SMART Mail

A SMART Platform for Mail Management

Ricardo Raminhos

, Eduardo Coutinho

, Nuno Miranda

, Maria Barbas

, Paulo Branco

Teresa Gonçalves

and Gil Palma

VIATECLA SA, Estrada da Algazarra nº72, Almada, Portugal

Instituto Politécnico de Santarém, Escola Superior de Educação, Complexo Andaluz, Apart.131, Santarém, Portugal

Universidade de Évora, Largo dos Colegiais 2, Évora, Portugal

Keywords: SMART Mail, Analytics, Visualization, Exploratory Data Analysis.

Abstract: Email is a key communication format in a digital world, both for professional and/or personal usage.

Exchanged messages (both human and automatically generated) have reached such a volume that processing

them can be a great challenge for human users that try to do it on a daily basis and in an efficient manner. In

fact, a significant amount of their time is spent searching and getting context information (normally historic

information) in order to prepare a reply message or to take a decision/action, when compared to the actual

time required for writing a reply. Therefore, it is of utmost importance for this process to use both automatic

and semi-automatic mechanisms that allow to put email messages into context. Since context information is

given, not only by historical email messages but also inferred from the relationship between contacts and/or

organizations present in the messages, the existence of navigation mechanisms (and even exploration ones)

between contacts and entities associated to email messages, is of fundamental importance. This is the main

purpose of the SMART Mail prototype, which architecture, data visualization and exploration components

and AI algorithms, are presented throughout this paper.

1 INTRODUCTION

In the universe of email management solutions,

where the volume of data is continuously increasing,

the existence of platforms/solutions that allow the

treatment of messages in a graphical and intelligent

way (supervised or not – i.e. using automatisms) is

more necessity than an optional feature.

The email has been, throughout the years, the

most ubiquitous way of digital communication, even

taking into consideration the strong growth of instant

messaging applications. The previous existence of

an email account limits a large part of the online

presence of an individual, from social network

authentication, online shopping, access to web

portals, as well as most forms of online

communication.

According to the latest report “Email Market,

2015-2019” from “The Radicati Group, Inc.” from

July 2015 (The Radicati Group, s.d.), there are 2.6

billion email accounts (in 2015) and this number is

expected to grow above 2.9 billion by the end of

2019.

Table 1: Estimated market growth of email platforms for

2015-2019 (The Radicati Group, s.d.).

Year

Worldwide

Users (M)

% var

Worldwide

Market

Revenues

($M)

% var

2015 2 586 $13 607

2016 2 672 3% $19 353 42%

2017 2 760 3% $25 934 34%

2018 2 849 3% $32 592 26%

2019 2 943 3% $38 917 19%

Email solutions market will exceed $13.6 billion

in 2015, being expected to exceed $38.9 billion by

the end of 2019, which represents an average annual

growth of around 30% -

Table 1.

The volume of emails generated, including both

business and personal emails is estimated to be

around 205 billion per day during 2015, increasing

to above 246 billion per day by 2019 –

Table 2.

378

Raminhos, R., Coutinho, E., Miranda, N., Barbas, M., Branco, P., Gonçalves, T. and Palma, G.

SMART Mail - A SMART Platform for Mail Management.

In Proceedings of the 18th International Conference on Enterprise Information Systems (ICEIS 2016) - Volume 2, pages 378-387

ISBN: 978-989-758-187-8

Table 2: Estimated email traffic for 2015-2019 (The

Radicati Group, s.d.).

Year

Total

Worldwide

Emails Per

Day (B)

% var

Business

Emails Per

Day (B)

% var

2015 205.6 112.5

2016 215.3 5% 116.4 3%

2017 225.3 5% 120.4 3%

2018 235.6 5% 124.5 3%

2019 246.5 5% 128.8 3%

While the vast majority of email management

platforms are focused on the technical component of

the messages exchanged, all “intelligence” needed to

organize, prioritise and discover historical content

regarding email messages is currently depending on

the actual end user intelligence.

Although there are some solutions (please refer

to the State of the Art section) that are capable of

presenting the user some data/email metrics in a

more analytical way, this is still considered a

secondary feature.

VIATECLA understands the need for R&D

effort on the creation of instruments that offer

mechanisms to support the user, including a change

of paradigm focused on the presentation and

exploration of email data versus the current focus –

to send and receive email messages.

