音樂歌單是由一組多首不同元素、風格的音樂所組成的，它包含了編輯者的個人品味以及因應主題、目的性產生而成。我們可以透過樂曲的律動、節奏、歌曲的主題精神，進而編輯一個相應契合的系列歌曲。當今的音樂收聽市場主要是在網路串流平台上進行隨時、隨地的聆聽，主要的平台有Spotify、Apple Music 以及KKBOX。各家業者不單只是提供使用者歌曲的搜索、單曲的聆聽，更提供訂閱專業歌單編輯者的歌單訂閱服務，甚至是讓一般的使用者參與歌單自訂編輯的過程。然而如何在有限的時間內針對使用者的聆聽習慣去介紹平台上豐富的音樂資源是個很大的挑戰。上述的過程我們稱之為推薦，而當前的音樂推薦研究大多是在對使用者進行相關歌曲的推薦，鮮少能進一步在更抽象層次上的歌單上進行推薦。這邊我們就此一推薦應用提供嵌入式向量表示法學習模型，在有著使用者、歌曲、歌單的異質性社交網路上，對使用者進行歌單的推薦。為了能有效的學習出歌單推薦的模型，我們更將使用者、歌單和歌曲的異質性圖形重組成二分圖（bipartite graph）， 並在此圖形的邊上賦予不等的權重，此一權重是基於使用者隱式反饋獲得的。接著再透過隨機漫步（random walk），根據邊上的權值進行路徑的抽樣選取，最後再將路徑上經過的節點進行嵌入式向量表示法的學習。我們使用歐幾里德距離計算各節點表示法的鄰近關係，再將與使用者較為相關的歌單推薦給使用者。實驗驗證的部分，我們蒐集KKBOX 兩年份的資料進行模型訓練並進行推薦，並將推薦的結果與使用者所喜愛的歌單進行準確度（Precision）評估， 結果證實所得到的推薦效果較一般熱門歌單的推薦來的好，且為更具個人化的歌單推薦。 / Music playlist is crafted with a series of songs, in which the playlist creator has controlled over the vibe, tempo, theme, and all the ebbs and flows that come within the playlist. To provide a personalization service to users and discover suitable playlists among lots of data, we need an effective way to achieve this goal. In this paper, we modify a representation learning method for learning the representation of a playlist of songs, and then use the representation for recommending playlists to users. While there have been some well-known methods that can model the preference between users and songs, little has been done in the literature to recommend music playlists. In light of this, we apply DeepWalk, LINE and HPE to a user-song-playlist network. To better encode the network structure, we separate user, song, and playlist nodes into two different sets, which are grouped by the user and playlist set and song as the other one. In the bipartite graph, the user and playlist node are connected to their joint songs. By adopting random walks on the constructed graph, we can embed users and playlists via the common information between each other. Therefore, users can discover their favorite playlists through the learned representations. After the embedding process, we then use the learned representations to perform playlist recommendation task. Experiments conducted on a real-world dataset showed that these embedding methods have a better performance than the popularity baseline. In addition, the embedding method learns the informative representations and brings out the personal recommendation results.

