walk — a system for explo­ring and fin­ding valu­able content on the In­ter­net; at its core based on ethical guide­lines, addres­sing the need for more thought­ful design online.

Bachelor Thesis by Kai Wanschura and Johanna Wellnitz

concept / core areas

walk is a concept addres­sing three core chal­len­ges for buil­ding an envi­ron­ment on the In­ter­net, people can navigate com­for­tably—based on their inten­tions. It aims to provide value to indivi­duals and society in the long run.

01/03

How can people find value in unstruc­tured data?

02/03

How might we enable diver­sity in topics and sources?

03/03

How might we foster controlled explo­ration based on intention?

Outcome

Throughout our thesis, we deve­loped an approach, that puts people to the center of finding infor­mation and inspi­ration, not by building a communi­cation platform, but by reen­gineering search and brow­sing and moving it to a more social context.

Example for Application

challenge / How will people be able to find value in unstructured data?

Humans as an Instance for declaring qualitative Content

subjective value, transparency

Respect the receivers’ intention as part of the applied sharing model

Humans as an Instance for decla­ring quali­tative Content

Silent Sharing Model

Algor­ithms are suited to filter for para­meters of objective quality, but to judge the subjective value, emotions or preferences, humans—due to their unique characters—are more suited. As trust is not uncon­ditional, it requires the approval of the receiver to allow recom­menda­tion to appear in the result list—for example by assigning them to an interest.

Example of Use
close
/

relevance, importance, sorting

sort personal recom­menda­tions over algo­rithmic ones

Humans as an Instance for declaring qualitative Content

Prioratisation of personal Recommendations

The Prioritization of Recom­menda­tion Types describes the impor­tance of dif­ferent types with otherwise iden­tical relevance (in terms of content). For every kind of recom­menda­tion view (Stroll, Interest, Search Query, ...), we defined para­meters and their impor­tance by which the recom­meda­tions are sorted. Repe­titions of the same or very similar content or showing multiple contents from the same plat­form or source after another should be avoided by our system.

Example of Use
close
/

flexibility, reach

Integrate external chan­nels and plat­forms of communication

Humans as an Instance for declaring qualitative Content

Integration of social networks and communication channels

To increase flexibi­lity and reach of the system, it offers inte­gration of exte­rnal channels for commu­nica­tion, such as getting recom­menda­tions from a social net­work or a Messanger. Thereby we can avoid requiring repeated sharing on diff­ering channels. By running through the same veri­fica­tion process as the other recom­menda­tions, they are still filtered according to fitness to an interest or search.

Example of Use
close
/

challenge / How might we enable diversity in topics and sources?

Diversity of Content through cross­plat­form Linking

usable cross-site linking

Enable Chronological »playing« of Audio/Visual content

Diversity of Content through crossplatform Linking

Chronological »playing« of content

Some content on the Web is consumed passively. Like listening to Music or Podcasts. To make cross-site linking usable, we want to avoid being bound to one App or Plat­form for listen­ing to Music or watching a Videos and instead make it possible to play Audio- or Video elements from multiple sites in a row. We thereby competes with native auto- play-features of sites like YouTube or Netflix, which lock users in using endless autoplay. We do believe however, that autoplay should be limited to playlists, the users assembled themselves.

Example of Use
close
/

respect content providers

Provide Recom­menda­tions by Topic instead of Platform

Diversity of Content through crossplatform Linking

recommendations according to topic instead of platform

We do not want to restrict content providers, so we instead add a layer on top of the existing sites, that creates new connect­ions among plat­forms. Neither are those thereby entirely managed by the website creators, nor are they organized by biased plat­form users. Instead, their themes are the con­nect­ing element. The recom­menda­tion can origi­nate from a person, who has never visited the website the user is currently viewing and could thereby offer an entirely new perspective.

Example of Use
close
/

reduce irrelevant content

Require recom­menda­tions to be full pages instead of items of content

Diversity of Content through crossplatform Linking

Require recommendations to be full pages instead of single items

While versatility of suppor­ted content and file types on the Inter­net is one of its strengths, we limit sharing to entire web­sites instead of single images or senteces. Doing so is supposed to minimize sharing content merely because it is visually pleasing or quotable, following the idea of the »slow Internet«, where it is more desireable to have on impactful article than many small snippets, that provide little value. An exception is being able to recom­mend content with a specified scroll position.

Example of Use
close
/

challenge / How might we foster controlled exploration based on intention?

An alterable topical Filter Bubbles supporting the Users‘ Intention

self-control, different mind-sets

Leave control over the topical Filter Bubble to the User

An alterable topical Filter Bubbles supporting the Users‘ Intention

Filter Bubble Sorting

Usually, platforms either hardcode fixed para­meters for topical closeness directly into their algo­rithms, or they try to assume the user’s intention using keywords and context. That reversely has the effect of recom­menda­tions seeming either random or limiting. If the algorithm decides for the user to look only at content that is similar to what they usually favor, it takes away the oppor­tunity for people to branch out of their comfort zones and limits the possi­bility of exploring different mind-sets.

Example of Use
close
/

preferences, filtering

Enable Interest based long-term Browsing

An alterable topical Filter Bubbles supporting the Users‘ Intention

Interests

Interests represent the preferences of a user. While they are mostly aiding the long term use, people can also use them for outcome-oriented browsing, that is supposed to go over the course of a day because they also rely on past direct or passive shares, that fit the category. The interest also serves as an effec­tive filter for content that is shared by other users.

Example of Use
close
/

browsing and searching

Support varying Inten­tions and their changes over time

Alterable Topical Filter Bubbles supporting the Users’ Intentions

Support change of Intention

A system should be able to support the users inten­tions, which are clustered into three types: outcome oriented search, outcome oriented browsing, interest-based browsing and free browsing. The system also has to be aware, that the users can change or adjust their intention any time. They should be able to adjust their recom­menda­tions accord­ingly by using either filtering options or entirely changing the view.

Example of Use
close
/