That is the purpose of the SMART Mail R&D

project (SMART Mail webpage, s.d.), a project

developed by VIATECLA (VIATECLA, 2015) and

supported by Universidade de Évora (Universidade

de Évora, 2015), Instituto Politécnico de Santarém

(Instituto Politécnico de Santarém, 2015) and GTE

Consultores (GTE Consultores, 2015), and co-

financed by QREN (Quadro de Referência

Estratégico Nacional) (National Strategic Reference

Framework (NSRF), s.d.).

2 THE SMART MAIL PROJECT

With the SMART Mail project, VIATECLA aims to

potentiate the importance of email, thus contributing

to an increase of productivity to its users, through a

software functional prototype that allows the current

standard/static system of email messages to have a

certain degree of intelligence.

Using SMART Mail it is intended for users to

reduce information overload, becoming information

more visual by representing email patterns as well as

propose priorities for processing email.

Therefore, SMART Mail goals are based on three

pillars:

• Enable a set of charts and exploration

controllers which applied to email information will

allow showing statistical data about email usage

(e.g. reception, sending and classification of emails),

through the application of statistical processes;

• Data normalization for both Organizations and

Contacts, which will make them more suitable to be

shared within the business context, not only its

specific information but also the meta-information

related to its relevance within the organizational

context;

• Application of knowledge and rule models,

directly to each email message data or to the

statistical numeric aggregated values from its history

and the definition of reading and reply messages’

priorities.

In this context, SMART Mail aims to be a very

valuable tool, improving the productivity of its users

and helping them in the decision making process.

The knowledge resulted from this investigation has

been materialized in a prototype for a generic

platform for email visualization and interaction.

This article presents the architecture, data

visualization and exploration components, Artificial

Intelligence and alert detection algorithms,

implemented in the SMART Mail. The article

“Email solutions – state-of-the-art and possible

evolutions” (Raminhos, et al., 2015), also written in

the context of the SMART Mail R&D activities,

presents the state of the art for email management

platforms in a greater detail.

3 STATE OF THE ART

Generally speaking, although email management

platforms have evolved during the execution period

of SMART Mail (2013-2015), no disruptive changes

have occurred, either at the internal data models,

interaction layer, graphic representation or AI

Algorithms. The main changes occurring in this area

have been driven by research and development

initiatives (and later creation of product) sponsored

by businesses to be applied to the corporative

environment, with very low contribution from the

academic sector.

Therefore, initiatives like Sidekick (Sidekick,

2015), Google Inbox (Inbox, 2015) and Verse from

IBM (Verse, 2015), end up contributing with

advances in the area of email management, through

the inclusion of graphical mechanisms and some

SMART Mail - A SMART Platform for Mail Management

379

intelligence on the comprehension and navigation of

emails and contacts. Each one of these initiatives is

currently a market leader, specifically:

• Under the form of plugin/add-on, using existing

email management platforms, and providing a layer

of intelligence on those platforms – Sidekick

approach;

• Reinventing the entire email experience,

natively integrated in the email client application –

Google Inbox approach;

• Under the form of an autonomous platform

based on the integration of email management

platforms and business analytics/documental

platforms – Verse (from IBM) approach;

On the other hand, Xobni plugin, one of the main

reference platforms in a recent past, having been

acquired by Yahoo! (Xobni Support Homepage,

s.d.), has since been withdrawn from the market.

The reason behind its acquisition (supposedly to be

integrated within Yahoo!’s email offer) is not clear

at the moment, and a practical outcome of any

integration with Yahoo!’s email has yet to be

observed.

From the state of the art analysis, it is observed

that there are not many tools that address the email

management and optimization of time spent using

email. The number of applications is even lower

when it comes to solutions available that are capable

of visually representing results and perform its

analysis.

From the technological point of view, there is a

large dependence of Microsoft Outlook under the

form of plugin creation for processing, visualization

and interaction with the information. In relation to

the approaches on the web visualization of email

indicators, these are also limited to a specific client

(Gmail in the case of Gmail Meter) or by using

internal resources (GetResponse (GetResponse

Homepage, s.d.) ) that do not allow to work with

services or clients of external emails.

