2
$\begingroup$

What ML techniques can be used to determine relevance or context between sentences?

For example:

"I am going to the event at 9pm. Can you meet me there?"

Here, there is context between the two sentences - the person is indicating they will be at the event at 9pm, and is asking the recipient to meet them there.

Conversely:

"I am going to the event at 9pm. Can you meet me at school tomorrow?"

Based on this text alone, it appears as if the sentences represent two disjoint thoughts.

Additionally, although the examples above are of adjacent sentences, the context may exist across multiple sentences, or sentences that are not adjacent:

"I am going to the event at 9pm. I've heard it is fabulous. Can you meet me there?"

$\endgroup$
  • 1
    $\begingroup$ RNNs with attention or memory. Read the question answering literature: github.com/dapurv5/awesome-question-answering Welcome to the site! $\endgroup$ – Emre Nov 22 '17 at 4:27
  • $\begingroup$ I would like to not go directly on the path of deep learning networks but look at something called as anaphora resolution to understand the relation between sentences. $\endgroup$ – Nischal Hp Jan 1 '18 at 23:38
1
$\begingroup$

What you want to do is to extract the relation within the sentence and in this case beyond - so basically a cross-sentence relation extraction.

Ambiguity and the context of words is a big deal when doing this kind of task. There might be multiple ways to get there, but it also depends on your goal (production environment, research or just doing it for fun to get into NLP).

I would recommend you looking into Graph LSTMs (e.g. this Paper)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.