Generating Supplementary Travel Guides from Social Media

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1670–1681, Dublin, Ireland, August 23-29 2014. Generating Supplementary Travel Guides from Social Media Liu Yang1,2, Jing Jiang2,∗, Lifu Huang1,2, Minghui Qiu2, Lizi Liao2,3 1Peking University / Beijing, China, 100871 2Singapore Management University / Singapore, Singapore, 178902 3Beijing Institute of Technology / Beijing, China, 100081 yang.liu@pku.edu.cn, jingjiang@smu.edu.sg {warrior.fu, minghuiqiu, liaolizi.llz}@gmail.com Abstract In this paper we study how to summarize travel-related information in forum threads to gener- ate supplementary travel guides. Such summaries presumably can provide additional and more up-to-date information to tourists. Existing multi-document summarization methods have limita- tions for this task because (1) they do not generate structured summaries but travel guides usually follow a certain template, and (2) they do not put emphasis on named entities but travel guides often recommend points of interest to travelers. To overcome these limitations, we propose to use a latent variable model to align forum threads with the section structure of well-written travel guides. The model also assigns section labels to named entities in forum threads. We then propose to modify an ILP-based summarization method to generate section-specific summaries. Evaluation on threads from Yahoo! Answers shows that our proposed method is able to generate better summaries compared with a number of baselines based on ROUGE scores and coverage of named entities. 1 Introduction Online forums and community question answering (CQA) sites contain much useful information from ordinary users, such as their personal experience, opinions, suggestions and recommendations. Extract- ing and summarizing information from these rich information sources has a wide range of applications. In this work, we study how to tap into user-generated content in forums such as Yahoo! Answers to generate supplementary city travel guides. Travel guides published by well-known publishers such as Lonely Planet are written by a small number of authors based on their travel experience. Presumably if we could summarize the large amount of information given by ordinary users about a city, such a summary could supplement the official travel guide and cover more up-to-date information. However, social media content is diverse and noisy because it is contributed by many different au- thors. Directly applying existing multi-document summarization methods to forum and CQA threads may not produce good travel guides for the following reasons: (1) Summaries produced by standard summarization methods are not structured, but travel guides usually follow a template structure. (2) Travel guides put much emphasis on points of interest, which are usually location entities, but standard text summarization methods are not entity-oriented. To illustrate our points, in Table 1 we show (i) the overall structure of a travel guide for Sydney from Lonely Planet, (ii) an excerpt from a summary generated by a state-of-the-art ILP-based summarization method (Gillick and Favre, 2009) from a set of threads related to Sydney, and (iii) excerpts of a structured summary generated by our proposed method. The comparison shows that the summary generated by the standard ILP method mixes information on different topics together and does not …

Content truncated to 500 words