Based on these conclusions, some trends/possible

evolutions foreseen on this domain are presented

1. Change of paradigm – from the current one,

which is focused on an isolated email message,

to an email message “in context”. Each message

should be put into historical context, in case it

exists (the history), especially with regard to

other Contacts and Organizations also present in

the conversation;

2. Integrated interface – following the suggested

change of paradigm, the way these

implementations are put in place cannot be seen

as something that is secondary/optional, but as a

new area that is always present and visible in the

user interface;

3. Higher intelligence – through reflection on which

artificial intelligence capabilities can be applied

to emails, either on the proposal of relevance

levels, or on the correlation of those levels;

4. Exploratory and Interactive – making graphic

controls that go beyond the graphical

representation of information, and enable the

user to have a certain degree of interaction

through: (i) filtering, (ii) search, (iii) definition of

temporal scope of the search, (iv) drilldown/roll

up mechanisms on the universe of data selected

to be analysed;

5. Collaborative – through the construction of

repositories (e.g. of Contacts, Organizations)

where information can be constructed and

consulted in a collaborative way in the context of

one entity (e.g. a company) or even in a

general/global way;

Considering the aforementioned points, it is

noted that there is urgency and concern with these

subjects. As it is an element that, for businesses and

organizations, involves many working hours of their

employees, so the possibility of optimizing that use

is entirely relevant, knowing that at any moment,

through metric analysis, whatever is being made

(whether well or not), and what can be corrected.

4 ARCHITECTURE

The Figure 1 shows a global vision for the SMART

Mail architecture. Following the client/server

paradigm, the solution is decomposed in four main

functional areas that are interrelated, namely (i)

SMART Mail Plugin, (ii) SMART Mail Core

Server, (iii) SMART Mail Catalogue and (iv)

SMART Mail Back office.

Being a prototype directly conditioned by a

standard email management platform that deals with

the email exchange component, Microsoft

Exchange/Outlook was selected as the most

adequate for the creation of a pilot capable of

demonstrating the capabilities of the developed

prototype.

Therefore, the MS Exchange Mail Server

component (external to the SMART Mail) is

responsible for the exchange of messages between

users, agenda management and message

synchronization with client components (e.g.

desktop, mobile). A plugin was developed for the

Microsoft Outlook Client – the SMART Mail

Plugin which deals with accessing email messages

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

380

locally present (and previously synchronized with

the MS Exchange Mail Server component)

processing them with the purpose of directly

obtaining data and metadata and indirectly present in

the messages through the Email Data Access

component.

Figure 1: General diagram for the SMART Mail platform

architecture.

The data access component is used by two

processing modules of the SMART Mail – Bulk

Processing and Continuous Processing. The Bulk

Processing component is used in the initial phase of

the SMART Mail Plugin and in case the user has a

high volume of historic emails that need to be

progressively processed in background. It is

intended that the historical volume be processed as

quick as possible, so this module guarantees that this

happens with transparency and in a non-intrusive

manner to the user’s daily operation – especially

during the peaks of information processing. Once

the historical information is processed, the

Continuous Processing module takes the Bulk

Processing’s place, by processing emails

incrementally as they are received. The data

obtained from the email messages – regardless of the

processing type – are kept in data optimized files

within the private area of the user’s file system (i.e.

Data Storage component), being the Data

Handling component in charge of the management

of all low-level data access.

Visually, two main graphical components exist –

Dashboard and Sidebar - integrated within the

SMART Mail Plugin. The Dashboard component

presents a global vision for email management

process – (i) presentation and navigation in emails

categorized as high priority items, (ii) navigation

and exploration of information from Organizations

and Contacts, (iii) possibility to create and update

Organizations and Contacts through the Content

Operations sub-component, (iv) overview of

Contact’s history and (v) intelligent interfaces for

search and identification of emails using multi-

filters.

On the other hand, the Sidebar component

always shows information in context according to

the specific email selected. As a result, it is possible

to visualize and navigate on Organizations and

Contacts present in the email, as well as to access a

set of metrics related to the same email, aggregated

by Organizations and Contacts or, in the case of a

conversation thread which extends in time, to

graphically represent its temporal iterations.

