紀要論文

ソーシャルタグを教師データとしたコンテンツ推薦システム

Content Recommendation System using Teacher Data created by Social Tag

jpn

18810667

https://doi.org/10.15002/00010534

法政大学大学院情報科学研究科

9

131

136

2014-03

Version of Record

Social tag as a significant feature in social bookmark service can be used for metadata which describes contents related to tags. The metadata is applied to generate teacher data. Generally, a tag used by a user can somehow reflect the user's interests and preferences. This research is focused on using the relations of tags and their associated contents as learning data for text classification. The resulting text classifiers take a text as input and generate a vector of values from 0 to 1 associated to tags. The text classifiers are applied to a user's visited text documents on SNS for generating the user's profile including a collection of vectors. Based similarity of any two vectors, the system recommends the most relevant contents to the user. In the experiments, 200 tweets are using for generating a user's profile including about 50 tags and the three types of contents, website, web advertising, and books from Amazon.com are recommended to the user. The result shows that the system can generate a user profile reflecting the user's interests to a certain degree if the user is an often Twitter user. For obtaining reasonable recommendation contents, it is important to have sufficient tags in the user profile.

Article

AA12222297