Explanation

做事情常常我都靠直覺, research也是一樣, 都是直覺的去感受該怎做就怎做, 只要心裡覺得make sense就好, but present自己的work的時候, 不能去假設別人也和自己擁有同樣的直覺, 因此, 在這裡要去詮釋心理的那種直覺為什麼是make sense的原因給人家聽, 當然自己覺得是trivial(小黃常說的口頭禪, 好懷念QQ), 也不能假設別人覺得是trivial, 最重要的重點, 是把自己的直覺化成可以讓大家聽得懂得論點, 人家覺得有道理, 而且自己實驗跑出來也可以驗證自己的直覺是有道理的, 那就是成功了!

Concentrate on one idea: CF

仔細想想, 我的work就是想貫徹一個精神: CF can be applied in Tag and Bibliography Recommendations and get good performance! 既然如此, 不需要想太多, 堅持CF, 主軸就在CF. 我該做的就是去Verify我的Assertion是正確的! 重要的是, 需要說服人家, 讓人家接受自己的idea.

Boys be ambitious

  • 要對自己有信心, 堅持自己的風格, 把所學的一切與自己堅持的理念, 在最後的oral test全部表達出來 !
  • 不能太過隨和, 對於每個事物的現象, 一定要有自己的想法和見解 !
  • 別人attack自己的觀點時, 不要害怕, 說明自己的立場與想法 !
  • 說出的話, 每一句都要符合邏輯, 合理, 切勿多說, 講重點即可 !
  • 一定要有信心, 熱誠, 來講自己的work, 別人才會有心情想聽 !

Friends

昨夜不小心認識了一位法律系的朋友
跟他玩五子棋
之後又聊聊了一下
覺得心情還不錯
他說的很好
朋友就是要互相幫忙
至少在道義上是要彼此有 symmetric relation 才算是 friend 吧
I like the feeling : )

大前天則是利用無名的 friend finder (模仿Facebook的嘛)
找到了國小好朋友ㄚ德
這位好朋友也算影響我滿深的
如果沒有認識他 我怎會有機會接觸GAME呢 ?
現在之所以能唸CSIE 而且還有熱枕
其實大多是從小時候一直累積的對電玩的熱情吧
妙的是
這位朋友一點都沒變 電玩一直是他最最最不可或缺的生活必需品呀
他的銘言: 我人生的1/4都花在打電動上了
哈 ! 堅持到底就是了 !
Anyway, 不管如何 我們都一定要堅持自己的夢想 而且要去守護自己喜歡的東西!

而在星期四的小meeting
也很感謝同組朋友們的幫忙
把我 presentation 的盲點找出來
讓我可以體會到如何用比較正確且吸引人的flow來 present my work

總而言之
本週還體會滿多東西的
有進步 值得 happy

keep going !!!

謹記

口試就是要好好的講一個故事給大家聽
最重要的就是故事在邏輯上要前後連貫且相互呼應
而且講者也要有熱誠來講這個故事
當可以引起人家的共鳴
而且能夠確實的把自己心裡的想法講給人家理解時
才算是講好一個完整的故事

[ Defense ]
想好所有步驟為什麼要這樣做
為什麼不用其他的方法
為什麼這方法在這裡可行
為什麼自己會想這麼做
這樣的作法有什麼理論根據或是假設

這些都要一一想清楚
才能上場擋下嘴砲

Notes

  • Rehearsal Dates: 5/21 (for oral defense)
  • Defense Notes:
    傳統的 CF 推薦系統
    是假設具有相似購買行為的使用者會買類似的產品
    在本 tag suggestion 機制
    也是利用類似的想法
    假設相似的paper會有相似的標註

    demographic recommender
    demographic data: e.g. 性別 年齡 職業
    demographic vector: 維度是寫死的 (not flexible)

    Information filtering systems can help users eliminate useless documents and bring to their attention only the relevant information. This implies that the system has to be able to recognize the users and to maintain a model for their interests.

    LifeStyleFinder: demographic recommender system based on profiling user in demographic clusters.
    SiteIF: web page recommender based on keyword matching

    SIFT: a content-based recommender for Netnews recommendations.
    http://wwwis.win.tue.nl/ah98/Stefani/Stefani.html

    Article related to tagging process
    http://en.wikipedia.org/wiki/Tag_%28metadata%29

    Introduction to Collaborative tagging systems
    Systems that allow users to user their created keywords to organize and share content

    Example:
    delicious: bookmarking,
    flicker: photo
    CiteULike: bibliography

    Motivation 1: How to enhance content sharing in tagging community
    1. create tag related to the content
    2. use popular tag

    Problem 1: However, sometimes we have no idea to tag an item we are not familar with.
    => we need tag suggestion mechanisms
    .
    Problem 2: Information overload (content accumulated to massive amount)
    1. too time-consuming to find useful information (explain def. for useful)
    => we need content recommendation mechanisms

    In our work, we want to create a collaborative tagging system for bibliography management. In addition, we proposed tag suggestion and content recommendation mechanisms to solve the problems above. (come with the illustration for problem definition)

  • Related work
    1. collaborative tagging system for bibliography management
    – CiteULike, Bibsonomy
    2. recommender system
    – demographic approach
    – content-based approach
    – collaborative-filtering approach

    Data analysis
    – CiteULike dataset
    – analysis in 3 views: user, tag, item (explain for long tail phenomenon)
    – sparsity and user correlation (item vs. tag)
    – conclusion for data analysis

    System components overview and Implementation
    – bib agent (with flow chat built by left)
    – bib management system (can come with database schema)
    – bib search engine (Lemur indexing)
    – tag suggestion engine (explain for CF approach) with evaluation
    – bib recommender (2 simple mechanisms) with evaluation

    Conclusion
    – What I have done in this work

Notes

  • TiVo application: It is a TV show recommender system implemented in decentralized architecture. Users’ preference on shows was presented in the form of rating (from score -N to +N) and the preference score was rated in the client side. The TiVo server periodically aggregate the clients’ rating and send the rating table via ring mechanism to the client. Each client received the rating table utilize the collaborative filtering mechanism to predict the rating score of the shows he/she has never seen. Next, after the predicting process was completed, the show recommender at client side will make show recommendations for its user.
  • Fab Application: It is a web pages recommender system integrated both the content-based and collaborative filtering approaches. In Fab, there are three main system components, namely collection agents, selection agents and the central router. Collection agents are designed to collect web page of a specific topic and central router will aggregate the collected pages. Next, central router will forward pages which related to users’ interest profiles to system users and users’ selection agents will make  selection to the forwarded pages (e.g. discard pages have seen by users). In this phrase, content based mechanisms are implemented to make recommendation from central router to users. In next phrase, the concept of collaborative filtering mechanisms are applied. Selection agents at user side will forward the pages to the nearest neighbors according to some metrics for similarity calculations. To sum up, Fab combine both the well-known recommendation approaches to produce hybrid recommendations, which was running under the distributed architecture.