关于训练数据其中的行为类型

[XGodLike]

请问，我看了在taobao的训练数据中，数据得预处理，其中行为类型好像没有用到？是这样吗？另外时间数据，只是作为一个排序来使用的，并没有考虑行为之间的时间差是吗？

[nwf5d]

论文中写的是有点击和购买等行为，只使用了点击行为数据。按时间排序，取200个，预测最后一个。数据预处理中确实没有过滤行为类型。加购、关注、购买、点击都算点击？

Taobao Dataset3 is a collection of user behaviors from Taobao’s
recommender system [12]. The dataset contains several types of
user behaviors including click, purchase, etc. It contains user behavior sequences of about one million users. We take the click
behaviors for each user and sort them according to time in an attempt to construct the behavior sequence. Assuming there are T
behaviors of user u, we use the former T-1 clicked products as
features to predict whether users will click the T -th product. The
behavior sequence is truncated at length 200.

另：楼上最后一个问题另外时间数据，只是作为一个排序来使用的，并没有考虑行为之间的时间差是吗？你说对了，确实不考虑时间差，把用户所有历史合并到若干channel的embedding中，建议再去看看论文。

[shuDaoNan9]

论文中写的是有点击和购买等行为，只使用了点击行为数据。按时间排序，取200个，预测最后一个。数据预处理中确实没有过滤行为类型。加购、关注、购买、点击都算点击？

Taobao Dataset3 is a collection of user behaviors from Taobao’s
recommender system [12]. The dataset contains several types of
user behaviors including click, purchase, etc. It contains user behavior sequences of about one million users. We take the click
behaviors for each user and sort them according to time in an attempt to construct the behavior sequence. Assuming there are T
behaviors of user u, we use the former T-1 clicked products as
features to predict whether users will click the T -th product. The
behavior sequence is truncated at length 200.

另：楼上最后一个问题另外时间数据，只是作为一个排序来使用的，并没有考虑行为之间的时间差是吗？你说对了，确实不考虑时间差，把用户所有历史合并到若干channel的embedding中，建议再去看看论文。

各位大佬有开源的TF2版本的实现吗？？