Research On Vision-Based Contextual Advertisement Keyword Retrieval Algorithm

Internet has became the most important advertising medium, its unique ability to display the various goods and services to all corners of the worlds with low-cost has attracted a large number of network advertising investment, and at the same, has boosted the development of the internet.A few years ago, because of the bankruptcy of many internet companies, there once had been a serious impact on online advertising investment. Such had been the situation until 2002 when it got alleviated. Incidentally, it was the emergence of a new form of advertising — search engine advertising, that had promoted the development of the online advertising.Up until now most search engine development has gone on at companies with little publication of technical details, our goal is trying to explore the details of such system and improve such systems in some aspect.In the first part, we introduce our work on the research of a new web page title retrieval algorithm. The algorithm combines the use of a web page title pattern extraction method and many visual features to create a model to extract the real web page title.In the second part, we introduce our work on the research of a new algorithm of web page main-text extraction. First, we use VIPS to segment the web page to get a tree of vision blocks. Then we use our algorithm introduced in the first part to extract the page title. In the last, we search the blocks which contain the title in the vision-block tree to obtain the main-text part of the web page.In the third part, we focus on our work on the extraction of keywords from web pages. We aim to find out a method to extract the keywords to maximize the semantic association between the website and advertisements.In the last part, we introduce the architecture of our contextual advertising system in details.