Application scenario

Introduction

Nowadays many people struggle with the information overload caused by all kinds of modern communication. Several principles have been presented to cope with this huge activity stream. First there is the issue of presenting information in a concise, hierarchical and yet informative way. To this end, filtering, clustering and ranking are essential. Another issue involves processing messages. Some should be removed immediately, others filed for later. In this course we are given the task to develop a product supporting this functionality in a human centered way.


Concept

The approach we present is similar to the one used in classic mail clients or feed readers. It uses the notion of clusters as an alternative to explicit folders created by the users. Clusters are user-defined, that is, the user specifies a pattern to relate content to one or more clusters. This pattern can be the source (e.g. YouTube), a set of keyword strings, the subject or title etc.

As with ordinary clients, the user can choose to show only messages of a specific cluster. Yet, the standard representation (‘general view’) shows a mixture of messages from various clusters. They get an overall rating, which allows to order them in descending order of importance. This allows to deal with the most urgent messages first.

The rating algorithm uses  initially user-defined cluster multipliers. That is, some clusters will be more important than others. The user’s favourite YouTube channel might be fun, it has probably lower importance than mails from business partners.

Within clusters the rating is defined in several ways.
A first one is time-based. As in classic approach, the older a messages gets, the lower its importance. Yet there is also a contrary measure for deadlines, the making of which will be discussed later. When a deadline comes close, the importance of the relevant message grows exponentially.  
A second one is string-based. The patterns resemble the clustering criteria. A user can for example define a set of strings the occurrence of which raises the rating. String could be found in the subject or description of the message, but could also be the sender, e.g. a feed from the company’s website.
A third way is the actual use of the system. Both the cluster multiplier and the in-cluster score are updated based on the messages the user reads and how long these are active. Similar messages (e.g. same source, same content profile,…) get a higher score in the future. The user can also explicitly mark a message as ‘important’, eliminating the need to define an explicit pattern.
The in-cluster scores are combined with the cluster multipliers to calculate an overall score. The patterns described here can also be used to create a blacklist e.g. to deal with spam. In some cases, it might be easier for the user to define what he doesn’t want to receive, instead of creating a whole network of filters to describe what he does want to get.


User scenario
Imagine Will, ceo of a busy company. Furthermore, Will likes to know what happens around him. To stay in touch with the world, he subscribed to a lot of newsletters from all kinds of news sources - videos, magazines, podcasts - both locally and international. Will is a very enthusiastic singer as well: every week, he meets his fellow choir members for rehearsals. Especially when the annual concerts come close, a lot of things have to be organised and communicated. On top of all that, Will's a sportsman. Once a month, he goes canoeing in the local stream with friends. Every summer, he likes to hike in Switserland as well. Unfortunately, all of his friends are very busy people, so it's always a mess to pick a date and deal with the practicalities.

 At times, all this information is really overwhelming him. He desperately needs a tool that can prioritize all his sources and allows for dealing with this overload in a convenient and efficient way.
Rankr is such a system. 

When Will logs on to the program, he sees all the clusters he has created on the left, with on top the ‘all’ cluster which consists of all messages. He could for example distinguish between updates from sports friends, job, his favourite YouTube channel, news feeds, choir mails etc. Next to the clusters, a stream of messages of the selected cluster is shown. In the beginning the main cluster is chosen, so the stream consists of all most important messages descending in importance. Furthermore, at any time the most important messages of four important clusters are shown at the right. The most urgent coming deadlines are shown as well, with the possibility to expand to a full-blown calendar. Will really likes this section because it is in fact a more general 'app section'. He can plug in any app he wants, the deadline being only one of the possible choices.

When Will clicks on a message in the stream, it will expand to the right.
When viewing a message, Will can choose to ‘file it for later’. In this case, a deadline has to be specified manually, which influences the rating as described earlier. Alternatively, he can select one or more messages and choose an action to perform from a dropdown list. He could for example remove several messages, saving them for a specified time in a thrash folder after which they are removed. Or he could create a new cluster with the selected messages. A basic set-up is derived from similarities in the messages. Of course, Will probably wants to fine-tune the cluster criteria later manually.