Query Expansion Based On Web Search Results For Sponsored Search

The prevailing business model of Web search relies heavily on sponsored search, whereas a few carefully-selected paid advertisements are displayed alongside organic search results. However, the query is often very short and commonly appears a few wrongly written or mispronounced characters, and users, consciously or not, choose query terms intended to retrieve the best search results rather than the best ads. Moreover, the information in one ad is limited, the ads are often formulated as abrupt, non-grammatical phrases intended to capture reader’s attention rather than to facilitate query matching. Therefore, it’s very difficult to select relevant ads for the user’s query.To solve the problem in ad retrieval, more and more researchers apply query expansion technology to computational advertising through different ways. As query expansion methods based on ads dataset is limited, some researchers try to use web search results as query expansion resource, they use feature like TF, TFIDF and so on to select query expansion terms, although they achieve some performance, but there are many noise.On the basis of using web search results as query expansion external resources, to reduce the noise and improve the performance in ad retrieve, we first apply query expansion method based on term co-occurrence to ad query expansion. Traditional TF and TFIDF features mainly rely on term’s frequence, they can’t take into consideration the semantic information, while the method based on term co-occurrence could select more semantic relevant expansion terms. What’s more, the quality of web search results is different, and the importance of each field in one web search result is different, so we import a filed quality factor which can control the influence of the field’s quality for selecting expansion terms in the traditional Rocchio expansion model, and we propose an ad query expansion framework based on multi-field, this framework is very flexible, we can get the field’s quality through many ways and also we can express the feature in field through many ways.Experimental results on the real ad dataset show that the ad query expansion based on term co-occurrence method and based on multi-field method can be effectively to reduce the noise in candidate expansion terms and improve the ad retrieval effectiveness.