.

2. **Proposed solutions**:
- **Rationale - based opinion summarization**: The author proposes a new framework RATION, which can extract representative opinions and corresponding rationales. RATION evaluates and selects rationales by defining four ideal properties (relatedness, specificity, popularity, and diversity) and uses the Gibbs sampling method for extraction.

### Main contributions of the paper

- **Proposing a new opinion summarization paradigm**: Rationale - based opinion summarization, making the summary more specific and supportive.
- **Designing the RATION model**: This model includes two components - Opinion Extractor and Rationales Extractor, which can extract representative opinions and rationales from comments in an unsupervised manner.
- **Verifying the effectiveness of RATION through automatic and manual evaluations**: The experimental results show that the rationales generated by RATION are better than those of existing methods, and its summaries perform better in terms of non - redundancy, coherence, and practicality.

### Markdown representation of formulas

- **Relatedness**:
\[
e(s, G) = \max_{o \in G} p_{\text{ent}}(s, o)
\]
\[
\text{rel}(s) = \frac{e(s, G_i)}{\sum_{G_k \in G_s} e(s, G_k)}
\]

- **Specificity**:
Use the DeBERTa model to estimate the specificity score of a specific sentence.

- **Popularity**:
\[
\text{pop}(s) = \text{the centrality of node } s \text{ in the graph}
\]

- **Diversity**:
\[
\text{div}(s_1:k) = -\frac{1}{k(k - 1)} \sum_{i = 1}^{k} \sum_{j \neq i} \cos(\mathbf{v}_{s_i}, \mathbf{v}_{s_j})
\]

- **Joint Probability**:
\[
P(s_1:k) \propto \exp\left( \sum_{i = 1}^k \text{sal}(s_i) + \gamma \cdot \text{div}(s_1:k) \right)
\]

Through these formulas and methods, RATION can effectively generate high - quality rationale - based opinion summaries, thereby improving users' decision - making efficiency.