The group administrator Final Fantasy Club in “Vkontakte” Zaur Shukurov has written for vc.ru a column about the experience of doing community. How to draw the minor drawbacks of the structure of the groups in the pros, to automate part of the process, to allow users to decide what they want to see on the main page, and how to “teach” them to like. In “Vkontakte” there are two types of communities.
Public and group. If the public is primarily intended to supply news groups – for communication. They are slightly different in structure. The first insignificant negative occurs when an administrator or an editor adds the image in any album.
Then they appear in the feed subscribers with special dice “added N photos” like its a separate entry. Minus the second album on the main page, which can not be removed. The problem could be solved is to convert the group to the public and forget about everything.
But then it turned out that the group will soon celebrate the 10-year anniversary and it would be good to stick to the roots, so this option was dropped. Had to come up with something new for the current structure. Came up with the idea. You can put some number of images in the album in the main, to appear in the ribbon user.
Then look at what more likes, and to pay to the master, thereby allowing subscribers to decide which content will be published. The appearance of additional manual work upset, but I remembered that a little know how to program. Before we talk about automation, I note that the Final Fantasy series of Japanese video games, which is available since 1987 and has 15 games in the main line, each of which is unique in plot, setting, characters and more.
In this regard, art, cosplay, and other content from the game series very much. Previously, a manual search of high-quality, unique and fresh images took a lot of time. Before upload, the image needs to find, sort, and process. In version 1.0 as sources of images used the services DeviantArt, Pinterest and Tumblr.
Through the API on DeviantArt and Pinterest you can search for popular for a particular query, which is very convenient. Less Tumblr friendly. There you can search only by tags, and in most cases come across inappropriate pictures. But a lot of blogs with a rich collection of images, already collected.
Find most popular on our subjects and parsim from there direct links to images in the database. It is important. Respect the authors, and do not forget to specify references to them in the description to images where you will be spread. We never had problems with owners, and neither should you.
When found the content, move on to the next step. Create two albums in the group. In the first daily at nine in the morning by the script automatically throws five random images from the database that will “fight” between themselves and the rest for the opportunity to get on the main, based on the number of likes on them. Do not forget to leave the album open to any subscriber could also send the image to participate in the “competition”.
The second album (and also “the way”) sent by the selected algorithm the image is the winner, so that it no longer participated in “the contest”. The selection algorithm was simple. It was considered the average number of likes on the album, chosen image, where the likes more than the average album, and already among these images some one accidentally went in the “way”.
Put in Cron a script to run in the morning, afternoon and evening, and are satisfied with the result, “drinking smoothies and poking at a tablet computers”. The disadvantages of this version.
It was therefore decided to make a second, more advanced version, but with a little human intervention. For the second version was written a simple website for administrators, which in real time can be watched, and filter images from DeviantArt, Pinterest, Tumblr and Instagram in popularity. As you can see from the screenshot showing the ribbon DeviantArt, under each image there is a button “publish”, and therein lies the fundamental difference from the first version.
When this button is pressed, the picture enters the buffer “selected”, where if you want you can delete some picture if it was added accidentally. Now every day at nine in the morning already from this buffer selected by the editors of image are laid out in the main album. Random images from the database are used if the buffer is nothing. Manual work has increased, but increased quality.
Otherwise almost nothing has changed. Was only slightly simplified algorithm of selection when sending the “way”. Cons disappeared. In my spare time I dabble in bots for a Telegram, so to pass this platform would be sacrilege, given its convenience, simplicity and high speed.
I created two bots. Public and private administrators. In private bot out of turn to select from DB or albums random picture and send it to the group. But lets not focus on Telegram, we turn to statistics.
Throughout the year (16 Feb 2016 16 Feb 2017) the system was tested, and thats what happened. To “pending orders” came 976 images that put 154 982 likes.
But not all of them come true in the main, and only 711, which set 140 720 likes, the rest were either rejected by editors, or are still waiting. So, on average, one entry has 198 likes. Read more:
The first version worked from February to August 2016, the second running from August 2016 to the present day. Through such simple actions failed: The “smart” ribbon. Send a private cases, in which you managed to improve (or, conversely, worsen) the performance of the project, [email protected]
Interesting experiments are sure to get heading Growth Hacks.