隨著手持裝置的普及與社群網路的興起，大眾可以隨時拍攝影片並且上傳至網路上與他人分享。但是一般使用者產生的影片若少了配樂，將失色許多。除了原本影片帶給人們的視覺觀感之外，配樂可以帶給人們聽覺的觀感，進而使得人們可以更容易了解影片的情感，也可以讓人們更能夠融入在影片中。背景音樂推薦的研究主要有兩大種做法，Emotion-mediated Approach與Correlation-based Approach。我們使用Correlational-based Approach的方法，利用Correlation Modeling找出影片特徵值與音樂特徵值之間的關係。但是由於目前Correlation-based Approach的研究只有考慮到全域特徵，因此在此論文中，我們提出了區域特徵。區域特徵利用時間序列表達影片細部的變化，並且將區域特徵與全域特徵結合至Correlation Modeling中，透過 MLSA、CFA、CCA、KCCA、DCCA、PLS、PLSR演算法找出其中的關係並且產生背景音樂推薦的Ranking List，實驗部份比較了各個演算法在背景音樂推薦上的準確率，並且觀察Global Features與Local Features之間的準確率。 / Background music plays an important role in making user-generated video more colorful and attractive. One of current research on automatic background music recommendation is the correlation-based approach in which the correlation model between visual and music features is discovered from training data and is utilized to recommend background music for query video. Because the existing correlation-based approaches consider global features only, in this work we proposed to integrate the temporal sequence of local features along with global features into the correlation modeling process. The local features are derived from segmented audiovisual clips and can represent the local variation of features. Then the temporal sequence of local features is transformed and incorporated into correlation modeling process. Cross-Modal Factor Analysis along with Multiple-type Latent Semantic Analysis, Canonical Correlation Analysis, Kernel Canonical Correlation Analysis, Deep Canonical Correlation Analysis, Partial Least Square and Partial Least Square Regression, are investigated for correlation modeling which recommends background music in ranking order. In the experiments, we first compare the results of only global features, only local Features and incorporating global and local Features among each algorithm. Then second compare the results of different clip numbers and Fourier coefficients.

近年來電子商務蓬勃發展，嚴重侵蝕實體通路業績，因此線下服務提供者更應善用資料科學技術，找出顧客未被滿足之需求，進而提供優質服務，其中脫穎而出的關鍵非推薦系統莫屬。 本研究以運用計算產品相似程度的「項目導向協同過濾」和計算使用者與商品蘊含特徵的「潛在因子」兩大類「協同過濾」推薦方法為核心，藉由實體零售通路累積的顧客消費紀錄，驗證「協同過濾」方法較傳統熱門商品推薦機制更符合消費者偏好，且「協同過濾」方法能達到完全個人化推薦之目標。 本研究使用的實體零售通路消費紀錄源於顧客真實購物行為，收集成本低，且數據量龐大，然而此類資料無法直接傳達顧客對於商品的喜好與滿足程度，因此被稱之為「內隱資料」，針對內隱資料處理上，本研究選擇以消費次數取代金額，提出短期重複行為計算閾值概念，以時間修正權重處理可能的偏好轉變與習慣性消費。 模型評估方面，透過強調推薦順序的「平均排名百分比」作為指標，利用傳統熱門商品推薦為基準，比較「項目導向協同過濾」和「潛在因子」兩大類「協同過濾」方法推薦品質的優劣，本研究顯示兩大類「協同過濾」方法達到的推薦品質皆優於熱門商品推薦，且前者遞交的推薦清單為完全個人化，運用本研究發展的推薦系統，將其導入與應用，讓線下服務提供者在與每位顧客接觸的關鍵時刻，能在洞悉對方需求的利基上，提供令顧客滿意的商品與服務，創造獨特且難以模仿的競爭優勢。

標記社群標籤是Web2.0以來流行的一種透過使用者詮釋和分享資訊的方式，作為傳統分類方法的替代，其方便、靈活的特色使得使用者能夠輕易地因應內容標註標籤。不過其也有缺點，除了有相當多無標籤標註的內容，也存在大量模糊、不精確的標籤，降低了系統本身組織分類標籤的能力。為了解決上述兩項問題，本研究提出了一種結合主題探勘與標籤聚合的自動化標籤推薦方法，期望能夠建立一個去人工過程的自動化標籤推薦規則，來推薦合適的標籤給使用者。 本研究蒐集了痞客邦部落格中，點閱次數大於5000次的熱門中文文章共2500篇，經過前處理，並以其中1939篇訓練模型及400篇作為測試語料來驗證方法。在主題探勘部分，本研究利用LDA主題模型計算不同文章的主題語意，來與既有標籤作出關聯，而能夠針對新進文章預測主題並推薦主題相關標籤給它。其中，本研究利用了能評斷模型表現情形的混淆度(Perplexity)來協助選取LDA的主題數，改善了LDA需要人主觀決定主題數的問題；在標籤聚合部分，本研究以階層式分群法，將有共同出現過的標籤群聚起來，以便找出有相似語意概念的標籤。其中，本研究將分群停止條件設定為共現次數最少為1次，改善了分群方法需要設定分群數量才能有結果的問題，也使本方法能夠自動化的找出合適的分群數目。 實驗結果顯示，依照文章主題語意來推薦標籤有一定程度的可行性，且以混淆度所協助選取的主題數取得一致性較好的結果。而依照階層式分群所分出的標籤群中，同一群中的標籤確實擁有相似、類似的概念語意。最後，在結合主題探勘與標籤聚合的方法上，其Top-1至Top-5的準確率平均提升了14.1%，且Top-1準確率也達到72.25%。代表本研究針對文章寫作及標記標籤的習性切入的做法，確實能幫助提升標籤推薦的準確率，也代表本研究確實建立了一個自動化的標籤推薦規則，能推薦出合適的標籤來幫助使用者在撰寫文章後，能夠更方便、精確的標上標籤。 / Tags are a popular way of interpreting and sharing information through use, and as a substitute for traditional classification methods, the convenience and flexibility of the community makes it easy for users to use. But it also has disadvantages, in addition to a considerable number of non-tagged content, there are also many fuzzy and inaccurate tags. To solve these two problems, this study proposes a tag recommendation method that combines the Topic Mining and Tag Clustering. In this study, we collected a total of 2500 articles by Pixnet as a corpus. In the Topic Mining section, this study uses the LDA Model to calculate the subject semantics of different articles to associate with existing tags, and we can predict topics for new articles to recommend topics related tags to them. Among them, the topics number of the LDA Model uses the Perplexity to help the selection. In the Tag Clustering section, this study uses the Hierarchical Clustering to collect the tags that have appeared together to find similar semantic concepts. The stop condition is set to a minimum of 1 co-occurrence times, which solves the problem that the clustering method needs to set the number of groups to have the result. First, the Topic Mining results show that it is feasible to recommend tags according to the semantics of the article, and the experiment proves that the number of topics chosen according to the Perplexity is superior to the other topics. Second, the Tag Clustering results show that the same group of tags does have similar conceptual semantics. Last, experiments show that the accuracy rate of Top-1 to Top-5 in combination with two methods increased average of 14.1%, and its Top-1 accuracy rate is 72.25%,and it tells that our tag recommendation method can recommend the appropriate tag for users to use.

對於大部分的人而言，寫作是一種深度的表達過程，需要詳盡而深入的描述能力及精準而嚴謹的語意呈現。對於非英語為母語的學習者(ESL/EFL)來說，英語寫作尤其是一個困難的過程，常常會因為單字、搭配字(collocation)、句型結構等方面的認知不足，或是受到母語認知的牽制影響，而造成用詞、語法、甚至語意的錯誤。 近年來，語料庫的發展帶來了豐富的語言資源，不僅可以對語言的使用提供許多統計分析上的資訊，也可以做為語言學習者在特定詞語使用上的參照對象。目前語料庫的詞語使用參照以concordance技術為主，提供以特定字詞為基準的上下文例句排列，這種功能對ESL/EFL學習者於寫作過程所提供的參照仍然相當有限；而對於不同的查詢條件，若沒有良善的輸入和比對機制，往往搜尋上的數量和例句品質無法滿足學習者的參照需求；除此之外，對於檢索結果語料庫也沒有評估與排序的概念，學習者往往需要在大量的資訊中篩選出有用的資訊；綜觀以上幾點，目前語料庫技術無法滿足不同程度的ESL/EFL寫作者在參照協助上的需求。 本研究提出一種文句推薦方法，針對ESL/EFL寫作者在寫作過程上的認知不足或不確定，從語料庫中尋找出可用的參照例句，進而提供針對性的協助。我們的文句推薦方法包含三個模組，第一模組是一個彈性的表達元素模組，能針對寫作過程中的語言資訊需求，以規定的表達元素呈現作者的認知需求。第二模組是一個檢索模組，指從語料庫中比對尋找符合作者語意需求的例句。第三模組是一個排序的模組，針對找出的例句，評估其符合作者語意需求的程度，並將之排序，以提升例句參照的使用效率。 本研究依據文句推薦方法實做雛形系統，以British National Corpus和科學人雜誌語料庫(Science America)當例句來源，並依據論文的客觀評估和學習者的問卷調查兩種評估方式來評量雛形系統的成效。經由上述兩種評估方式，本研究驗證雛形系統能針對ESL/EFL於寫作過程中給予一定程度的協助成效，證明本研究的文句推薦方法確實能達成寫作協助的既定目標。 / For most people, writing is a process of profundity. It requires well description of capabilities and excellent semantic of representations. For ESL/EFL (English as a Second Language/English as a Foreign Language) learners, English writing is a particularly difficult process. They have error in words, syntax or semantics because their expressions are constrained by their mother tongue or by lacking of the awareness of vocabularies, collocation, or sentence-structures. In recent years, according to the development of corpus technique, we have rich resource in language. The resource can not only provide many statistical analyses of informations in language research but also can be used as a reference in the use of specific words for the learners. Current corpus technique use concordance which displays the words with their surrounding text to present