Service description file

which is fundamental for the service-oriented architecture

can be used by the IoT sensing devices to accelerate resource discovery and searching processes.Currently

these files are mostly written manually by the device developers

this process is inefficient and fallible.The state-of-art method SPITFIRE can generate the devices' description in a semi-automatic way

but its configuration can be trivial and its accuracy still can be improved.In this paper

we proposed a novel automatic device description method

with which devices can automatically generate their individual description.We designed a clustering algorithm based on DBSCAN to infer the description of sensing device

taking advantages of existing descriptions and the data features of series gained by data sampling.A metric learning algorithm is also implemented to optimize the parameter used by the clustering algorithm.All the routines run independently on different devices

and no manual intervention is needed during the self-description process.Through simulation

we show that this method has a prominent advantage of precision over other state-of-art methods

making our method more suitable for the massive IoT devices.