Email Classification Incorporating Social Networks and Thread Structure

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1336–1345 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1336 Email Classification Incorporating Social Networks and Thread Structure Sakahr Alkhereyf ∗, Owen Rambow † ∗Columbia University, † Elemental Cognition New York, NY, USA sakhar@cs.columbia.edu, owen.rambow@gmail.com Abstract Existing methods for different document classification tasks in the context of social networks typically only capture the semantics of texts, while ignoring the users who exchange the text and the network they form. However, some work has shown that incorporating the social network information in addition to information from language is effective for various NLP applications including sentiment analysis, inferring user attributes, and predicting inter-personal relations. In this paper, we present an empirical study of email classification into “Business” and “Personal” categories. We represent the email communication using various graph structures. As features, we use both the textual information from the email content and social network information from the communication graphs. We also model the thread structure for emails. We focus on detecting personal emails, and we evaluate our methods on two corpora, only one of which we train on. The experimental results reveal that incorporating social network information improves over the performance of an approach based on textual information only. The results also show that considering the thread structure of emails improves the performance further. Furthermore, our approach improves over a state-of-the-art baseline which uses node embeddings based on both lexical and social network information. Keywords: text classification, email classification, social networks, graph algorithms 1. Introduction There has been much work on using social networks to predict user characteristics. This work exploits homophily (for example, young people are more likely to communi- cate with other young people). In contrast, there has been far less work that uses the communication network (the network induced by conversations) to improve document classification of the communications themselves. This is a harder problem, since homophily is not relevant when characterizing the communications themselves: in various document classification tasks, the document category might not be directly inferred from the relationship of the partic- ipants when the same participants exchange different types of documents. For instance, the same people might ex- change both personal and business emails, or urgent and nonessential emails. In this paper, we study document clas- sification in the context of written conversations. As our task, we choose classification of email into personal or busi- ness emails. There are several reasons for this choice. 1. We are interested in how personal relationships affect communication, taking into account that the same pair of people may have multiple types of relationships. 2. The task we choose is relevant. Email remains a cru- cial communication medium for both individuals and organizations for both personal and business commu- nications. Kiritchenko and Matwin (2011) show that a typical user daily receives 40-50 emails. And despite the massive growth of other social media over the past decade, company email is still used for personal pur- …

Content truncated to 500 words