Enriching Cold Start Personalized Language Model Using Social Network Information

Computational Linguistics and Chinese Language Processing Vol. 21, No. 1, June 2016, pp. 51-68 51  The Association for Computational Linguistics and Chinese Language Processing Enriching Cold Start Personalized Language Model Using Social Network Information1 Yu-Yang Huang, Rui Yan, Tsung-Ting Kuo, and Shou-De Lin Abstract Personalized language models are useful in many applications, such as personalized search and personalized recommendation. Nevertheless, it is challenging to build a personalized language model for cold start users, in which the size of the training corpus of those users is too small to create a reasonably accurate and representative model. We introduce a generalized framework to enrich the personalized language models for cold start users. The cold start problem is solved with content written by friends on social network services. Our framework consists of a mixture language model, whose mixture weights are estimated with a factor graph. The factor graph is used to incorporate prior knowledge and heuristics to identify the most appropriate weights. The intrinsic and extrinsic experiments show significant improvement on cold start users. Keywords: Language Model, Factor Graph, Social Network Analysis, Smoothing, Cold-Start Problem. 1. Introduction Personalized language models on social network services are useful in many aspects (Xue et al., 2009; Wen et al., 2012; Clements, 2007). For instance, if the authorship of a document is in doubt, a language model may be used as a generative model to identify it. In this sense, a language model serves as a proxy of one’s writing style. Furthermore, personalized language models can improve the quality of information retrieval and content-based recommendation 1 A preliminary version of this paper appears as a short paper in Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Short Papers), pages 611–617, Baltimore, Maryland, USA, June 23-25 2014.  Graduate Institute of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan. E-mail: {r02922050, d97944007, sdlin}@csie.ntu.edu.tw  Computer and Information Science Department, University of Pennsylvania, Philadelphia, U.S.A. E-mail: ruiyan@seas.upenn.edu 52 Yu-Yang Huang et al. systems, where documents or topics can be recommended based on the generative probabilities. It is challenging, however, to build a personalized language model for users who just entered the system since the content posted by these users is insufficient to characterize them. These users are referred to as “cold start” users. Since the popularity of a system largely depends on whether new users will continue to stick to the system, it is even more critical to generate good recommendation results for new users. Therefore, this paper focuses on how to obtain a better personalized language model for cold start users. To achieve the aforementioned purpose, the content written by friends on a social media site is exploited. It can be a reply or a post written by friends on sites like Facebook or Twitter. Here, the hypothesis is that friends, who usually share common interests, tend to discuss similar topics and use similar words than non-friends. In other words, we believe that a cold start user’s language model can be …

Content truncated to 500 words