A method is presented to identify some pieces of specific information in multi-carrier data streams byfeature words and based on PinYin matching. An effective knowledge approximation method is used to judge therelation between feature words and context by statistics theory. The part of speech transfer-value as systemknowledge can be obtained by inductive learning of training corpus. When data streams are evaluated, theevaluation value can be gained according to the system knowledge by matching all feature words and based on theirPin Yin, which examines the comparability with context regular of part of speech between all feature words in datastreams and themselves in training corpus. Further more, if the evaluation value exceeds the threshold, the datastreams will be shielded. Experimental results show that the effect of the experiment system based on this method isefficient for identifying ill information and monitoring & controlling their spreading by multi-carrier data streams.