A probabilistic model for semantic advertising

A Knowledge-based Model for Semantic Oriented Contextual Advertising

“…The goal of contextual advertising is to help advertisers reach consumers with a strong preference for their products [2,7,9]. In this context, contextual advertising systems attempt to automatically accomplish this task through the placement of the most relevant commercial textual ads within the content of generic publisher Webpages [10,11].…”

“…As stated by Kumar et al [11], contextual advertising employs automated techniques that display advertisers' ads that are relevant to the user's interests as well as publisher website's content. Several approaches have been proposed to realize contextual advertising in practical and real-world application domains [7][8][9][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27]. The authors of [1], [7] and [13] have exploited knowledge encoded in Wikipedia to interpret the textual content of ads.…”

“…Hence, the precision of ad matching techniques degrades due to these reasons. Other researchers propose to exploit text summarization and probabilistic techniques to improve the matching process between ads and their corresponding Webpages [9,15,25]. Using text summarization techniques, text fragments of Webpages are summarized into short textual descriptions to minimize the matching space and reduce semantic dimensionality problems [15,25].…”

“…However, these approaches rely on using text processing algorithms that ignore the semantic aspects of the extracted texts and therefore, result in producing low-precision matching results. As far as probabilistic models are concerned, their performance is limited due to their dependence on other resources such as existing search engines as demonstrated in the context of the work proposed in [9]. As stated by the authors, using search engines such as Google and Baidu to measure the relatedness between terms is a limiting factor to their proposed approach since if the search engine performs poor for some terms, their proposed method may produce poor results as well.…”

“…This dataset is composed of 93 documents that we used to represent Webpages and a collection of 11429 textual segments that we used to represent our ads collection. It is important to point out that instead of utilizing web crawling techniques to collect Webpages in the same manner as performed in [9], we decided to use the documents in this dataset since they are available online and can be utilized for the same purpose by other researches in the field. In addition, the language used to express documents in this dataset is English, and since our exploited ontologies and semantic similarity techniques are applied on English words, we were not able to reuse datasets in languages other than this language.…”