There are a number of HTML documents which include metadata on the Web and a number of information services which provide metadata embedded using metadata standards across domains. Those metadata are, however, encoded in various different schemas and in different serialization formats, which makes it hard to automatically extract and interpret the metadata. The primary reason of the difficulty is the lack of interpretation rules of the metadata, e.g., lack of definition of metadata vocabularies, lack of definition of encoding syntax and so forth. This paper proposes a model to support interpretation of embedded metadata without formal schema by linking a metadata instance to DCMI Description Set Profiles (DSP). An XPath expression addresses a metadata instance encoded in HTML, and DSP define metadata schema. We propose extending DSP to include XPath for linking a metadata instance to a metadata schema. This paper also shows an experimental system which extracts metadata using extended DSP.