sentences, but that approach provides very limited references in the writing process for ESL/EFL learners. Also considering the different inquired queries, if we have no well-defined input and retrieval module, the quantity and quality of results could not meet the demand for the learners. In addition, corpus technique doesn’t provide the ranking and evaluating skills to the retrieval results. As a result of that, learners often require filtering a large number of information to find something useful. In view of the above points, current corpus technique is unable to satisfy different degrees of ESL/EFL learners in the writing process. This paper presents a Sentence Recommendation Approach technique. With regard to the inadequate knowledge in the writing process for ESL/EFL learners, we search for useful sentences from the corpus and provide them to learners. Our Sentence Recommendation Approach technique has three modules: one is Expression Element Module. It allows the learners to use some expression elements to represent their demand. Another is Retrieval Module. It means the process of searching the corresponding sentences based on the expression elements in Expression Element Module. The last is Sort Module. It means the process of ranking the results derived from the Retrieval Module. Our research establishes the experimental system to verify the performance of our Sentence Recommendation Approach technique. We use British National Corpus and Science America Corpus for the source of sentences. The evaluation is divided into our objective evaluation and the questionnaire evaluation. Both of them prove that our experimental system does some favor in the writing process for ESL/EFL learners. In other words, our Sentence Recommendation Approach technique really helps the learners in the writing process.

本研究欲探討分析師推薦對管理當局所釋出資訊量之關聯性，當分析師越強力推薦公司時，公司的管理當局將願意提供較多的資訊給分析師做為預測的依據，則分析師對公司盈餘的預測也越準確。研究結果發現給予公司較佳推薦的分析師，預測準確性確實比給予公司較差推薦的分析師高。 之後再利用台灣證券暨期貨市場發展基金會設立的資訊揭露評鑑系統，探討資訊較為透明的公司，因為管理當局自願提供較多的資訊，即便分析師強力推薦，可能也無法得到額外的資訊，所以分析師推薦的效果應比資訊揭露較不透明的受評公司差。實證結果發現資訊揭露較透明的受評公司，分析師的推薦效果確實比資訊訊揭露較不透明的受評公司差。 / This thesis examines directly whether that managers provide more (less) information to analysts with more (less) favorable stock recommendations, based on the Barron et al. model (1998). Prior study documents the relative forecast accuracy of analysts before and after a recommendation issuance under the assumption that increases (decreases) in management-provided information will increase (decrease) analysts’ relative forecast accuracy. In contrast, this paper directly measure amount of information based on Barron et al. model (1998), and examine whether amount of information varies between pre- and post- a recommendation. Contrary to our prediction, the results show no significant difference in amount of information after and before recommendation issuance. However, we do find that analysts issuing more favorable recommendations experience a greater increase in their relative forecast accuracy compared with analysts with less favorable recommendations. In addition, we also find that the association is smaller for firms with higher information transparency than those with lower information transparency. The information transparency is measure by whether firms are listed in Taiwan Securities & Futures Information Center’s Information Disclosure and Transparence Ranking System (therefore TSFIC).