Both Dashboard and Sidebar include four main

components throughout the process, although with a

few variants according to the space available for the

presentation of the information, (i) presentation and

exploration charts (i.e. Chart component), (ii)

aggregated or non-aggregated numeric values – e.g.

by organization, contact, or temporarily (i.e. usage

Metrics component), (iii) application and extraction

of relevant words from the email message, and

possible relationships with other messages that share

the same set of contents (i.e. Keyword Navigation

component) and (v) presentation and management of

Alerts and suggestions (i.e. Alert/Recommendation

SMART Mail - A SMART Platform for Mail Management

381

Engine component).

The SMART Mail Core Server represents the

server layer responsible for the persistence of

Contacts and Organizations’ information, as well as

some general metrics related to the relationship

between a user and their specific Contacts and

Organizations. Due to privacy issues, no information

regarding specific email messages is kept in this

repository – this shall be responsibility of MS

Exchange Mail Server. Therefore, all accesses to

the server layer (based in human action – e.g. back

office authentication request, or based in software

programs/automatisms) is performed through the

Access Control API. Upon successful

authentication, access to the SMART Mail Rest

API is made available on the form of a web service

REST for CRUD Organizations management (i.e.

Organization Manager module), Metrics (i.e.

Metrics Manager module), Contacts (i.e. Contact

Manager module) and keyword extraction (i.e.

Relevant Content module). On a lower level, all the

data is kept in persistence in a Relational Database,

being the Raw Content Controller’s responsibility

to access and manipulate them.

Being possible to manipulate the information

from Organizations and Contacts through the

Dashboard and Sidebar components included in

the SMART Mail Plugin, the SMART Mail Back

office enables access to all contents created and their

management and administration in a web browser

environment. Apart from the consultation and

manipulation processes, the administrator has access

to the contents being able to decide if it should be

removed or whether the access should be or not be

restricted to some users. The Authentication

component is responsible for the user’s

authentication in the back office, where, if the

authentication is successful, a Web Interface is

made available. Through the Workflow Engine,

which implements a set of access policies to

contents, it is possible to remove/restrict the access

to these contents.

Finally, the SMART Mail Catalogue provides a

web access to the Organizations and Contacts

catalogue, from which information is used by the

Dashboard and Sidebar components. Thus, after a

successful Authentication by the user, it is possible

to search for Organizations and Contacts through the

Search module in natural language terms. Whilst the

Results Handler manages the results obtained in an

intelligent manner, the sub-modules Organization

Handler, Contact Handler and Metrics Handler

deal with the visualization of Organizations,

Contacts and metrics associated to the user that has

performed the search. The Navigation Controller

module guarantees the navigation process between

all entities involved.

5 SMART VISUAL

COMPONENTS – DASHBOARD

SMART Mail, as previously presented in the

“Architecture” section, has three main graphical

interfaces to communicate and interact with the end

user regarding statistics on Emails, Contacts and

Organizations, namely (i) SMART Dashboard, (ii)

SMART Sidebar e (iii) SMART Catalogue. Due to

the extent required for describing in detail each of

these interfaces, a decision was made to present on

extent only the SMART Dashboard interface in the

current article.

The SMART Dashboard area can be accessed via

a ribbon present at Microsoft Outlook top toolbar

section. While the Dashboard is activated only “on

request” by the user via ribbon, the Sidebar

graphical component is always visible and is located

on the right side pane of Microsoft Outlook

application (Figure 2).

Figure 2: SMART Dashboard – global vision while

integrated on Microsoft Outlook client.

The central Dashboard area comprises metrics,

graphics and global actions applied to all received

email and contacts associated to these emails and the

lateral Sidebar area displays content always in

context with the selected email (just one email),

presenting the related information (e.g. organization,

contacts, metrics, charts) accordingly.

In a general way Dashboard works as a “memory

panel” which never forgets its previous

content/usage, keeping historic behaviour for the

latest actions performed on its five main areas,

namely (i) SMART Email Overview, (ii) SMART

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

382

Contacts, (iii) SMART Alerts, (iv) SMART

Overview and (v) SMART Search.

Below the description and presentation of each

of these functional areas is presented.

5.1 SMART Email Overview

The Email Overview area (Figure 3) is the first on

display at the Dashboard component. It presents an

area chart, where the volume of email messages is

proportional to the area used by each of the email

classifiers, namely:

(i) Need Attention: Main focus for the user

regarding received messages, and not yet processed.

