A Pattern Matching Based Model for Implicit Opinion Question Identification

Hadi Amiri, Zheng-Jun Zha, Tat-Seng Chua

Abstract

Identifying Implicit Opinion Questions (IOQs) is a significant challenge in community Question Answering (cQA) as these queries often lack explicit sentiment markers. This paper proposes a novel framework that combines automated lexicon expansion, sequential pattern mining, and bipartite graph matching to identify IOQs. By focusing on structural patterns rather than just keywords, the proposed "OpFact" model achieves superior performance across multiple domains, significantly outperforming traditional unigram and lexicon-based baselines.

Introduction: The Challenge of Implicit Subjectivity

In community Question Answering (cQA) platforms like Yahoo! Answers, distinguishing between factual inquiries and opinion-seeking questions is vital for routing queries to appropriate solvers or aggregation engines. While explicit opinion questions are easily identified by sentiment words (e.g., "best," "worst"), Implicit Opinion Questions (IOQs) pose a greater challenge. These questions seek subjective advice or diverse perspectives using neutral vocabulary.

Examples of Implicit Opinion Questions (IOQs) which lack explicit sentiment markers but seek subjective answers.

As shown in the examples above, questions like "What can one do for a 50th wedding anniversary?" contain no obvious opinion words but clearly request subjective suggestions. Traditional sentiment analysis tools often fail here due to limited lexicons and a lack of structural understanding.

Methodology

The authors propose a multi-stage pipeline designed to recognize the "shape" of an opinion question:

  1. Automated Lexicon Expansion: To capture informal or domain-specific sentiment, the system identifies new opinion words through "interchangeability." Words that share similar linguistic contexts with known sentiment markers are added to the lexicon.
  2. Sequential Pattern Mining: Using the PrefixSpan algorithm, the system extracts frequent word sequences (e.g., <{what}{can}{do}>). This allows the model to identify structural indicators of subjectivity.
  3. Bipartite Graph Matching: To validate newly discovered patterns, the system models the similarity between a new pattern and a known explicit pattern as a bipartite graph, seeking a maximum weight match to confirm structural analogy.
  4. One-Class SVM for IOQs: For the most difficult cases—questions with entirely neutral vocabulary—a one-class Support Vector Machine (SVM) is trained on n-gram features of known implicit patterns.

Experimental Results

The model was evaluated on two primary datasets: a general benchmark (D1) across five categories and a specialized IOQ dataset (D2) from the Travel and Family domains.

Performance comparison (F1-score) of the proposed OpFact model against various baselines across five domains.

As illustrated in the results table, the full "OpFact" model achieved an average F1-score of 75.5% on the D1 dataset, significantly outperforming the Unigram baseline (68.9%) and models relying solely on existing lexicons (55.8%). The improvement was particularly pronounced in informal categories like Sport and Science. Furthermore, on the specialized D2 dataset targeting pure IOQs, the model achieved an accuracy of 79.0%.

Conclusion

By prioritizing structural patterns over individual keywords, the proposed framework effectively bridges the gap in identifying implicit subjectivity. This approach allows QA systems to evolve alongside changing internet slang and informal language. Future work will focus on extending this pattern-based methodology to other user-generated content types, such as micro-blogs and product reviews.

Summary updated

Abstract

This paper presents the results of developing subjectivity classifiers for Implicit Opinion Question (IOQ) identification. IOQs are defined as opinion questions with no opinion words. An IOQ example is "will the U.S. government pay more attention to the Pacific Rim?" Our analysis on community questions of Yahoo! Answers shows that a large proportion of opinion questions are IOQs. It is thus important to develop techniques to identify such questions. In this research, we first propose an effective framework based on mutual information and sequential pattern mining to construct an opinion lexicon that not only contains opinion words but also patterns. The discovered words and patterns are then combined with a machine learning technique to identify opinion questions. The experimental results on two datasets demonstrate the effectiveness of our approach.
BibTeX
@inproceedings{amiri2013pattern,
  author = {Hadi Amiri and Zheng-Jun Zha and Tat-Seng Chua},
  title = {{A Pattern Matching Based Model for Implicit Opinion Question Identification}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2013},
  doi = {10.1609/aaai.v27i1.8604},
}
Figure