AFAIK, you're looking at the wrong approach anyway; RTDs are better, and thermocouples slightly moreso, for extremely stable temperature measurements (mK to uK). Which figures -- they're less sensitive, but less noisy; but the overall performance is better than a thermistor (which contributes the inherent noise of the ceramic material used).
But anyway, that doesn't account for changes. Sounds like EMC. Crappy front end? No RF filtering? Common mode noise generated by the logger?
Anything else connected to the logger, like a power supply? A switching supply will kick out more noise when producing more power.
Simplest thing to try, place a cap at the input pin to GND, maybe 0.01uF.
Tim