This paper presents the Topic-Aspect Model (TAM), a Bayesian mixture model which jointly discovers topics and aspects. We broadly define an aspect of a document as a characteristic that spans the document, such as an underlying theme or perspective. Unlike previous models which cluster words by topic or aspect, our model can generate token assignments in both of these dimensions, rather than assuming words come from only one of two orthogonal models. We present two applications of the model. First, we model a corpus of computational linguistics abstracts, and find that the scientific topics identified in the data tend to include both a computational aspect and a linguistic aspect. For example, the computational aspect of GRAMMAR emphasizes parsing,

Abstract. In this study, we describe a conversational agent designed to support collaborative learning interactions between pairs of students. We describe a study in which we independently manipulate the social capability and goal alignment of the agent in order to investigate the impact on student learning outcomes and student perceptions. Our results show a significant interaction effect between the two independent variables on student learning outcomes. While there are only a few perceived differences related to student satisfaction and tutor performance as evidenced in the questionnaire data, we observe significant differences in student conversational behavior, which offer tentative explanations for the learning outcomes we will investigate in subsequent work.

With the proliferation of user-generated articles over the web, it becomes imperative to develop automated methods that are aware of the ideological-bias implicit in a document collection. While there exist methods that can classify the ideological bias of a given document, little has been done toward understanding the nature of this bias on a topical-level. In this paper we address the problem of modeling ideological perspective on a topical level using a factored topic model. We develop efficient inference algorithms using Collapsed Gibbs sampling for posterior inference, and give various evaluations and illustrations of the utility of our model on various document collections with promising results. Finally we give a Metropolis-Hasting inference algorithm for a semi-supervised extension with decent results. 1

In this paper we investigate the effect of the context of interaction on the extent to which a contributor’s perspective bias is displayed through their lexical choice. We present a series of experiments on political discussion data. Our experiments indicate that (i) when people quote contributors with an opposing view, they tend to quote the words that are less strongly associated with the opposing view. (ii) Nevertheless, in quoting their opponents, the displayed bias of their word distributions shifts towards that of their opponents. (iii) The personal bias of the speaker is displayed most clearly through the words that are not quoted, (iv) although characteristics of the quoted message do have a measurable effect on the words that are included in the contribution. And, finally, (v) posts are influenced by the displayed bias of previous posts in a thread.

Modeling reciprocity in social interactions with probabilistic latent space models

This paper describes cross-collection latent Dirichlet allocation (ccLDA), a probabilistic topic model that captures meaningful word co-occurrences across multiple text collections. The model is applied to three different applications: discovering cultural differences in blogs and forums from different countries, discovering research topics across multiple scientific disciplines, and comparing editorial differences between multiple media sources. A variety of qualitative and quantitative evaluations of ccLDA are performed, including log-likelihood measurements and performance measurements of the model used as a generative classifier. Improvements over previous work are demonstrated. Finally, possible extensions and modifications to the model are presented with promising results. 1

Latent variable models can be enriched with a multi-dimensional structure to consider the many latent factors in a text corpus, such as topic, author perspective and sentiment. We introduce factorial LDA, a multi-dimensional model in which a document is influenced by K different factors, and each word token depends on a K-dimensional vector of latent variables. Our model incorporates structured word priors and learns a sparse product of factors. Experiments on research abstracts show that our model can learn latent factors such as research topic, scientific discipline, and focus (methods vs. applications). Our modeling improvements reduce test perplexity and improve human interpretability of the discovered factors. 1

Abstract. Twitter as a new form of social media can potentially contain much useful information, but content analysis on Twitter has not been well studied. In particular, it is not clear whether as an information source Twitter can be simply regarded as a faster news feed that covers mostly the same information as traditional news media. In This paper we empirically compare the content of Twitter with a traditional news medium, New York Times, using unsupervised topic modeling. We use a Twitter-LDA model to discover topics from a representative sample of the entire Twitter. We then use text mining techniques to compare these Twitter topics with topics from New York Times, taking into consideration