近年來，網路廣告已成為最常用使用的市場通路。「搜尋贊助廣告」和「內容關聯廣告」是兩種最主要網路廣告類型。我們的研究主要針對「內容關聯廣告」類型。根據之前的文獻指出，「內容關聯廣告」面臨著是同義詞、多媒體內容、廣告中的文字數量有限，和顯示特性等四大問題。我們主要的研究目的是想要去提出一個新的「內容關聯廣告」的工具用來媒合廣告與網頁。我們提出的計畫工具主要的優點在於可以避免多媒體內容和文字數量有限問題。 在文獻上，目前用來配對廣告和網頁的現有方法主要有兩種類別。分別為向量空間模式和關鍵詞模式，但它們都各自有不同的缺點。像是向量空間模式的方法，它難以建立成文字向量對於一些非文字的廣告，因為這些廣告是多媒體型態而且擁有有限的文字。而對於關鍵詞模式，會造成大部分網頁的文字沒有被使用與利用，當我們如果只有單純只選擇使用幾個關鍵詞而非使用全部完整文字向量來表達一個網頁。 因廣告商對於他們的廣告相當了解，我們研究在這樣的假設之下，可藉由廣告商提供相關的關鍵詞標籤加在多媒體廣告上，把這樣的假設建立在我們提出推薦廣告給網頁的方法之上。我們結合了兩種傳統模式，將網頁文字轉換成文字向量，讓廣告商在多媒體廣告加上關鍵詞標籤。然後提供一個機制去計算文字向量和關鍵詞標籤的文字相似度。執行實驗與評估後，透過展示出我們方法的成果，發現比傳統資訊檢索工具好。 In recent years, web advertising has become one of the most commonly-used marketing channels. Sponsored search and contextual advertising are the two main categories of text-based web advertising. Our research focuses on the contextual advertising. As indicated in the literature, the contextual advertising suffers from the four major problems, including synonyms, multimedia content, limited amount of text in advertisement, and display characteristics. In this work, our aim is to propose a new contextual advertising method to match ads and web pages. The advantages of our proposed method are that it can avoid the problems caused by multimedia content and limited amount of text problems. In the literature, the methods to match ads and web pages can be categorized into two main categories, which are vector space model and keyword based model. Both approaches have their own different weaknesses. For vector space model approach, it is difficult to build vectors for non-text-based ads because these ads include mainly multimedia material and have only very limited text; for keyword based model approach, most information in the web page is not used since we only select few keywords rather than a full vector to represent a web page. Since advertisers understand their ads well, this work assumes that the multimedia ads would be associated with keyword tags provided by their advertisers, based on which we propose a new approach to recommend ads to web pages. In our approach, we combine the two traditional approaches and represent a web page by a term vector and represent an ad by the keyword tags proposed by its advertiser. Then, a matching mechanism is developed to compute the similarities between the vectors and keywords tags. An experiment and evaluation are carried out to demonstrate the performance of the proposed method. The results show that it performs better than traditional information-retrieval methods.