在目前網路和資訊科技迅速發展的環境底下，我們可對推薦系統做多方面的應用，而我們可知在一般家庭底下或多或少都會面臨一些狀況。為瞭解決大多數家庭都會面臨到的問題並省卻一般以人工來解決問題的方式，我們將提供一套線上解答系統並適時推薦有效答案給使用者。首先可利用FAQ的機制，讓使用者自行在系統上面尋找解決的方法和管道，提供一般的基本解答；然後結合推薦系統的功能，讓此問題的解答更加個人化。有別於一般的推薦系統只以個人為推薦單位，此系統將以家庭為單位的基礎底下實現推薦機制，以期能夠適時給予屬於整個家庭為核心之最適當的建議。並藉由使用者本身對系統之解答所做出的回饋來持續修正系統的準確度，以提高使用者的滿意度。 / Under current E-business environment, it is a great issue to keep customer's relationship by improving their satisfaction. On the other hand, it is usual to have some parent-children problems in families. Therefore, in order to assist to the problem-solving that most families would encounter, this research proposes an on-line system to recommend answers to users in appropriate time. The system would combine the mechanism of FAQ, which offer the general basic answer, and the function of recommender system to allow the personalized answers further. The proposed system is different from other recommender systems because we take the whole family as a unit, not just the looking-for-help parent or child. In addition, we would collect feedbacks provided by users who have applied the system answers. It is hoped to improve the user’s satisfactions, solve their family problems.

隨著全球資訊網的發展，人們享受了資訊快速流通的便利，也造就了搜尋引擎的發展。針對學術文獻，ACM, IEEE等學術組織也將學術文獻數位化，並提供關鍵字查詢文獻的功能。此外，Google也發展了Google Scholar搜尋全球資訊網上的學術文獻。Google在回傳查詢結果時，除了考慮文獻內容與查詢關鍵字的相似度之外，也利用PageRank技術來考量文獻間的引用關係。但是，有時後使用者想查詢的是與查詢相關的重要參考文獻。這些文獻的內容與查詢未必有很高的相似度。 因此本論文的研究目的在研究並發展推薦重要參考文獻的技術。我們先利用蜘蛛程式( spider)與剖析程式( parser)擷取分析ACM Digital Library上所收錄的論文後設資料，並解析出論文篇名、作者、摘要、關鍵字、分類、參考文獻等論文的重要組成要素。接著利用Mixed Media Graph（MMG）以描述關鍵字與參考文獻間關係的MMG 模型。當使用者輸入關鍵字，利用MMG做random walk因此可以找出與輸入關鍵字相關性最高的參考文獻。 / The rapid development of the Internet, people enjoy the rapid flow of information to facilitate, but also created a search engine of development. ACM and IEEE have developed the digital libraries to provide literature search. Moreover, there exist some search engines for academic literature, such as Google Scholar. Google Scholar collects academic literatures from WWW and provides users the capability to query literatures by keywords. However, sometimes what users need is to search for important citations specified by authors, such as seminal survey papers or books. The aim of this thesis is to investigate and develop the mechanism for search for important citations. In the developed mechanism, first the spider crawls and collects the literature from ACM Digital Library. Then the parser parse and extract the meta information for each literature. The Mixed Media Graph is employed to capture the relationships between keywords and citations. Given a set of query keywords, the important citations are generated by random walk over the constructed Mixed Media Graph. Performance analysis shows that the proposed mechanism performs well.

