Could please point out any approach that would be the best to achieve this ?

What have you tried? What was the result?

The approach I'd take would be to create a pseudo state machine that changes state based on input. If I was able to attach graphics to a PM post I'd give you a nice state diagram for that, but here goes with a text only version

        +------------------------------------+
        |                                   yes
        v                                    |
     start----->"TAGSTART" ---> intag ---> "TAGEND?" <------+
                                             |              |
                                           process input ---+

The bit "process input" you essentially read in a line, depending on the tag you just add it to a hash as appropriate.