Find importance score for each entity using BERT

Develop a python Class "NodesBertScoring"

Input: output of https://github.com/korobool/MM-demo/issues/6

.all_nps() > [(6,7),...] .all_ners() > [(9,9),...] .all_vps() > [(1,2),...] .tokenized() > ['I', 'love', 'you', 'so', 'much', 'my', 'beautiful', 'life', 'in', 'US', '.' ]

Output: Sorted by 'score' list of tuples:

[ ('text of node1', 0.013, 'np'), ('text of node2', 0.004, 'ner'), ... ]

Proposals: Hypothesis: if element (words) of any node are simply predicted, the meaning importance weight of this phrase is LOW because doesn't contain any surprising meaning put hear by author

Different strategies are possible:

individual predicting of each word of phrase causes final score as (linear?) combination
predicting of a whole phrase

korobool / MM-demo

Find importance score for each entity using BERT #7