2011年1月14日 星期五

Seminar Paper Report about Social Media Recommendation

Social Media Recommendation based on People and Tags

Social Media Recommendation based on People and Tags
由 Ido Guy, Naama Zwerdling, Inbal Ronen, David Carmel, Erel Uziel所提出
發表於 SIGIR'2010

--

摘要

網路的時代,多媒體社群網站的蓬勃發展,使用者可以透過社群網站創造或分享各式各樣的資訊,伴隨著各種資訊而來的是標籤(Tag)、排名(rating)和意見(comments),使用者透過這些動作替自己的資訊做註解,進而達到分類與排名的效果,同時也透過這些動作把資訊分享給更多的朋友,然而,社群網站的使用者每天都要面對無數的分享資訊,選擇有興趣的資訊變成一個相當大的難題,而什麼樣的資訊才是使用者真正有興趣的也是一個值得探討的課題。

一個好的社群網站,其網站經營的目標之一,就是留住舊用戶並且吸引新用戶加入社群。

此篇論文探討在限定的社群網站裡,以People和Tags做為資訊推薦系統(recommender)的推薦基礎,然後評估其效益與成果,其優點是我們只需要擁有明確的網頁資訊即可有效做為推薦系統的推薦依據。

--

系統

1. Social Media Platform - Lotus Connection [18]
A social software application suite for organization.
All 7 items. (profiles, activities, bookmarks, blogs, communities, files, and wikis.)

2. Relationship Aggregation - SaND [5,27]
Models relationships through data collected across all LC applications.
Aggregates any kind of relationships between people, items, and tags.

Builds an entity-entity relationship matrix: "direct relations" and "indirect relations"

3. User Profile
Person-person relations [15,16,17]
Aggregate direct and indirect people-people relations into a single person-person relationship.

User-tag relations
used tags:
direct relation based on tags the user has used.
incoming tags:
direct relation based on tags applied on the user.
indirect tags:
indirect relation based on tags applied on items related on the user.

4. Recommendation Algorithm (略)

5. Evaluation (略)
(In total, 412 participants completed our survey, originating from 31 countries and spanning the different organizational units)

--

Conclusion

hybrid people-tag-based recommender other advantages:
* low proportion of expected items.
* high diversity of item types.
* richer explanations.

leading to a 70:30 ratio between interesting and non-interesting items when explanations are included.

--

討論

對於實驗中的時間因子控制,常見方法是使用半衰期。
(假設時間越久遠的效力就越低)

論文中所使用的是推薦系統中常見的方法,然而這類型研究較困難的地方多半在於如何評估其成效,要思考如何證明實驗評估的基準有足夠的可信度,一般來說是以使用者參與(User Study)的經驗來做實驗評估,然後依靠大量的實驗參與者,來提高實驗的可信度。

0 留言:

張貼留言

Related Posts Plugin for WordPress, Blogger...