Identifying the libraries used by a web application is an important task for sales intelligence, website profiling, and web security analysis. Recent work uses tree structures to represent the property relationships of the library at runtime, realizing automatic library identification without pinpointing versions. But when assessing the security risks associated with these web libraries or conducting fine-grained software analysis, it becomes essential to determine the specific version of the library in use. However, existing tree-based methods are not directly applicable to version detection due to the huge storage requirements for maintaining separate trees for a large number of versions. This paper proposes a novel algorithm to find the most unique structure out of each tree in a forest so that the footprint of the features can be greatly minimized. We implement this algorithm into a web library detection tool. Experimental evaluations on 556 web libraries, encompassing 30,810 versions, reveal that our tool reduces space requirements by up to 99%, achieves more precise version detection compared to existing tools, and detects 190 vulnerabilities on 200 top-traffic websites.