Deep Reinforcement Learning-based Text Anonymization against Private-Attribute Inference

Deep Reinforcement Learning-based Text Anonymization against Private-Attribute Inference Ahmadreza Mosallanezhad Ghazaleh Beigi Computer Science and Engineering Arizona State University {amosalla, gbeigi, huanliu}@asu.edu Huan Liu Abstract User-generated textual data is rich in con- tent and has been used in many user be- havioral modeling tasks. However, it could also leak user private-attribute information that they may not want to disclose such as age and location. User’s privacy concerns man- date data publishers to protect privacy. One effective way is to anonymize the textual data. In this paper, we study the problem of textual data anonymization and propose a novel Re- inforcement Learning-based Text Anonymi- zor, RLTA, which addresses the problem of private-attribute leakage while preserving the utility of textual data. Our approach first ex- tracts a latent representation of the original text w.r.t. a given task, then leverages deep rein- forcement learning to automatically learn an optimal strategy for manipulating text repre- sentations w.r.t. the received privacy and util- ity feedback. Experiments show the effective- ness of this approach in terms of preserving both privacy and utility. 1 Introduction Social media users generate a tremendous amount of data such as profile information, network con- nections and online reviews and posts. Online vendors use this data to understand users prefer- ences and further predict their future needs. How- ever, user-generated data is rich in content and ma- licious attackers can infer users’ sensitive infor- mation. AOL search data leak in 2006 is an ex- ample of privacy breaches which results in users re-identification according to the published AOL search logs and queries (Pass et al., 2006). There- fore, these privacy concerns mandate that data be anonymized before publishing. Recent research has shown that textual data alone may contain suf- ficient information about users’ private-attributes that they do not want to disclose such as age, gen- der, location, political views and sexual orienta- tion (Mukherjee and Liu, 2010; Volkova et al., 2015). Little attention has been paid to protect users textual information (Li et al., 2018; Zhang et al., 2018; Anandan et al., 2012; Saygin et al., 2006). Anonymizing textual information comes at the cost of losing utility of data for future applications. Some existing work shows the degraded quality of textual information (Anandan et al., 2012; Zhang et al., 2018; Saygin et al., 2006). Another re- lated problem setting is when the latent represen- tation of the user generated texts is shared for different tasks. It is very common to use recur- rent neural networks to create a representation of user generated text to use for different machine learning tasks. Hitaj el al. show text represen- tations can leak users’ private information such as location (Hitaj et al., 2017). This work aims to anonymize users’ textual information against private-attribute inference attacks. Adversarial learning is the state-of-the-art ap- proach for creating a privacy preserving text em- bedding (Li et al., 2018; Coavoux et al., 2018). In these methods, a model is trained to create a text embedding, but we cannot control the privacy- …

Content truncated to 500 words