Masters Project: Integrating interest-based content recommendation into social feeds

Yaron Yitzak summarizes the results of his masters project Integrating interest-based content recommendation into social feeds:

A social feed such as Twitter’s home timeline displays follow based content; updates (e.g. tweets), in chronological order, from followed accounts. Not necessarily all the content shared by these persons and organizations is considered of interest by the following user. This study investigated whether the insertion of purely interest based recommended updates into Twitter’s social feed increases the level of interestingness and whether receiving such non follow based content is actually desired.

An experiment was conducted to test the hypothesis that the insertion both increases the interestingness and is desired. An algorithm called the Social Data Mining and Ranking Algorithm was devised to generate the interest based content recommendations. For a multitude of topics of interest, which participants could choose from in the experiment, the RSS/XML feeds from related online publications were collected. The latest (i.e. today published) items (i.e. news articles, blog posts) from these feeds were identified and sorted by a variable called Facebook’s total count (i.e. total number of likes, shares, comments for a specific article). The top 3 articles, for each of these topics, were posted (title + link) to for the experiment specially created Twitter accounts (e.g. @tip_technologie, @tip_muziek).

Participants were presented their latest 10 Twitter social feed updates (as embedded tweets on a blank HTML page, generated by a for this study created Twitter application) and were asked to rate them interesting or not. Furthermore, they were presented the same tweets yet with a percentage of them replaced by some of the tweets from the earlier mentioned Twitter accounts containing potential recommendatory updates for their chosen topics of interest. They were presented with three pages containing interest based content recommendations. One with 10% interest based recommended tweets, one with 30% and one with 50%. The mean interestingness of the page containing 30% interest based recommended updates and even more so the page containing 50% interest based recommended updates proved to be significantly higher than the first page, solely containing original social feed updates (follow based). Furthermore, qualitative data showed that a clear majority of the participants appreciated the insertions, for reasons such as the decreased need to search for interesting Twitter accounts to follow in order to receive desired updates on the social feed.

Posted in Masters Projects

Leave a Reply