Wikimedia Mobile engineering/Contributory Funnel

Introduction

A useful metaphor for addressing the decline in Wikipedia editors is the conversion funnel. Since the goal is to convert more readers to editors, and to engage users to become productive contributors, the funnel in this case is focused on contributions.

Here is an early draft:

Mobile as a driver of contributions

As evident in the diagram above, mobile is a growing part of Wikipedia readership. There are several reasons to think of mobile as a key channel for increasing contributions:

mobile is essentially a new frontier for Wikipedia and can innovate more freely

mobile apps are particularly flexible ways of "sandboxing" and testing new ideas

mobile is the most prevalent method of net access in many parts of the world

mobile delivery can target device types in specific ways

mobile usage is closely aligned with short forms of messaging, and there are various tasks that can leverage this usage pattern

In terms of the types of tasks suitable for mobile, there is quite a range. The traditional Wikipedia world is divided into editors and non-editors. But in the emerging micro-focused world, there are many forms of indirect editing that still contribute to article quality and creation.

For example, the simple act of providing input about an article, in terms of its content, style or references, can help editors improve articles. But the concern from the perspective of established editors is that most such comments would be useless or even malicious. Therefore, some mechanisms for encouraging useful feedback and harvesting good feedback are required.

There are also many types of so-called editing that are really about review and curation. Some of these tasks are ideal for mobile usage patterns. For example, waiting in line at a bank, or sitting on a train, a user could flick through a list of revisions or new images and make simple indications of what to do next.

The power law distribution

In an ideal world, all contributions would operate in a balanced ecosystem where tasks are shared amongst a population in ways that reinforce momentum and the cycling of resources.

This concept must account for a general principle of all social systems, the power law distribution^[1]. It has been observed by Clay Shirky and others that in many, if not all social systems, a small minority of members is the most productive, exponentially so. The curve looks like this:

Since the distribution is so skewed to the left, the law of averages does not really apply. The majority of work is performed by a tiny minority of contributors, so if we were to look at the average number of contributions per contributor, the number would be deceptively low. The vast majority perform a small number of tasks, whereas just a few contributors do much more than their share.

The top left of the curve is equivalent to the bottom point of the Contributory Funnel. To get some idea of how small this point is, take a look at this article.

The power law distribution tells us a number of things. First, we can infer that the rate of growth of the most active users is a fraction of the rate of growth of overall contributors; therefore, the larger the population of contributors, the more likely that a few more of them will become active users.

Second, not all contributors are equal. It is not reasonable to expect that the contributors at a certain level of activity are equivalent to or overlapping with contributors at another level of activity. There are different contributor populations.

These two observations may seem contradictory, but in fact they can co-exist. Contributors can change which part of the curve they belong to over time, and this cannot be measured without the notion of the rate of contributions. The rate of contributions is simply the number of contributions over a small time period, such as one month. So contributors who make three contributions per month belong to a different part of the curve than contributors who make 50 contributions per month. Over time, their rates of contribution may change, so that they move from one part of the curve to another. However, it is important to recognize that their rates of contribution signify which part of the curve they belong to at a certain point in time.

One obvious implication of this line of thinking is that measuring contributions by cumulative number of edits per user is limited.

Of course, not all contributions are equal. Minor proofreading edits are certainly less difficult and valuable than researching and adding references, or composing whole sections of an article. This complex topic is addressed below, in Task taxonomy. In general, the more difficult and valuable contributions also tend to fall to the left of the curve, and a simple weighting of tasks by complexity would make complex tasks equivalent to some multiple of simple tasks.

The virtuous circle

Let's get back to the overall ecosystem. Given the power user distribution, different populations of contributors can co-exist and in fact cooperate, creating a virtuous circle. Small tasks can be handled by the majority of contributors, whereas more complex tasks can be handled by the power users. A virtuous circle is a convenient way to visualize a balanced, flowing ecosystem.

There are two useful concepts when considering the interplay between power users and the majority of contributors. The first is that power users are suitable candidates for defining and requesting tasks. The second is that some tasks can be farmed out to a broad population of contributors in a crowdsourcing fashion. Automation can assist both kinds of activities.

