Cross-Cultural Analysis of Blogs and Forums with Mixed-Collection Topic Models

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1408–1417, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Cross-Cultural Analysis of Blogs and Forums with Mixed-Collection Topic Models Michael Paul and Roxana Girju University of Illinois at Urbana Champaign Urbana, IL 61801 {mjpaul2, girju}@illinois.edu Abstract This paper presents preliminary results on the detection of cultural differences from people’s experiences in various countries from two perspectives: tourists and lo- cals. Our approach is to develop proba- bilistic models that would provide a good framework for such studies. Thus, we pro- pose here a new model, ccLDA, which extends over the Latent Dirichlet Alloca- tion (LDA) (Blei et al., 2003) and cross- collection mixture (ccMix) (Zhai et al., 2004) models on blogs and forums. We also provide a qualitative and quantitative analysis of the model on the cross-cultural data. 1 Introduction In today’s society, people from different cultural backgrounds have to understand each other, inter- act on a daily base and travel to or work in more than one country. Understanding cultural diver- sity, as well as addressing the need to communi- cate effectively across cultural divides, have be- come imperative in almost every aspect of life. These constitute an important language aspect since the lack of such cultural awareness can lead to misinterpretations. This paper presents preliminary results on the detection of cultural differences from people’s ex- periences in various countries from two perspec- tives: tourists and locals. Since the advent of Web 2.0, user-generated data in the form of blogs and newsgroup messages have reached high propor- tions. In this paper we take advantage of such resources of blogs and forums to perform various cross-cultural analyses. Our approach is to develop probabilistic mod- els that would provide a good framework for such studies. Thus, we propose here a new model, ccLDA, which extends over the Latent Dirichlet Allocation (LDA) (Blei et al., 2003) and cross- collection mixture (ccMix) (Zhai et al., 2004) models. Our contribution is as follows: (1) Unsupervised topic models such as LDA are elegant and flexible approaches to clustering large collections of unannotated data. These models, however, have conceptually focused on one single collection of text which is inadequate for compar- ative analyses of text. We thus develop an LDA-based model that can not only discover topics but also model their simi- larities and differences across multiple text collec- tions. (2) We improve on similar previous work by craft- ing a model that can better generalize data and is less reliant on user-defined parameters. (3) We apply our new model on blogs and forums to identify cross-cultural differences. Thus, different models can be compared to re- flect different hypotheses about the data. The paper is organized as follows. In Section 2 we summarize relevant previous work and give a detailed description of the model in Section 3. Section 4 details the model’s parameter estima- tion. Experimental results are presented in Sec- tion 5, followed by discussion, future work, and conclusions. 2 Previous Work …

Content truncated to 500 words