The Research Of Web Advertising Model Based On Semantic Annotation

With the rapid development of the Internet, Web Advertising has been rapid development. Compared with traditional media advertising, Web Advertising through multi-media, all-weather, global exposure, with low-cost, highly interactive features. Web Advertising favored by more and more advertisers, and became the research subject of many scholars.In recent years, both in the commercial application and research field, Web Advertising has been a great development. However with the explosion of Internet Advertising delivery the click through rate (CTR) of the banner is falling. The prospect of the Web Advertising has been questioned. In order to enhance the efficiency of Web Advertising, how to reach the best match among target page, current user and web ads becomes the focus of the industry and research field.In this paper, the best match is defined as followings, the content of target page and web ads are related, at same time, the content of web ads are match with the interest of current user to some extent. According to the above definition, in order to resolving the mismatch problem among ad entities, we firstly proposed a formalization of the ad loading procedure, and then used a semantic based method to resolve the mismatch problem. Additionally, a system model is created in order to capture the whole web ads loading procedure, so we can analyze and research the key algorithms, including the semantic associations based web ads ranking algorithm and the user interest based web ads ranking algorithm.First, a semantic annotation based algorithm is proposed to solve the match problem during the loading process of web ads. The detail solutions are as followings:At first, after target web page, user interest and web page ad are annotated using semantic and their semantic features are also retrieved. Then, retrieve the similarity features between target page and web ads, and rank them by the similarity features, now we can get the frist round ads ranking result. At last, retrieve the similarity features between user and web ads, and re-rank the first round ad ranking result by using the similarity features. At whole, the semantic association based algorithm can convert the similarity match problem of ad entities to text semantic similarity ranking problem, so mature text processing techniques can be used to analyze the similarity among the ad entities.Second, a formalization and optimization model for web page content based ad ranking algorithm is proposed in order to solve the similarity match problem under a uniform formalization model. In this phase, the mainly focus is to research the similarity match problem between target page and web ads. It will be devided into two steps:at first, retrieve the similarity features between target page and ads. Here the difference between our proposed algorithm and others is that our proposed algorithm not only uses conventional feature based on VSM and semantic association features, but also the newly introduced statistical match features and hidden topic match features. Then, considering the problem of bad efficiency using only one feature, a RSVM based web ad ranking algorithm is proposed in order to get amalgamation of all features.Third, in the affect of user interest amalgamation, web ad re-ranking algorithm is proposed to implement the personalized web advertising. After analysis of the reading habit of user, cluster interest is used to represent the stable interest of the user in this paper. Here, cluster interest is a combination of the centroid vector and user interest measure, so it can solve the computation problem of the similarity during the amalgamation process. In order to reach a similarity match of all the ad entities, a user interest based re-ranking algorithm is proposed rather than ranking solely based on user interest.Finally, we proposed and implemented a semantic assocication based web advertising system. By using the system modeling techniques, the proposed algorithms about ad entities semantic annotation and web page ad ranking are deployed and implemented systematically. From the system aspect, the dependent relations among all the part are taken into consideration, and the solutions for integration existing system are also included. By using hierarchical techniques, the system model is well designed, and also the whole system is fully evaluated in test and application environments.From the above, our research on the related problems of web advertising model based on semantic annotation is a beneficial step to apply NLP techniques in computational advertising field. It will helpful to make high level consistent amony target page, user interest and web ads. And it has significance for search engine optimization and web information retrieval.