Please use this identifier to cite or link to this item:
|Title:||Deep Learning Approach for Automatic Email Classification|
|Keywords:||Computer Science & Engineering|
|Abstract:||E-mail usage is one of the mechanism of communication which has advanced very rapidly. It has grown from simple exchange of text to one of the most used means of communication in organisations for important information. Management of e-mails has thus become an important problem to be solved. The problem of email looping in very common, as people forward e- mails to other department and this leads to late replies of client emails. Classi- fication of mails based on department and employee roles can help eradicate this problem. Supervised classification provide an efficient solution for such problems. One key difficulty with these algorithms is that they require huge , often prohibitive, number of labelled training examples to learn accurately. Labelling requires experts to manually annotate the data which is cost as well as time expensive process. Our technique classifies unlabelled e-mails using deep learning. We use word2vec to build word vectors and utilise them to build our own document vectors. These are then clustered using various clustering algorithms in accor- dance with the taxonomy of classification required. Thus, this method semantically captures the meaning of e-mails without making use of any labelled information. We evaluate our system on a set of manually annotated data of publishing mail inbox with promising results. We also utilise supervised learning approaches to find the differences in both approaches.|
|Appears in Collections:||01. CSE|
Files in This Item:
|Deep Learning Approach for Automatic Email Classification.pdf||763.27 kB||Adobe PDF||View/Open Request a copy|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.