Collective Tweet Wikification based on Semi-supervised Graph Regularization

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 380–390, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Collective Tweet Wikification based on Semi-supervised Graph Regularization Hongzhao Huang1, Yunbo Cao2, Xiaojiang Huang2, Heng Ji1, Chin-Yew Lin2 1Computer Science Department, Rensselaer Polytechnic Institute, Troy, NY 12180, USA 2Microsoft Research Asia, Beijing 100080, P.R.China {huangh9,jih}@rpi.edu1, {yunbo.cao,xiaojih,cyl}@microsoft.com2 Abstract Wikification for tweets aims to automat- ically identify each concept mention in a tweet and link it to a concept referent in a knowledge base (e.g., Wikipedia). Due to the shortness of a tweet, a collective inference model incorporating global ev- idence from multiple mentions and con- cepts is more appropriate than a non- collecitve approach which links each men- tion at a time. In addition, it is chal- lenging to generate sufficient high quality labeled data for supervised models with low cost. To tackle these challenges, we propose a novel semi-supervised graph regularization model to incorporate both local and global evidence from multi- ple tweets through three fine-grained re- lations. In order to identify semantically- related mentions for collective inference, we detect meta path-based semantic rela- tions through social networks. Compared to the state-of-the-art supervised model trained from 100% labeled data, our pro- posed approach achieves comparable per- formance with 31% labeled data and ob- tains 5% absolute F1 gain with 50% la- beled data. 1 Introduction With millions of tweets posted daily, Twitter en- ables both individuals and organizations to dis- seminate information, from current affairs to breaking news in a timely fashion. In this work, we study the wikification (Disambiguation to Wikipedia) task (Mihalcea and Csomai, 2007) for tweets, which aims to automatically identify each concept mention in a tweet, and link it to a concept referent in a knowledge base (KB) (e.g., Wikipedia). For example, as shown in Figure 1, Hawks is an identified mention, and its correct ref- erent concept in Wikipedia is Atlanta Hawks. An end-to-end wikification system needs to solve two sub-problems: (i) concept mention detection, (ii) concept mention disambiguation. Wikification is a particularly useful task for short messages such as tweets because it allows a reader to easily grasp the related topics and en- riched information from the KB. From a system- to-system perspective, wikification has demon- strated its usefulness in a variety of applica- tions, including coreference resolution (Ratinov and Roth, 2012) and classification (Vitale et al., 2012). Sufficient labeled data is crucial for supervised models. However, manual wikification annota- tion for short documents is challenging and time- consuming (Cassidy et al., 2012). The challenges are: (i) unlinkability, a valid concept may not ex- ist in the KB. (ii) ambiguity, it is impossible to determine the correct concept due to the dearth of information within a single tweet or multiple correct answer. For instance, it would be diffi- cult to determine the correct referent concept for “Gators” in t1 in Figure 1. Linking “UCONN” in t3 to University of Connecticut may also be ac- ceptable since Connecticut Huskies is …

Content truncated to 500 words