Week 12-13 Sem 2


Things get really interesting now. Now that we have our topic summarization algorithm ready, we shall try to understand how to figure out the poem remixing part.

6.1            Selecting Lines based on user input

First and foremost we try to summarize the user input. After that we do the following

6.1.1    Input Message’s Super Sense Matching

We search the poems database with the highest ranked salient measures of the input super senses and we shortlist an initial bunch of lines.
6.1.2    Input Message’s Semantic Similarity Checking

Once the list is narrowed down we apply further drilling by letting only those lines stay which have specific topics pass a fitness function of semantic similarity with the user input’s specific topics. This test has the following factors

1.       Is the same word (similarity value =3) or Part of the same synset or other similar relations (similarity value =2).

2.       Hypernym Relatedness Factor (for more details look in the previous section)

3.       Gloss overlap factor (for more details look in the previous section)

Now we have a bunch of lines which are quite relevant to the poem.

6.2            Recombination Based Poem Generation Algorithm

Here we randomly pick 3 lines (considering dependencies if any) and then we combine them in all possible ways and see which produces a higher salient measure based on the input. If the measure passes the fitness test we leave that as the final output. Else this process is repeated for 20 times and the best output is shown.

6.3            Overview of the recombination based poem generation process

The following is the overview of the proposed recombination algorithm.