Content discovery systems and methods

1. A computer implemented method comprising:

accessing a document;

generating a document vector in respect of the document;

generating a sentence vector for each sentence in the document;

calculating a sentence similarity score for each sentence in the document, the sentence similarity for a given sentence being calculated based on a similarity between the sentence vector for the given sentence and the document vector;

identifying one or more representative document sentences for inclusion in a document summary, the one or more representative document sentences being identified based on their sentence similarity scores.

2. The computer implemented method of claim 1 , further comprising:

generating a summary order in which the representative document sentences identified for inclusion in the summary should be presented, wherein the summary order is based on the order in which the identified sentences appear in the document.

3. The computer implemented method of claim 1 , wherein generating a sentence vector in respect of a given sentence comprises:

identifying relevant words in the given sentence;

retrieving, from a language model, word vectors in respect of each relevant word identified in the given sentence; and

summing the retrieved word vectors to generate the sentence vector.

4. The computer implemented method of claim 1 , wherein generating a sentence vector in respect of a given sentence comprises:

identifying relevant words in the given sentence;

calculating a weighted word vector for each word identified in the given sentence; and

summing the weighted word vectors to generate the sentence vector.

5. The computer implemented method of claim 4 , wherein calculating a weighted word vector for a given word comprises:

retrieving, from a language model, a word vector in respect of the given word;

applying a term frequency-inverse document frequency weighting to the retrieved word vector.

6. The computer implemented method of claim 1 , wherein prior to generating a sentence vector for each sentence in the document, the document is tokenized to identify the sentences in the document.

7. The computer implemented method of claim 1 , wherein generating a document vector in respect of the document comprises:

accessing the document;

tokenizing the document to identify document sentences and document words;

processing the tokenized document to generate an initial document vector; and

normalizing the initial vector to generate the document vector.

8. The computer implemented method of claim 7 , wherein processing the tokenized document to generate the initial document vector comprises:

retrieving, from a language model, word vectors in respect of each relevant word identified in the document; and

summing the retrieved word vectors to generate the initial document vector.

9. The computer implemented method of claim 7 , wherein processing the tokenized document to generate the initial document vector comprises:

calculating a weighted word vector for each relevant word identified in the document; and

summing the weighted word vectors to generate the initial document vector.

10. The computer implemented method of claim 9 , wherein calculating a weighted word vector for a given word comprises:

retrieving, from a language model, a word vector in respect of the given word;

retrieving, from a frequency model, a training set frequency in respect of the given word, the training set frequency in respect of the given word being the frequency of the given word in a training set of data;

calculating the frequency of the given word in the document; and

applying a term frequency-inverse document frequency weighting to the retrieved word vector, the term frequency being the training set frequency in respect of the given word and the document frequency being the frequency of the given word in the document.

11. A computer system comprising:

a processor;

a communication interface; and

a non-transitory computer-readable storage medium storing sequences of instructions, which when executed by the processor, cause the processor to implement a method comprising:

accessing a document;

generating a document vector in respect of the document;

generating a sentence vector for each sentence in the document;

calculating a sentence similarity score for each sentence in the document, the sentence similarity for a given sentence being calculated based on a similarity between the sentence vector for the given sentence and the document vector;

identifying one or more representative document sentences for inclusion in the summary, the one or more representative document sentences being identified based on their sentence similarity scores.

12. The computer system of claim 11 , wherein the sequences of instructions further cause the processor to implement the method further comprises:

generating a summary order in which the representative document sentences identified for inclusion in the summary should be presented, wherein the summary order is based on the order in which the identified sentences appear in the document.

13. The computer system of claim 11 , wherein generating a sentence vector in respect of a given sentence comprises:

identifying relevant words in the given sentence;

retrieving, from a language model, word vectors in respect of each relevant word identified in the given sentence; and

summing the retrieved word vectors to generate the sentence vector.

14. The computer system of claim 11 , wherein generating a sentence vector in respect of a given sentence comprises:

identifying relevant words in the given sentence;

calculating a weighted word vector for each word identified in the given sentence; and

summing the weighted word vectors to generate the sentence vector.

15. The computer system of claim 14 , wherein calculating a weighted word vector for a given word comprises:

retrieving, from a language model, a word vector in respect of the given word;

applying a term frequency-inverse document frequency weighting to the retrieved word vector.

16. The computer system of claim 11 , wherein prior to generating a sentence vector for each sentence in the document, the document is tokenized to identify the sentences in the document.

17. The computer system of claim 11 , wherein generating a document vector in respect of the document comprises:

accessing the document;

tokenizing the document to identify document sentences and document words;

processing the tokenized document to generate an initial document vector; and

normalizing the initial vector to generate the document vector.

18. The computer system of claim 17 , wherein processing the tokenized document to generate the initial document vector comprises:

retrieving, from a language model, word vectors in respect of each relevant word identified in the document; and

summing the retrieved word vectors to generate the initial document vector.

19. The computer system of claim 17 , wherein processing the tokenized document to generate the initial document vector comprises:

calculating a weighted word vector for each relevant word identified in the document; and

summing the weighted word vectors to generate the initial document vector.

20. The computer system of claim 19 , wherein calculating a weighted word vector for a given word comprises:

retrieving, from a language model, a word vector in respect of the given word;

retrieving, from a frequency model, a training set frequency in respect of the given word, the training set frequency in respect of the given word being the frequency of the given word in a training set of data;

calculating the frequency of the given word in the document; and

applying a term frequency-inverse document frequency weighting to the retrieved word vector, the term frequency being the training set frequency in respect of the given word and the document frequency being the frequency of the given word in the document.