近年來，由於網際網路的興起，網際網路已成為新聞媒體重要的傳播管道之一，許多新聞網站如雨後春筍般的成立，而讀者也樂於使用這類更加便利、高互動性的新聞網站。但是媒體使用網路作為傳播管道，同時也面臨在傳統傳播模式所未遭遇的新挑戰，網路新聞媒體被迫需要創造獨特的內容吸引使用者，也需發展具黏性的社群經營服務，才能與其他具有類似社群互動機制的Web 2.0網站一較長短，留住廣大的使用者群。 本研究嘗試利用新聞為日常生活人們獲得資訊不可或缺管道的獨特優勢，提出一套有效利用新聞使用社群集體智慧(Collective Intelligence）機制，能夠自動化依據使用者顯隱性回饋，針對每篇新聞分析出分歧度、熱門度、話題性三個社群資訊，並以上述三個社群資訊挑選出合適的焦點新聞，以此促進新聞社群使用者對於焦點新聞的討論與互動，進而提昇新聞傳播的效益與新聞社群的凝聚力。實驗結果證實，本研究所提出的機制確實能夠探勘出滿足大多數使用者關注焦點新聞資訊的需求，並且對於輔助記者掌握讀者對於新聞資訊需求及促進新聞社群經營方面都有很大的助益。 / In recent years, due to the rise of the information and communication technologies, the internet has become one of most important communication channel for Journalism. A long with drastically flourished on-line Journalism, models of readers’ information reception changed while they are enjoyed more convenient and interactive websites providing instant information. At the same time, while mass media utilize internet as communication channel, it has also brought unprecedented challenge to traditional communication. On-line Journalism has not only need to create unique content (information) to attract readers; but it also need to develop a more engaging community management services to interact with other communities with similar mechanisms of Web 2.0 sites to retain user’s attention. This study attempts to exam the proposed on-line journalism system for University Press community, which could automatically analyze readers’ community dataset of University newspaper; including opinion deviation indicators, popularity indicator, and topicality indicator of each news (information). This system selects targeted news (information) according to above indicators to promote discussion and interactivity within readers’ community in hope to promote efficiency of news (information) communication and engagement within readers’ community. Experiment results reveal this proposed mechanism could satisfy most readers’ need for headline news; as well as assist Journalists’ understanding on their readers’ need while promoting on-line journalism social networking management.

近來，旅遊資訊廣被旅遊者在網路上使用。雖然網路上的資訊十分豐富，但是使用者仍常常難以找尋到精準的資訊。而旅遊商品的特性為無形的，所以使用者不能實際地來評估這個服務直到他實際地體驗之後。也就是因為此種特性，所以如何讓使用者在真正體驗到之前能夠取得可信與真實的旅遊資訊變得十分重要。為了解決此問題，語意網絡的概念即出現來解決人與電腦間溝通的問題。而一個本體即是由一個正式化的、某一具有精確規格概念的領域來提供之可實行的平台來發展可信的旅遊資訊服務。 在本論文中，我們探討了旅遊推薦系統的發展、其遭遇的問題、語意網相關之技術包含了：網路本體語言、資源描述架構、和一些目前現有的旅遊本體發展的情況。此外，為了要能提供更智慧化的旅遊行程規劃推薦服務，我們將語意的想法帶入了此領域。我們會提出一個方法讓智慧型旅遊行程推薦服務能在本體論的基礎上實現。所以，一系列的旅遊本體會被建構發展，來讓我們的芻形系統能夠做出行程推薦的服務。此提出的系統能夠驗證語意網的概念在旅遊推薦領域的可行性。它亦能利用屬性與之間的關係來推薦出更智慧型的資訊，找出個人化的景點、活動與行程給旅行者。 / Nowadays, travel information is increasing to appeal the tourists on the web. Al- though there are numerous information provided on the web, the user gets puzzled in nding accurate information. The tourism product has an intangible nature in that cus- tomers cannot physically evaluate the services on oer until practically experienced. This makes access to credible and authentic information about tourism products before the actual experience very valuable. In order to solve these problems, the concept of seman- tic web comes into existence to have communication between human and computer. An Ontology being a formal, explicit specication of concepts of a domain provides a viable platform for the development of credible tourism information services. In this paper, we discuss the development of travel recommendation system, the problems it encounters, the related technology about semantic web including OWL, RD- F/RDFS, and some current circumstances of the existing tourism ontologies as well. Futhermore, in order to make more intelligent travel planning recommendation services, we bring the idea of semantic into tourism domain. We will present an approach aimed at enabling intelligent recommendation services in tourism support systems using ontolo- gies. A suite of tourism ontologies was developed and engaged to enable a prototypical tourism system with recommednation capabilities. The proposed system can verify the feasibility and concept of taking semantic web technology into tourism recommendation systems domain. It also can recommend more intelligent information using properties, relationships of travel ontology, and is responsible for nding personalized attractions, activities and a trip itinerary for travelers.