一种无位置偏见的广告协同推荐算法

An Advertisement Collaborative Recommendation Algorithm Without Position Bias

在广告推荐系统中,页面与广告的相关性是用户是否点击广告的重要因素,一般利用点击率计算相关性,但是广告展示位置的不同会影响页面-广告相关性计算的准确性,从而导致相关性低的广告被当成相关性高的广告进行错误推荐。针对该问题,提出一种无位置偏见的广告协同推荐算法。利用贝叶斯定理改进位置模型,排除历史数据中的位置影响,计算页面-广告相关性。通过协同过滤技术,为页面找到与其相似的其他邻居页面,实现准确的广告推荐。在腾讯搜搜广告日志数据上进行实验,结果表明,与传统协同过滤算法相比,该算法的推荐准确率、召回率以及F度量值均提高了40%以上,具有较好的广告推荐效果。

In an advertisement recommendation system,an advertisement which has a low relevance with the query has a high Click-through Rate（ CTR） because of the high position where it is shown. Advertisement-query relevance which attracts users to click advertisement plays an important role at advertisement recommendation,and a wrong relevance,for example, CTR with position bias, may lead to a bad recommendation to user and loss of interest. Aiming at these problems,this paper proposes an advertisement collaborative recommendation algorithm without position bias. It finds similar page with other neighbors page by collaborative filtering technology to realize accurate advertising recommended. An experiment on tencent soso data about advertisement logs shows that this algorithm has a better recommendation result （at least 40% higher） of higher recall,precision and F-measure than baseline traditional Collaborative Filtering（CF） algorithm without the effect of position bias,and it has good advertising recommendation effect.