Since SMART Mail is used mainly in corporate

context this area is sub-divided in three: “Work”,

“Clients” and “Remainders”;

(ii) My Organizations: A dynamic area

changes according to the number of Organizations

the user works for. This separation is mainly

relevant in multi-company environments as

displayed in the example;

(iii) Family & Friends: Even if the SMART

Mail focus is mainly corporate it is not possible to

isolate familiar/friends message interchange during

work time, thus a specific classifier exists for that

purpose;

(iv) Promotions: Promotion / advertisement

emails which although secondary are not considered

SPAM messages;

(v) Newcomers: Reference to emails for

which there is not enough classification information,

thus being classified in this area (rather any of the

previous ones).

Figure 3: SMART Dashboard panel.

Besides the number of received emails messages

associated to any classifier area, the main related

Contacts and Organizations are also presented in the

case of the areas “Need Attention” and “Family &

Friends”. In the classifier “My Organizations”, each

Organization is identified by its name in the specific

area.

If the user selects any of the “Need Attention”

internal areas, he is positioned in its details

represented as a two dimension matrix with all

Organizations involved in sent email messages

(horizontally) and two columns with “New” and

“Follow up” classifications (vertically).

5.2 SMART Contacts and

Organizations

The “SMART Contacts and Organizations” area is

activated any time the user (i) selects a Contact or

Organization within any Dashboard area and intends

to see its detail, (ii) navigates specifically to the area

– using the left-side anchor button (Figure 3) or by

scrolling on the Dashboard panel.

Figure 4: SMART Dashboard – Organization listing.

While in the first case the detailed record is

presented immediately for the selected Contact or

Organization, in the second a list with Contacts and

Organizations available is shown.

As an example, Figure 4 presents a list with the

Organizations previously created, while pressing the

“People” tab will present, in an equivalent way, a

Contact listing.

While in listing mode each Organization/Contact

is characterized by its logo/photo, name and

business area/business role, access to its detail is

performed by clicking on the full extent on the

listing entry. Also, in this panel, it is possible to

create a new Organization/Contact (according to the

selected tab) or to identify a content using textual

search.

SMART Mail - A SMART Platform for Mail Management

383

Figure 5: SMART Dashboard – Contact detail.

Figure 5 presents the detail for a Contact where,

besides its associated information, a set of statistics

regarding previous email interaction history between

the user and the selected Contact is shown.

5.3 SMART Alerts

SMART Mail presents two types of alarms in the

Dashboard area.

The “Losing Contact” (Figure 6) alarm refers to

personal Contacts which, according to its previous

email exchange record, detects that the frequency of

exchange emails has been progressively decreasing

(e.g. in the last days, weeks, months) accordingly to

the message volume exchanged in the past.

Figure 6: SMART Dashboard – “losing contact” area.

Figure 7: SMART Dashboard – “follow up” area.

Thus, in order not to “lose contact”, an alert is issued

(initially in the form of a popup message) where the

user is invited to “reconnect” with the Contact (or

ignore this proposal). This feature can be of relevant

importance especially in commercial context or

when associated with networking actions.

Consulting the Dashboard the user can see which

alerts have been issued (and are active/not ignored),

with indication on the Contact and associated

Organization (if existing) and the period extension in

which no contact has occurred. The “Ignore” action

will cancel explicitly the alarm while the “Contact

Now” action will create a new email message for the

selected Contact.

The second type of alarm – “Follow up” (Figure

7), is always defined explicitly by the user. Through

the Sidebar, and having selected an email, it is

possible to associate a future moment/date to

perform a follow up action. Once this period is

attained a popup message is presented to the user

alerting that is time to act. On the other hand, the

user can, by accessing Dashboard, consult the

scheduled follow up actions, ordered

chronologically, and decide to cancel them or

provide an early reply. A special highlight is

presented for those alarms for which the initial

date/time to reply has been exceeded.

5.4 SMART Overview

The SMART Overview control (Figure 8) enables a

graphical display using an area representation, where

received email messages are mapped to the

associated Organization within a time frame. The

area size is proportional to the volume of received

emails within the defined time frame.

Figure 8: Graphical overview for the received emails

associated with Organizations in a time frame.

