News Feed Filtering with Explanation Using Textual Concepts and Social Contacts

近年來數位資訊的資料量伴隨著網路型態的轉變而益加龐大。人們不但吸收著傳統媒體發佈的訊息，另一方面每個人也都成為資訊的來源或者轉送樞紐。隨著網路應用涉入越來越多日常生活，資訊氾濫的問題影響越加廣泛。傳統上處理資訊氾濫的問題，有一大研究方法是採用推薦系統的思維。也就是依據一些經驗法則和機率模型來推薦使用者有興趣的物件。這些方法大多依據使用者行為或者物件彼此之間的相似度做判斷。然而，在社群網路之上的資訊交流者彼此之間的親疏關係，也影響著資訊的重要程度。過往的方法著重在推薦使用者有興趣的內容，而忽略了社交關係對資訊交流的影響。而另一大處例資訊氾濫的問題的方法，便是做資料重點彙整。系統將整體內容或者與使用者指定的關鍵字相關的內容，將重點標示出來。 讓使用者能夠以較簡短的時間，了解所有資料的大綱。不過，以作者的認知，鮮少有結合推薦與彙總的研究。 在本論文中，我們認為考慮社交關係對於資訊篩選有正面的回饋，並且，經由常識庫的輔助，可以提昇彙總的準確度。我們研究了以個人資料推測出的社交親疏程度。並且將重點彙總以簡單明瞭的方式呈現，讓使用者能夠加掌握最近的消息主題。最後，附加簡短的說明讓使用者更加了解系統推薦的原因，對於整體推薦系統的幫助。 實驗結果顯示，在許多對使用者而言無意義的資料中，單只靠社交關係的資訊，難以找出使用者能有興趣的資訊。但若提供重點彙總後，使用者可瀏覽大致主題，並自行選閱。此時若配合社交關係與說明可提昇使用者對於資訊的興趣。In recent years, the amount of information flowing on the internet is raising. People absorb messages from media and tranfer information to others. As more applications and web services step involved in our daily life, the affection of information overload spreaded. Historially, many studies try to deal with this problem by using recommendation system. That is, recommending users items they might like based on user profiling or content similarity. However, the importance of sharing information might partially depends on the closeness between users. Another popular way to solve information overload is to summarize documents. System gives a general summarization or part of content related to specified keywords. In this thesis, we investigate if social contact information and summarization help users geting more interested in sharing content. The summarization for sharing content is based on ConceptNet which stores sematic relations between words. In addiction, explanations for each sharing contents address the relevance to summarized topics and the strength of social relations with senders. Experiment was established as a Facebook Application. The result showed that for users receiving many unimportant sharing content, social contact was hardly to filter out interesting ones out of them. However, summarization gave users an overview of sharing information and users felt much more interested in them with explanations provided