We believe that you should always be able to discover new people to engage with, and our goal is to make this easier for you. Therefore, this past Sunday, Sid (front end engineer) and I (Ferenc, a data scientist) were hard at work, coming up with a few simple, yet powerful upcoming features that we hope will improve user experience around PeerGroups. In this post we wanted to give a little sneak peek into a small widget we came up with, called “Groups you may like”.
This is one of those features which does what it says on the tin; it sits on your dashboard and recommends groups that we think you will want to check out. We wanted to build on our existing strength (having detailed deep analysis of people’s activity and influence) to create a simple prototype, that’s immediately useful to you.
We started by thinking about the question “Why should a group interest me?”, and came up with these answers:
- overall popularity: Not all groups are created equal. Some are just more interesting and popular than others. If we had no other information about you, the least we can do is to suggest you check out groups that are trending and popular amongst our users.
- overlap in community: You may want to check out groups whose members include your friends and people you engage with. We have already identified people who influence you the most, and have data on people you converse with most often, so if a group has a large fraction of these people, we’ll recommend that group to you.
- overlap in interests: We have a very fine and detailed analysis of each user’s activity and influence in over 1500 topics (and the number is growing). We use that information to detect if your areas of interest overlap with that of members in a particular group. Here, we put more emphasis on smaller topics: you’re highly active in a niche topic like “Peer Influence”, and we happen to find a group whose members specialise in this very topic (such as the PeerIndex Team), than that group is probably highly relevant to you.
- closeness to you: PeerIndex are a multinational community: we have active users from the U.K, U.S, France, Germany, Romania, Brazil, Spain and several other countries. Whilst the French may be interested in French Marketing Professors, Aussies will want to check out Top Australian Social Media People instead. We can therefore also take into account your location and language when recommending groups.
On our Sunday hack-day we’ve implemented overlap-in-community aspect, and it alone started generating surprisingly sensible recommendations to us. In the following days we will keep adding further criteria, and testing the widget on the PeerIndex Team.
The rewarding part of working at a startup like PeerIndex is that the feature may well go live already by next week. Once we roll it out we will start to have finer data on which groups you actually like. For example, by observing which recommendations you click on, which groups you tweet about, which pages you spend most time interacting with, we can start turning the knobs on our system to ensure the widget always recommends stuff that are actually useful to You.