The time frame definition is performed on the

right most control where the user can select the full

extent of time (from the time he/she received the

first email message until current time) or define a

specific time interval.

The user, when selecting an area associated to an

Organization, visualizes a list for those emails

(ordered chronologically).

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

384

5.5 SMART Search

The last visual control present on the Dashboard is

the SMART Search which enables the user to

identify email messages according to the relations

between Organizations, Contacts and Conversation

Threads.

The usage of this control can be triggered in one

of two ways: (i) in context – when selecting an

Organization, Contact or Conversation Thread, the

control is invoked, making the selected value to be

automatically pre-selected in the dynamic search

filters or (ii) without context – directly accessing the

control via Dashboard where none of the dynamic

search filters present pre-selected values.

Thus, by defining values to the Organization,

Contact or Conversation Thread filters, associated

emails are dynamically filtered. Associated to each

filter area, there’s the possibility to perform a textual

search in order to faster determine the filter value to

apply. Further, each filter area can be

expanded/collapsed in order to gain further space

area when the number of search results/available

filter options is high.

When the user selects a specific email the

Dashboard area is hidden and the email is selected in

the Microsoft Outlook client, being this information

completed by Sidebar that presents the associated

metadata regarding the email.

6 SMART RELEVANT CONTENT

The recognition of keywords is a subarea of

knowledge from information extraction that intends

to identify and classify relevant elements on the text

– e.g. which mention pre-defined categories as

people’s names, organizations, locations, time

expressions, monetary values and percentages. For

those detected keywords (retrieved from the email’s

corpora) which are found to be transversal to

multiple / different users, can be proposed as a

dynamic classification attribute.

There are several approaches to the problem

(Nadeau & Sekine, 2007), from systems that use

rules defined by people (Aberdeen, et al., 1995) to

systems based in automatic learning techniques

(Mitchell, 1997) through the use of classification

algorithms as decision trees (Baluja, Mittal, &

Sukthankar, 2000) or support vector machines

(Takeuchi & Collier, 2002).

The process implemented for SMART Mail for

the automatic classification uses a generic inference

process (commonly designated by “learning”) from

which a classifier is automatically built.

During the learning phase, several examples of

email text with keywords manually classified are

provided, from which the algorithm concludes the

features that define each one of those entities. This

conclusion is reached using a set of attributes that

characterize each word that can be found in the

email text. As a result, a model of knowledge that

summarizes the rules of identification of keywords is

obtained.

Later in the classifying phase, the model

obtained is used to identify entities on new email

text.

There are two main phases during the learning

and classifying process. One is responsible for the

extraction of spelling and morphological attributes

from each word; the other one is responsible for

identifying and classifying keywords within the text.

The morphological attributes are obtained from

SVMTool (Gimenez & Marquez, 2004), a word

labeller that together with the spelling attributes is

used in a second stage for the construction of the

entities classifying model.

The SVMTool is a sequential label generator

based on support vector machines (Cristianini &

Shawe-Taylor, 2000) (the same automatic

classifying algorithm already mentioned above). As

such, for it to be used, it is necessary to have a

model for the language in which the email messages

are written in. As the Portuguese model is not

included in the available language models (i.e.

Spanish, Catalan and English) it has been necessary

to develop it.

On an operational level, the two stages have a

component in common that is based on automatic

classification algorithms. The entity recognizer, like

the labeller, uses support vector machines.

The choice of this algorithm was due to a set of

tests made initially in order to assess the

performance of some algorithms recommended by

the scientific and academic literature as the most

suitable (Quinlan, 1993), (Zhang, 2004), (Caruana &

Niculescu-Mizil, 2006), (Keerthi, Shevade,

Bhattacharyya, & Murthy, 2001), (Witten & Frank,

2005).

Table 3: Spelling Attributes.

Spelling Attributes

! ? Uppercase Capitalization

( ) Lowercase Alphanumeric

“ ‘ Unique character Numeric

; : Upper and lowercase Letters

+ - Initial on its own

« » hyphenated words

SMART Mail - A SMART Platform for Mail Management

385

Table 4: Morphological Attributes.

