Our paper "Alecsa: Attentive Learning for Email Categorization using Structural Aspects", with Azadeh Shakery, and Maryam S. Mirian, has been published at the Knowledge-Based Systems Journal. \o/
Due to the enormous volume of email data generated each day, email management has become a vital area of research. Among the email management tasks, automatic email categorization is one of the most interesting problems. However, the dynamic nature of email data makes the email categorization problem difficult to address for traditional machine learning approaches.
We propose ALECSA (Attentive Learning for Email Categorization using Structural Aspects) as an attentive learning approach for automatic email categorization. Alecsa aims to simulate the dynamic behavior of users while they attempt to categorize a new email. For this purpose, email categorization problem in Alecsa is cast to a decision-making problem, and an attention control framework is employed to dynamically choose a sequence of structural aspects of the email as the distinguishing factors for categorization. We have analytically evaluated the proposed approach on the Enron–Bekkerman datasets. The evaluation results indicate the unprecedented power of Alecsa toward modeling the dynamic essence of the email categorization problem in terms of effectiveness as well as efficiency.
The main challenge which is addressed in Alecsa is handling dynamicity of categorization criterion. To get a better intuition, consider a user that categorizes his/her emails in the following 10 folders:
- Call For Papers (Conferences/Journals)
- PhD Council
- SIGIR Mailinglist
- My Supervisor
- 2015 _ Summer
- Machine Learning Course _ Spring 2014
- Machine Learning Course _ Spring 2015
- Samira’s email _ 2014–2015
As can be observed from the name of the folders, the user trying to categorize emails in the above folders does not make the decision based on the content only. An initial guess would be that the emails in folders 1, 2, and 3 are categorized based on the topics of their content, while emails that are associated with labels of folders 4, 5, and 6, are grouped based on their senders, which is either a specific person or a particular group of people. Emails in folder 7 are categorized regarding the time they are received. Emails in folders 8 and 9 are also archived based on their topics as well as the time they are received. Folder 10 contains emails that are grouped together regarding their sender who is a specific person and also the time interval that they are received. This example reveals the type of dynamism existing in the decision-making process while users try to categorize incoming emails. This dynamism can be linked to the fact that each email has a structured content, and thereby can be viewed from different perspectives and aspects.
In Alecsa, the email categorization problem is represented as an attentive decision-making problem in which Active Decision Fusion Learning (ADFL) 1 is used to model the problem. The modeled problem is then solved using Reinforcement Learning. Alecsa not only is a powerful approach for imitating the dynamics of user behavior in email categorization effectively, but also tries to improve the efficiency of automatic email categorization. This is due to the fact that along with achieving a high accuracy, minimizing the cost is a key concern in the ADFL framework, which is employed as the attentive decision-making framework in Alecsa. Compared to the existing categorization methods, Alecsa can be regarded as a meta-classifier, which provides a policy for dynamic feature selection for each new instance. The selected features are then exploited for classifying the email. In fact, ADFL, which is the core of Alecsa, is a pre-processing layer on top of the traditional classifiers.
To know more about the ALECSA, please read our paper:
- Mostafa Dehghani, A. Shakery, M. S. Mirian. "ALECSA: Attentive Learning for Email Categorization using Structural Aspects". Knowledge-Based Systems Journal, Volume 98, 2016, pp. 44-54.
- M.S. Mirian, M.N. Ahmadabadi, B.N. Araabi, R.R. Siegwart, Learning active fusion of multiple experts’ decisions: An attention-based approach, Neural Comput., 23 (2) (2011), pp. 558–591