Shorter data points are always ranked more influential then longer ones

In some cases I cannot normalize text length. Please add some factor for normalization here. In some cases I don't ant to analyze text influence token by token but want to analyze the entire data points influence as one block