N2One: Identifying Coreference Object Among User Generated Content with Siamese Network
Lists
Only you see this
Abstract
With the boom of social network, various service is provided based on the user generated content (UGC). Some of these UGCs semantically point to the same entities in the real world. Recognizing the same real-world entity that is referred by various user contents is crucial for providing personalized service, such as recommender systems and user profiling. Recently, most recommender systems recommend content based on keywords detection or structured data. These approaches cannot fully grasp latent semantic of UGC, so it is difficult for them to discover the coreference entities behind UGCs. As a result, when there are less obvious keywords or preset user preference, the experience of such services are poor. In this paper, we formalize this problem as a N to 1 task, which means performing entity resolution among UGCs, and a Siamese Network is presented to deal with it, namely ‘N2One’. Extensive experiments are conducted on two commonly used datasets Yelp and IMDB to evaluate the effectiveness of the proposed model, and the results demonstrate its superiority.
BibTeX
@inproceedings{yuan2020n,
author = {Wei Yuan and Peng Wang and Mengyao Yuan and Yue Guo and Tieke He},
title = {{N2One: Identifying Coreference Object Among User Generated Content with Siamese Network}},
booktitle = {WISA},
year = {2020},
doi = {10.1007/978-3-030-60029-7\_26},
}Loading related content...