Morphological Attributes

Determiners Common noun Verbs

Proper noun Contraction (determiner) Prefix

Conjunction Contraction (adverb) Adverbs

Adjectives Contraction (pronoun) Pronoun

Interjection Punctuation Preposition

Following, the attributes used in the

classification of words as point of entry for the

keyword extractor, are described.

The spelling attributes, in a total of 25 binary

attributes, are extracted from the words orthographic

characteristics.

Table 3 lists all attributes considered.

The morphological attributes, in a total of 15 binary

attributes, indicate the morpho-syntatic class of each

word in the text. Table 4 lists all attributes

considered.

To improve the efficiency of the process and to

add information from the context of the word

analysed, a “window of context” is created, which

consists in joining the describing attributes of the

neighbour words (before and after the word that is

being analysed).

As an example, for a window of context of size

5, the analysed word is present at a central position

and the two previous and two posterior words are

also considered in the classification process, as

depicted bellow.

word - 2 word - 1

analysed

word

word + 1 word + 2

Therefore, the base pillars for the operation of the

keyword extraction mechanism have been listed: the

window of context, extraction of spelling attributes,

extraction of morphological attributes via support

vector machines and finally, the final classification

of keywords using support vector machines, and

consuming all available resources obtained from the

stages previously mentioned.

7 EVALUATION AND FUTURE

WORK

For the evaluation of the SMART Mail prototype

two validation test pilots have been performed – one

conceptual and another functional. The SMART

Mail conceptual pilot was performed by creating a

test case simulation/ scenario /story using a set of

images (resulting from the design mockups) for the

different interface screens. Using Microsoft

Sketchflow technology (Corporation, 2015) it was

possible to create interaction areas in the design

mockup images and simulate a navigation flow,

making it a dynamic and living experience. This way

it is possible to attain a conceptual prototype (even if

static and image based) for testing the concept,

organization, perception and navigation flow.

The test case environment took place during 3

days (with 1 week interval between each) where new

prototype versions were presented according to the

collected feedback in the previous version. In the

first session both project, concept, objectives and the

simulation test scenario itself were presented to the

tester team involved in the process.

The main feedback recoiled from these sessions

focused the following points:

• Layout issues – both comprehension related and

ergonomics – i.e. positioning, size, colours and

graphical elements;

• Discussion, proposal and conception of the

“Losing Contact” area (integrated in the Dashboard

component);

• Inclusion of social networks information

associated to both Contacts and Organizations;

• Proposal on removal the “Attachments” visual

control (integrated in the Sidebar component) which

would aggregate links for the attachment files

present in an email message.

The functional pilot testing had the duration of

one month and was performed on a continuous way

using the SMART Mail prototype software which

was available at the time. Also, during this period,

feedback (of a more functional nature) was

collected, namely:

• Layout issues – both on understanding and

ergonomics – i.e. positioning, size, colours and

graphical elements;

• Issue identification and solving;

• Discussion, proposal and conception of the

“SMART Search” area (integrated in the Dashboard

component);

• Proposal and integration of “Take me there”

functionalities which would potentiate

georeferenced information (mostly Organization

addresses) integrated with external map applications

for both Desktop, tablet and mobile environments;

• Proposal and inclusion on the “SMART

Catalogue” interface of high level statistics

regarding the communication between the end user

and other Contact or Organization;

• Inclusion of a “Clients” section in the

Dashboard’s “Need Attention” area (previously the

proposal was focused on “Work” and “Reminders”

sections).

In a complementary way, also during the

functional prototype testing, some ideas/feedback

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

386

regarding the possible commercial positioning of

SMART Mail were addressed/discussed:

• Development of a 100% web interface which

would not be dependent on a Desktop based email

client (such as Microsoft Outlook which was used as

a first testing environment);

• Possible SaaS business model (“Software as a

Service”) where the end user would refund the

SMART Mail promotor according to the actual

platform usage;

• Possible unification of both Contact and

Organization repositories at a higher macro level

which could be used “openly” by different

independent entities;

During the operational prototype testing, in order

to access the usefulness of SMART Mail, a set of

tasks (in the format of a script) was provided to user

subjects. Group A performed tasks supported with

SMART Mail while Group B used a standard

desktop email client. Users using SMART Mail, and

after the initial learning curve, proved to be 5% -

15% more productive (i.e. time per task) than Group

B users.

