利用眼动轨迹包含的参数来预测顾客是否喜欢某些商品,对于电子商务的推荐系统具有重要意义。通过采集顾客观察某物品的眼动参数,借鉴Finds算法概念学习的思想,提出眼动轨迹语义提取算法,该算法首先学习先验知识,然后通过让样例正反例距离最大实现确定眼动参数包括注视时间、瞳孔大小、眨眼次数以及回视次数的权重,利用SEBET(Semantic extraction based eye tracking)算法,通过di=SX(∑()KF(∑(m′=k1,i′=k2,i=λm′=1,i′=1,i=0)(xNm′(i)-xoi′(i))2w(i)KF)k1×k2SX)计算样例正反例之间的距离,依照距离的远近来判断顾客是否喜欢某商品,从而实现从眼动轨迹进行语义提取。实验中,记录被试观察水果图片的眼动参数,分析出被试的喜好,与被试实际喜好进行了比较,发现样本与正例的距离为0.91,与反例的距离为3.01,与实际情况相符。
It is important for recommendation system of E-business to predicate consumer's habits by eyes gaze tracking. We explore algorithms to identify people's enjoyment by referencing Finds algorithm, it includes getting the weight of eyes parameters by get- ting the distance of object examples and negative examples, eyes parameters include the eye gazing time, the pupil size, the blink times and the looking back times. We use SEBET (Semantic Extraction Based Eye Tracking) algorithm to calculate the distance of object examples and negative examples, so we can decide whether consumers enjoy the goods or not from the distance, we extract the emotion semantic from eve tracking successfully. Exoeriments show the efficiency of our algorithm.