基于网页查询结果的广告查询扩展研究

发布时间：2018-10-05 19:50

【摘要】：在计算广告学中,为用户查询返回相关的广告一直是研究的热点。然而用户的查询一般比较简短,广告的表示也局限在简短的创意和一些竞价词上,返回符合用户查询意图的广告十分困难。为了解决这个问题,该文提出利用多特征融合的方法进行广告查询扩展,先将查询输入到搜索引擎中,获得Top-k网页查询结果,将它们作为获取扩展词的外部资源,由于采用一般的特征选取方法获取扩展词采用的特征比较单一,缺乏语义信息,容易产生主题漂移现象,该文通过计算扩展词和查询词在网页查询结果中的共现度,并融合传统的TF特征和词性信息,获得与原始查询语义相关的扩展词。在真实的广告语料上的实验结果显示,基于多特征融合的选择广告扩展词的方法能有效地提高返回广告的相关性。
[Abstract]:In computational advertising, returning related advertisements for user queries has been a hot topic. However, the query of the user is usually brief, and the expression of the advertisement is limited to the brief idea and some bidding words, so it is very difficult to return the advertisement which accords with the purpose of the user's query. In order to solve this problem, this paper proposes a multi-feature fusion method for advertising query expansion. First, the query is input into the search engine to obtain the query results of Top-k pages, which are regarded as the external resources to obtain the extension words. Because the general feature selection method is simple and lacks semantic information, this paper calculates the co-occurrence degree of extended words and query words in the query results of web pages. The extended words related to the original query semantics are obtained by combining the traditional TF features and part of speech information. The experimental results on the real advertising materials show that the method based on multi-feature fusion can effectively improve the relevance of the returned ads.
【作者单位】：大连理工大学信息检索研究室;
【基金】：国家自然科学基金资助项目(60673039,60973068) 国家863高科技计划资助项目(2006AA01Z151) 教育部留学回国人员科研启动基金高等学校博士学科点专项科研基金资助(20090041110002)
【分类号】：TP391.3

【共引文献】