Difference between revisions of "Ritter et al NAACL 2010. Unsupervised Modeling of Twitter Conversations"
m |
m |
||
| Line 8: | Line 8: | ||
This [[Category::Paper]] describes a [[UsesMethod::topic model]] based approach to model dialogue acts. | This [[Category::Paper]] describes a [[UsesMethod::topic model]] based approach to model dialogue acts. | ||
| − | Whereas previous work has often required the manual construction of a dialogue act inventory, this paper proposes a series of ''unsupervised'' conversation models, where the discovery of acts amounts to clustering utterances with similar conversational roles. | + | Whereas previous work has often required the manual construction of a dialogue act inventory, this paper proposes a series of ''unsupervised'' conversation models, where the discovery of acts amounts to clustering utterances with similar conversational roles. Specifically, the authors address this task using conversations on Twitter. |
== Brief description of the method == | == Brief description of the method == | ||
| + | |||
| + | The authors propose 2 models to discover dialogue acts in an unsupervised manner. | ||
| + | |||
| + | === Conversation Model === | ||
| + | |||
| + | The base model, the Conversation model, is inspired by the content model proposed by [[RelatedPaper::Barzilay and Lee 2004 Catching the drift: Probabilistic content models, with applications to generation and summarization|Barzilay and Lee (2004)]] for multi-document summarization. | ||
| + | |||
| + | [[File:ritter-naacl2010-cmodel.png|400px]] | ||
| + | |||
| + | Here, each conversation <math>C</math> is a sequence of dialogue acts <math>a</math>, and each act produces a post, represented by a bag of words shown using the <math>W</math> plates. The assumption is that each post in a Twitter conversation is generated by a single act. | ||
| + | |||
| + | === Conversation + Topic Model === | ||
| + | |||
| + | Since twitter conversations are not restricted to any particular topic, the Conversation Model tends to discover a '''mixture''' of dialogue and topic structure. In order to address this weakness, the authors propose an extended Conversation + Topic model. | ||
| + | |||
| + | [[File:ritter-naacl2010-ctmodel.png|400px]] | ||
| + | |||
| + | In this model, each word in a conversation is generated from one of three sources: | ||
| + | # The current post's dialogue act | ||
| + | # The conversation's topic | ||
| + | # General English | ||
| + | The model includes a conversation-specific word multinomial <math>\theta_k</math> that represents the topic, as well as a universal general English multinomial <math>\psi_E</math>. A new hidden variable, <math>s</math> determines the source of each word, and is drawn from a conversation-specific distribution over sources <math>\pi_k</math>. | ||
== Experimental Result == | == Experimental Result == | ||
| Line 20: | Line 42: | ||
== Related Papers == | == Related Papers == | ||
| − | The conversation model is inspired by [[RelatedPaper::Barzilay and Lee 2004 Catching the drift: Probabilistic content models, with applications to generation and summarization|Barzilay and Lee | + | The conversation model is inspired by [[RelatedPaper::Barzilay and Lee 2004 Catching the drift: Probabilistic content models, with applications to generation and summarization|Barzilay and Lee (2004)]]. |
== Study Plan == | == Study Plan == | ||
Revision as of 17:51, 30 September 2012
Contents
Citation
Alan Ritter, Colin Cherry, and Bill Dolan. Unsupervised Modeling of Twitter Conversations. In Proc of NAACL 2010
Online Version
Unsupervised Modeling of Twitter Conversations.
Summary
This Paper describes a topic model based approach to model dialogue acts. Whereas previous work has often required the manual construction of a dialogue act inventory, this paper proposes a series of unsupervised conversation models, where the discovery of acts amounts to clustering utterances with similar conversational roles. Specifically, the authors address this task using conversations on Twitter.
Brief description of the method
The authors propose 2 models to discover dialogue acts in an unsupervised manner.
Conversation Model
The base model, the Conversation model, is inspired by the content model proposed by Barzilay and Lee (2004) for multi-document summarization.
Here, each conversation is a sequence of dialogue acts , and each act produces a post, represented by a bag of words shown using the plates. The assumption is that each post in a Twitter conversation is generated by a single act.
Conversation + Topic Model
Since twitter conversations are not restricted to any particular topic, the Conversation Model tends to discover a mixture of dialogue and topic structure. In order to address this weakness, the authors propose an extended Conversation + Topic model.
In this model, each word in a conversation is generated from one of three sources:
- The current post's dialogue act
- The conversation's topic
- General English
The model includes a conversation-specific word multinomial that represents the topic, as well as a universal general English multinomial . A new hidden variable, determines the source of each word, and is drawn from a conversation-specific distribution over sources .
Experimental Result
Data: The corpus consists of about 1.3 million conversations in a 2 month period during the summer of 2009, with each conversation containing between 2 and 243 posts. The dataset was formerly available at http://homes.cs.washington.edu/~aritter/twitter_chat/ (asked by Twitter to be taken down).
Discussion
Related Papers
The conversation model is inspired by Barzilay and Lee (2004).
