This is a test for word2vec
Wed Nov 07 16:47:19 2018
dir of model1: ./model/window3_ min_count2_worker4_sg0_sess1105/size_80.model
dir of model2: ./model/window3_ min_count2_worker4_sg0_sess1105/size_110.model
80 110 150
绑定 1
关联 1
厨打 1
促销
支付 1
预约 1
导入 1
哪里
导入
关闭 1
size80 的效果出乎意料的好,可能也是考虑到我们目前的训练数据并不是特别多,
除了相似度高意外,很符合我们对近义词的要求,可以有效的解决歧义
厨打
====== model1 ======
[('厨房', 0.7792487144470215), ('KDS', 0.6969343423843384), ('厨房打印机', 0.6915861368179321), ('kds', 0.6875752210617065),
====== model2 ======
[('厨房', 0.7365704774856567), ('厨房打印机', 0.6782543063163757), ('总控', 0.6597431898117065), ('kds', 0.6522904634475708),
====== model3 ======
[('厨房', 0.7174404859542847), ('厨房打印机', 0.643281102180481), ('总控', 0.641669750213623), ('kds', 0.6321718692779541), ('后厨', 0.6275204420089722),
后台
====== model1 ======
[('云后台', 0.7980374693870544), ('前台', 0.7327364683151245), ('云端', 0.6401246190071106), ('后天', 0.6294926404953003)
[('云后台', 0.7991924285888672), ('前台', 0.6874397993087769), ('后天', 0.6474512815475464), ('云端', 0.6466808319091797),
[('云后台', 0.7783473134040833), ('后天', 0.6452266573905945), ('前台', 0.6173823475837708), ('云端', 0.5968232750892639),
size高有助于识别错别字,但是考虑到错别字出现的频率,如果出现的频率很高的话,可能也可以在低维就识别出来