With the booming of online business,the recognition of product entity has been widely applied in Business Intelligence.In our paper,we use the CRF model and select a series of lexical,syntactic and semantics features as feature space.In addition we use cross evaluation methods to evaluate the classifier's performance and also guide previous feature selection.In corpus construction step,we adopt two different labeling strategies,which we analyze in the evaluation section.The Experiment has achieved satisfactory result.By comparison with max entropy model,we further demonstrate the efficiency of CRF model we use in this experiment.%随着互联网经济的飞速发展,信息抽取领域的产品命名实体识别在商务智能领域有着广泛的应用。本文采用条件随机场(CRF)模型,选取词汇、词法和词形上一系列的特征进行训练,通过交叉验证对识别效果进行评价,并通过识别效果指导特征的选取。实验中比较了两种标注方式(BRAND/TYPE和PROD),并取得了令人满意的识别效果。在与最大熵模型对比中,验证了CRF模型对于产品实体识别的优越性。
展开▼