Tasks can be categorized along several dimensions, and the following is helpful for understanding the interplay between tasks themselves and contributors:

time available
interest areas, usually related to content
expertise or skills

This is a user-centric view of tasks, as opposed to the task taxonomy discussed below. In trying to understand the widest part of the Contributory Funnel and the flow of tasks that can create a virtuous circle, this user-centric breakdown can be helpful.

A mobile example

Here is an example. Let's consider the interplay between several new proposed features in a mobile context.

The mobile team has begun to think about watchlists as not only a useful feature, but also as a tool for engaging users who are from the broad universe of readers. Watchlists are currently an editor-oriented feature on the desktop site, but there is a way they can be used as bookmarks or a reading list. This is a common request among mobile readers, and when the mobile apps introduced saving pages offline, it was quite evident that some users were using the feature to save large numbers of pages. A reading list can be stored locally on the device, and in some iteration it could be linked to saving pages offline.

Now let's consider an anonymous reader who wants to save a reading list. We can enable that via a local tracking device, such as a cookie. And we can make clear the benefit of creating an account - the reading list will be stored in a safer way and it can be accessed on any device. Maybe the user doesn't create an account right away, but we gently remind him or her at various points of the benefits of registering. This is the Proto-Account in action (see the Funnel diagram above, and please note that there may be a better user-facing name for this concept, such as "Early Account").

As soon as the user does register, we have an opportunity to deepen engagement. Now the user has a reading list that is accessible from any device. What if we did this - encourage feedback on the articles in the reading list with the mobile version of the Article Feedback Tool? Specifically, we can surface a simplified feedback widget that not only allows quick feedback, but also shows other comments that the user can review and "like." This looks a lot like other social tools that are popular but serves some specific purposes:

comments from other users can help shape this user's own comment
this user becomes more attentive to the content of the article
this user sees that articles are living, growing things
the "like" action adds a form of moderation to the comments, reducing the review burden for experienced editors

Taking a look at this task from the perspective of the user, we can see that:

it fits the small time window typical in mobile usage
it is directly related to the user's interests - namely the articles on the users' reading list
it fits the basic level of expertise that suits newly registered users

Of course, this scenario will require many refinements and design considerations to perfect the flow, but in essence it can create a virtuous circle between a broad population of new contributors and the tiny population of experienced contributors.

Task taxonomy

The universe of tasks is diverse and difficult to categorize. This is partly due to the virtue and curse of wikis: nearly anything is possible (it just may not be easy or elegant).

Nevertheless, it is helpful from an internal perspective to organize tasks so they can be rationalized within the Funnel and virtuous circle.

Thanks to Dario Taraborelli, here is a general taxonomy of tasks and note that task review can be organized in the same way:

Task
- Size/Effort
  - Microtask - tasks that require less than an edit to complete
    - voting
    - rating
    - tagging
    - flagging
    - bookmarking
  - Macrotask - tasks that require at least one edit to complete
    - inline editing
    - full editing
- Expertise
  - None - tasks that virtually anyone could perform, like fixing typos
  - Topic - tasks that require topic expertise (like finding an appropriate citation for an unsourced statement)
  - Scholarly - tasks that require a particularly high level of topic expertise
  - Local - tasks that require local knowledge (like determining if the coordinates of an article are accurate)
  - Community - tasks that require knowledge of community policies and guidelines (like determining if a topic is notable enough for inclusion)
- Crowdsourceability
  - None - the task cannot be broken down and distributed into modular subtasks
  - Partial - part of the task can be broken down and distributed into modular subtasks
  - Complete - the entire task can be broken down and distributed into modular subtasks
- Delivery
  - User - task queues delivered to individual users with an account
  - Reader - task queues generated to readers with no account
  - Article - task queues attached to invidual articles
  - Topic - task queues are attached to a topic and are delivered via categories or WikiProjects
  - Location - task queues are attached to locations and are delivered via location-aware features
  - Task repo - tasks queues generated in a central repository where they can be searched
Task review - same structure

Research

Here are some relevant links:

Internal

External

References

↑ Clay Shirky, Here Comes Everybody, p. 123

[1] Clay Shirky, Here Comes Everybody, p. 123

[1]