Both pilots strongly contributed for testing,

validation and evolution of SMART Mail.

Being available at the current development stage

a first functional version of SMART Mail, future

work will mainly be directed to the promotion and

support of “live” clients in real environments.

According to the feedback collected from both users

and enterprises, future SMART Mail clients, a

roadmap (both technological and business oriented)

will be defined in order to contribute to the platform

further refinement and evolution.

REFERENCES

Aberdeen, J., Burger, J., Day, D., Hirschman, L.,

Robinson, P., & Vilain, M. (1995). {MITRE}:

description of the Alembic system used for {MUC-6}.

MUC6 '95: Proceedings of the 6th conference on

Message understanding (pp. 141-155). Morristown,

NJ, USA: Association for Computational Linguistics.

Baluja, S., Mittal, V., & Sukthankar, R. (2000). Applying

Machine Learning For High Performance Named-

Entity Extraction. In Proceedings of the Conference of

the Pacific Association for Computational Linguistics,

(pp. 365-378).

Caruana, R., & Niculescu-Mizil, A. (2006). An empirical

comparison of supervised learning algorithms. ICML

'06: Proceedings of the 23rd international conference

on Machine learning (pp. 161-168). New York, NY,

USA: ACM.

Corporation, M. (2015). Sketch Flow. Retrieved from

Cristianini, N., & Shawe-Taylor, J. (2000). {An

Introduction to Support Vector Machines}. Cambridge

University Press.

GetResponse Homepage. (n.d.). Retrieved 2015, from

GetResponse: http://www.getresponse.com/

Gimenez, J., & Marquez, L. (2004). {SVMTool: A general

POS tagger generator based on Support Vector

Machines}. Proceedings of the 4th LREC.

GTE Consultores. (2015). Retrieved from

http://www.gte.pt/

Inbox. (2015). Retrieved from https://inbox.google.com.

Instituto Politécnico de Santarém. (2015). Retrieved from

http://www.ipsantarem.pt/

Keerthi, S., Shevade, S., Bhattacharyya, C., & Murthy, K.

(2001). {Improvements to Platt's SMO Algorithm for

SVM Classifier Design}. Neural Comput., 13(3), 637-

649.

Mitchell, T. (1997). Machine Learning. McGraw-Hill.

Nadeau, D., & Sekine, S. (2007). A survey of named

entity recognition and classification. Linguisticae

Investigationes, 30(1), 3-26.

National Strategic Reference Framework (NSRF). (n.d.).

Retrieved 2015, from http://www.qren.pt/np4/home.

Quinlan, R. (1993). C4.5: Programs for Machine

Learning. San Mateo, US: Morgan Kaufmann.

Raminhos, R., Coutinhho, E., Miranda, N., Barbas, M.,

Branco, P., Gonçalves, T., & Palma, G. (2015). Email

solutions – state-of-the-art and possible evolutions.

Sidekick. (2015). Retrieved from http://www.

getsidekick.com/

SMART Mail webpage. (n.d.). Retrieved 2015, from

http://www.viatecla.com/inovacao/smart-mail.

Takeuchi, K., & Collier, N. (2002). Use of support vector

machines in extended named entity recognition.

COLING-02: proceedings of the 6th conference on

Natural language learning (pp. 1-7). Morristown, NJ,

USA: Association for Computational Linguistics.

The Radicati Group, I. (n.d.). Retrieved Julho 2015, from

http://www.radicati.com/wp/wp-

content/uploads/2015/07/Email-Market-2015-2019-

Executive-Summary.pdf.

Universidade de Évora. (2015). Retrieved from

http://www.uevora.pt/

Verse. (2015). (IBM) Retrieved from http://www

.ibm.com/social-business/us/en/newway/

VIATECLA. (2015). Retrieved from http://www.

viatecla.com.

Witten, I., & Frank, E. (2005). Data Mining: Practical

machine learning tools and techniques (2nd ed.). San

Francisco, US: Morgan Kaufmann.

Xobni Support Homepage. (n.d.). Retrieved 2015, from

Xobni Support: https://support.xobni.com/home.

Zhang, H. (2004). The optimality of naive Bayes.

Proceedings of the 17th International FLAIRS

conference. AAAI Press.

SMART Mail - A SMART Platform for Mail Management

387