[Topic-models] Joint topic-sentiment modeling

Mark Dredze mdredze at seas.upenn.edu
Sun May 17 00:23:20 EDT 2009


Hi Maojin,

Take a look at these papers:
- Supervised LDA (SLDA) Blei & McAuliffe, 2007
- A Joint Model of Text and Aspect Ratings for Sentiment
Summarization. Ivan Titov, Ryan McDonald, ACL 08
- Modeling Online Reviews with Multi-Grain Topic Models. Ivan Titov,
Ryan McDonald. WWW 08

Mark


On Sat, May 16, 2009 at 6:13 PM, Maojin Jiang <jianmao at iit.edu> wrote:
> LDA and similar models have been mostly studied and applied to modeling
> topics. With more texts that are rich-in subjectivity available on the
> Internet, we are facing new challenges to opinion/sentiment or
> subjectivity analysis, which may find their applications in product
> recommendation, reputation track and analysis, automatic e-rule making,
> etc. So, in this regard, LDA and the like may be used to model topic and
> sentiment simultaneously. A possible hidden variable to be modeled
> governs a series of strong association between topics and sentiments.
> These bindings of topics and sentiments may reveal purchase preference
> of buyers (e.g. prefer new features to new look or on the contrary when
> people buy new mobile phones); or they reveal underlying ideology that
> governs people's political orientation (e.g. prefer free market or
> prefer more government control).
>
> Existing LDA-based topic models see no difference among words in terms
> of semantics. So, several basic and preliminary tasks need to be solved,
> including topic detection, sentiment extraction and association of
> topics and sentiments, among other things. But some simplified treatment
> may be made for these tasks to facilitate the application of graphical
> models to topic-sentiment analysis.
>
> Want to know if there is previous or ongoing work of applying LDA or
> similar models (generative or discriminative) to applications such as
> business intelligence, review analysis, political orientation
> categorization and customer behavior categorization, to name a few.
>
> For those who might be interest in this topic, there is now a relevant
> event to submit interesting work or results in this regard. TSA'09 is a
> CIKM workshop on topic-sentiment analysis. CFP is now available on its
> website here:
>
> http://sites.google.com/site/tsa2009workshop
>
> Thanks!
> --
> Maojin Jiang
> _______________________________________________
> Topic-models mailing list
> Topic-models at lists.cs.princeton.edu
> https://lists.cs.princeton.edu/mailman/listinfo/topic-models
>


More information about the Topic-models mailing list