情報処理学会 第77回全国大会講演要旨

People now spend more time on social media than other internet activities. On Facebook, however, it is possible to miss posts from friends who have similar interests but who are only weakly associated with the user, since Facebook ranks posts based on affinity and type of post but rarely inspect the contents of the posts. We are therefore trying to discover friends with similar interests and recommend their posts to the user. To calculate similarities we use tf-idf and cosine similarity to measure how closely related one document is to others in a collection. We combine that with hypernyms to extract the generic meanings of words, which yields a better similarity metric. Using this approach we can connect Facebook users with weakly-associated friends whose posts are demonstrating similar interests.