Contributions from community members in the form of valuable ideas are seen as strategic assets in the success of IdeaScale initiatives. The larger the community of participants the more diverse views are likely to appear; more diversity increases the chances of producing valuable ideas. But the chance for success also increases, when the behavior within communities is also understood. In this blog article, I provide evidence that demonstrates the presence of common patterns in the behavior of IdeaScale communities’ members. By applying Machine Learning techniques, we found that the evolution of user actions (i.e., member registration, idea submission, comment posting, and vote casting) within communities over the first 12 months of life is shaped by five well-defined patterns. See figure below.
Collective Patterns. For most communities (142 out of 166, 85%), the evolution of registrations over the first year of their life follows patterns 1, 3, or 5 (see the table below for the list of patterns). In behavioral pattern 1, which we call Q1 peak and gradual descent, 55 (33%) of the communities show to have a burst of registrations during the first three months of the year and then the number of new members gradually decreased or remained somehow constant until the end of the period. Communities that follow behavioral pattern 3, which we call Q1 peak and rapid decent, (53 out of 166, 32%) show, however, a more prominent peak of registrations during the first quarter. In fact, between 50 and 75% of registrations occurred in that period of time. Then, from the second quarter on, the proportion of registrations falls remaining stable around 25%. Behavioral pattern 5, which we call Q1 peak and super rapid descent, represents a more extreme case of pattern 3. Here, between 75 and 100% of member registrations happened in the first quarter. Then the number of member registrations decays drastically and remains very low until the end of the period. A quite different pattern is followed by 13% of the communities, which corresponds to behavioral pattern 2, which we call Q2 peak and very rapid descent. Instead of having large proportions of registrations at the beginning, they concentrate their registration activities during the second quarter (from month three to half-year). After that period, the registration of members falls down to quite low levels. Finally, very few communities (4 out of 166, 2%) show peaks of registrations towards the end of the year (behavioral pattern 4, which we can call Q4 latter peak). This type of behavior could be considered more an outlier than a pattern.
Interestingly, for the rest of the actions, i.e., idea submissions, comment posting, and vote castings, communities follow the same patterns. However, the distribution of communities per pattern is different as shown in the table below. Although the distribution of communities in each pattern show to be different from action to action, a general trend can be seen: patterns 1, 3, and 5 are followed by the majority of the communities. Pattern 2 depicts the behavior of about 6 to 15% of the communities for each action while pattern 4 is rather negligible. One third of communities (55 out of 166) follow the same collective behavior for all of the action types. Such commonality suggests overall attention peaks, where contributions —in all forms might— follow member registration.
Influence of moderation in collective behavior. Different factors may influence the collective behavior of communities. We have no information about the external factors, such as promotional events, incentives, or other public events because they are not registered in the data set. Other factors are internal and in this analysis, we investigated if there was a relationship between moderator interventions and behavioral patterns, understanding moderator intervention as all submissions (ideas, comments and votes) performed by moderators and organizers of communities. Interestingly, communities that follow patterns 1 and 3 are at the same time those that show the strongest presence of moderators. On average, moderators intervened 2.5 times (69.71 vs. 27.92 interventions in average) more in communities in which their ideation actions are shaped by patterns 1 and 3 than in communities that follow patterns 2, 4, and 5. Similar numbers were found when studying the participation of moderators in communities where commenting and voting are governed by these patterns. By splitting interventions into quarters, we observed that periods with high level of activity correspond to quarters of high activity by moderators.
Take action as early as possible. Overall, communities follow the same collective behavior pattern for all action types, i.e., for member registration, idea submission, comment posting, and vote casting. From the results that we reported earlier, patterns that show higher activity levels at the beginning of the life of communities prevail. This common behavior might be the effect of the early enthusiasm occurring soon after the launch of a community or the result of additional external factors such as the promotion of the initiative outside the IM platform or the incentive offered by the organizations to the participants. The implication of this behavior is that organizers or moderators who want increase the volume of interactions by members of the community should focus their efforts during this early period of high activity and high rate of member registrations, as opposed to leaving such efforts towards a later time.
For more details about the methods used in the analyses and the study in general please refer to the publication and the website of the study. The dataset and the R scripts used in the analyses are available at the following Github repository. All data were collected using the IdeaScale API client, IdeaScaly.
University of Trento, Italy
Catholic University of Asuncion, Paraguay