Guiding Theme A3: Opinion and Sentiment - extrapropositional aspects of discourse

Guiding theme A3 extends the contributions of A1 and A2 by analyzing extra-propositional aspects of meaning: factivity vs. modality, as well as the sentiment associated with an entity or an event. These aspects of meaning are not only relevant for content extraction for summarization, but lead directly to aspect-based summarization, which highlights different perspectives and opinions about specific contents, or the pros and cons of a situation.

Successful Ph.D. projects may address different aspects of the characterization of extra-propositional meaning in a multi-document summarization setting. Core aspects are the recognition of extra-propositional aspects of meaning, their categorization, and the scope they may take beyond the sentence level. Beyond recognition and classification based on traditional models, more fine-grained sentiments, their polarity and their scope will be encoded in a graph-based discourse representation. Criteria for the identification of scope include tense, mood and discourse markers, but also coherence indicators.

Learning discourse models for the assignment of extra-propositional meaning involves sentiment classification as well as argument structure to identify opinion or sentiment holders. Of particular importance in AIPHES are domain and genre effects in the recognition and classification of extra-propositional meanings. Important questions involve compositionality and the representation of unmarked scopes within and across sentences (see e.g. [1] Socher et al. 2013). Cross-document alignments in a multi-document summarization task support the categorization of unmarked structures, but may also identify conflicting perspectives. Specific learning methods may include ranking (C2) or deep learning (C3).

This guiding theme will closely interact with themes A1 and A2 in Area A and with B2 in Area B.

Poster (in German)

Example thesis topics

  • Sentiment and opinion in discourse graphs
  • Sentiment and opinion in extended contexts
  • Motifs over sentiment and opinion in aspect-oriented multi-document summarizations

References

[1] Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, Christopher Potts (2013): Recursive Deep Models for Sentiment Compositionality Over a Sentiment TreebankProceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, Seattle, Washington, USA.

Pang, B. und Lee, L. (2008). Opinion Mining and Sentiment Analysis. Foundations and Trends in Information Retrieval, 2(1–2):1–135.

Liu, B. (2012). Sentiment Analysis and Opinion Mining. Morgan & Claypool Publishers.

A A A | Drucken Print | Impressum Impressum | Sitemap Sitemap | Suche Search | Kontakt Contact | Webseitenanalyse: Mehr Informationen
zum Seitenanfangzum Seitenanfang