Transferring User Interests Across Websites with Unstructured Text for Cold-Start Recommendation

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 805–814, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Transferring User Interests Across Websites with Unstructured Text for Cold-Start Recommendation Yu-Yang Huang and Shou-De Lin Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan {r02922050,sdlin}@csie.ntu.edu.tw Abstract In this work, we investigate the possibility of cross-website transfer learning for tackling the cold-start problem. To address the cold- start issues commonly present in a collabora- tive �ltering (CF) system, most existing cross- domain CF models require auxiliary rating data from another domain; nevertheless, un- der the cross-website scenario, such data is of- ten unobtainable. Therefore, we propose the nearest-neighbor transfer matrix factorization (NT-MF) model, where a topic model is ap- plied to the unstructured user-generated con- tent in the source domain, and the similarity between users in the latent topic space is uti- lized to guide the target-domain CF model. Speci�cally, the latent factors of the nearest- neighbors are regarded as a set of pseudo ob- servations, which can be used to estimate the unknown parameters in the model. Improve- ment over previous methods, especially for the cold-start users, is demonstrated with experi- ments on a real-world cross-website dataset. 1 Introduction While collaborative �ltering (CF) approaches are one of the most successful methods for building rec- ommender systems, their performance deteriorates dramatically under cold-start situations. That is, low prediction accuracy is observed for users/items with very few ratings. Content-based recommender sys- tems may also suffer from the cold-start problem. For instance, content-based nearest-neighbor mod- els (Pazzani and Billsus, 2007) might not be as ef- fective if some users contain too few information to generate a meaningful set of neighbors. Two types of solutions have been proposed to ad- dress the cold-start problem. The �rst is to cre- ate hybrid recommendation models that impose a content-based model on a CF model to enrich the information for users/items with sparse rating pro- �les (Burke, 2002; Burke, 2007). The second is to transfer the information from auxiliary domains as a remedy to the cold-start individuals (Deng et al., 2015). This paper aims at bringing a marriage be- tween these two types of strategies. Although transfer learning gradually gains pop- ularity in handling the cold-start issue (Roy et al., 2012), most of them assume a homogeneous model where observations in both domains are of the same type. That is, to transfer knowledge to a rating- based/text-based recommender system, the source system must also be rating-based/text-based. Some earlier works even require the ratings from both do- mains to be in the same format (Li et al., 2009), or assume speci�c structured text, such as user- provided tags (Shi et al., 2011; Deng et al., 2015). In this work, by contrast, no source-domain ratings are available and unstructured user-generated content is treated as the auxiliary data. We propose a hetero- geneous transfer learning framework to utilize un- structured auxiliary text for a better target-domain CF model. As there is no …

Content truncated to 500 words