Hierarchical Label Propagation and Discovery for Machine Generated Email

Abstract

Machine-generated documents such as email or dynamic web pages are single instantiations of a pre-defined structural template. As such, they can be viewed as a hierarchy of template and document specific content. This hierarchical template representation has several important advantages for document clustering and classification. First, templates capture common topics among the documents, while filtering out the potentially noisy variabilities such as personal information. Second, template representations scale far better than document representations since a single template captures numerous documents. Finally, since templates group together structurally similar documents, they can propagate properties between all the documents that match the template. In this paper, we use these advantages for document classification by formulating an efficient and effective hierarchical label propagation and discovery algorithm. The labels are propagated first over a template graph (constructed based on either term-based or topic-based similarities), and then to the matching documents. We evaluate the performance of the proposed algorithm using a large donated email corpus and show that the resulting template graph is significantly more compact than the corresponding document graph and the hierarchical label propagation is both efficient and effective in increasing the coverage of the baseline document classification algorithm. We demonstrate that the template label propagation achieves more than 91% precision and 93% recall, while increasing the label coverage by more than 11%.

Online template induction for machine-generated emails

“…For example, processing the cluster of emails belonging to a single template enables the ability to determine portions of these emails that are fixed across all instantiations. Resultant fixed text and even fixed images have been shown to be useful for classification tasks [30,37].…”

Learning from User Interactions in Personal Search via Attribute Parameterization

“…• Categories -This attribute type models the content using a fixed small set of topics or categories. It might be, for instance, an output of some textual classifier or clustering algorithm that runs over the private content [5,17,29].…”

“…• Structure -This attribute type models the content using its inherent structure, regardless of the content topic. For instance, for email corpora, a structure can be represented via structural templates [3,29]. For personal files, it can be represented, among other options, as a file type.…”

Template Induction over Unstructured Email Corpora

“…Kiritchenko and Matwin deal with the sparsity issue by using a co-training algorithm to build weak classifiers, then label unlabeled examples, and add the most confident predictions to the labeled set [13]. Somewhat similarly, Wendt et al utilize a graph-based label propagation algorithm to label unlabeled emails from a small set of labeled emails, but do so at the template level to improve scalability in very large mail provider systems [25].…”

“…While there is very little published work on using structural templates for processing commercial email data, templates have been used for annotating semantic types within the DOM trees of emails [27] and used in hierarchical classification of emails [25]. To our knowledge, no techniques have yet been published that propose template induction for plain text email content.…”