Text generation apparatus, text generation learning apparatus, text generation method, text generation learning method and program
A text generation apparatus includes a content selection unit that acquires a reference text based on an input text and information different from the input text and a generation unit that generates a text based on the input text and the reference text, wherein the content selection unit and the generation unit are neural networks based on learned parameters, so that information to be considered when generating a text can be added as text.
1. A text generation apparatus comprising:
a memory; and
a processor configured to:
determine, based on a predetermined measure, a score associated with each word in an input text using a first supervised learning model;
extract a first set of words from the input text based on the score and generate a reference text based on the extracted first set of words; and
generate the output text based on the input text and the reference text using a second supervised learning model,
wherein the output text represents a summary text of the input text based at least on a part of words in the reference text,
the second supervised learning model includes an input text encoder and a reference text encoder, and
vector representations encoded by the input text encoder and the reference text encoder are cross-referenced in a multi-layered network.
2. The text generation apparatus according to claim 1 , wherein the second supervised learning model includes an input text encoding and a reference text encoding as distinct text encodings cross-referencing each other in a multi-layered network.
3. The text generation apparatus according to claim 1 , wherein the processor is configured to be connected with an external knowledge source including a plurality of texts;
determine, based on the predetermined measure, a knowledge-source word score associated with either each word in the plurality of texts or each of the plurality of texts, or both, using the first supervised learning model;
extract, based on the knowledge-source word score, one or more texts from the plurality of texts;
generate the output text based on the input text and the extracted one or more texts using the second supervised learning model.
4. A text generation training apparatus comprising:
the text generation apparatus according to claim 1 ,
wherein the second supervised learning model is trained using the generated output text as training data.
5. A non-transitory computer-readable recording medium having computer-readable instructions stored thereon, which when executed, cause a computer including a memory and a processor to execute processing as the text generation apparatus according to claim 1 .
6. A text generation apparatus comprising:
a memory; and
a processor configured to:
extract, based on a score of each word in an input text and a length of output text to be generated, a set of words from the input text, the score being determined based on the predetermined measure using a first supervised learning model; and
generate the output text,
wherein the output text is output by a second supervised learning model receiving the input text and the extracted set of words as input,
the second supervised learning model is trained on training data,
the training data includes a plurality of words in the input text, the extracted set of words, and corresponding output text for training,
the second supervised learning model includes an input text encoder and a reference text encoder, and
vector representations encoded by the input text encoder and the reference text encoder are cross-referenced in a multi-layered network.
7. The text generation apparatus according to claim 6 , wherein the processor is configured to be connected with an external knowledge source including a plurality of texts;
determine, based on the predetermined measure, a knowledge-source word score associated with either each word in the plurality of texts or each of the plurality of texts, or both, using the first supervised learning model;
extract, based on the knowledge-source word score, one or more texts from the plurality of texts;
generate the output text based on the input text and the extracted one or more texts using the second supervised learning model.
8. A text generation training apparatus comprising:
the text generation apparatus according to claim 6 ,
wherein the second supervised learning model is configured to be trained using the generated output text as training data.
9. A non-transitory computer-readable recording medium having computer-readable instructions stored thereon, which when executed, cause a computer including a memory and a processor to execute processing as the text generation apparatus according to claim 6 .
10. A text generation apparatus comprising:
a memory; and
a processor to:
determine, based on a predetermined measure, a knowledge-source word score associated with either each word in a plurality of texts or each of the plurality of texts, or both, using a first supervised learning model;
extract, based on the knowledge-source word score, one or more texts from the plurality of texts;
generate the output text,
wherein the output text is output by a second supervised learning model receiving the input text and the extracted one or more texts as input,
the second supervised learning model is trained on training data,
the training data includes the input text, the extracted one or more texts, and corresponding output text for training,
the second supervised learning model includes an input text encoder and a reference text encoder, and
vector representations encoded by the input text encoder and the reference text encoder are cross-referenced in a multi-layered network.
11. A text generation training apparatus comprising:
the text generation apparatus according to claim 10 ,
wherein the second supervised learning model is trained using the generated output text as training data.
12. A non-transitory computer-readable recording medium having computer-readable instructions stored thereon, which when executed, cause a computer including a memory and a processor to execute processing as the text generation apparatus according to claim 10 .
13. A text generation apparatus comprising:
a memory; and
a processor configured to:
extract, based on a score of each word in an input text and a length of output text to be generated, a set of words from the input text, the score being determined based on the predetermined measure using a first supervised learning model; and
generate the output text,
wherein the output text is output by a second supervised learning model receiving the extracted set of words as input,
the second supervised learning model is trained on training data,
the training data includes the extracted set of words and corresponding output text for training,
wherein the second supervised learning model includes an input text encoder and a reference text encoder, and
vector representations encoded by the input text encoder and the reference text encoder are cross-referenced in a multi-layered network.
14. The text generation apparatus according to claim 13 , wherein the processor is configured to be connected with an external knowledge source including a plurality of texts;
determine, based on the predetermined measure, a knowledge-source word score associated with either each word in the plurality of texts or each of the plurality of texts, or both, using a first supervised learning model;
extract, based on the knowledge-source word score, one or more texts from the plurality of texts;
generate the output text based on the extracted set of words and the extracted one or more texts using the supervised learning model.
15. A text generation training apparatus comprising:
the text generation apparatus according to claim 13 ,
wherein the second supervised learning model is configured to be trained using the generated output text as training data.
16. A non-transitory computer-readable recording medium having computer-readable instructions stored thereon, which when executed, cause a computer including a memory and a processor to execute processing as the text generation apparatus according to claim 13 .
17. A method, comprising:
extracting, based on a score of each word in an input text and a length of output text to be generated, a set of words from the input text, the score being determined based on the predetermined measure using a first supervised learning model;
encoding, by using a second supervised learning model, the input text into a first vector representation of the input text;
encoding a reference text into a second vector representation of the reference text; and
generating, based on the first vector representation of the input text and the second vector representation of the reference text, the output text,
wherein the output text is output by the second supervised learning model receiving the input text and the extracted set of words as input,
the second supervised learning model is trained on training data,
the training data includes a plurality of words in the input text, the extracted set of words, and corresponding output text for training,
the second supervised learning model includes an input text encoder and a reference text encoder, and
respective vector representations encoded by the input text encoder and the reference text encoder are cross-referenced in a multi-layered network.
18. A method, comprising:
determining, based on a predetermined measure, a knowledge-source word score associated with either each word in a plurality of texts or each of the plurality of texts, or both, using a first supervised learning model;
extracting, based on the knowledge-source word score, one or more texts from the plurality of texts;
encoding, by using a second supervised learning model, the input text into a first vector representation of the input text;
encoding a reference text into a second vector representation of the reference text; and
generating, based on the first vector representation of the input text and the second vector representation of the reference text, the output text,
wherein the output text is output by the second supervised learning model receiving the input text and the extracted one or more texts as input,
the second supervised learning model is trained on training data, and
the training data includes the input text, the extracted one or more texts, and corresponding output text for training,
the second supervised learning model includes an input text encoder and a reference text encoder, and
vector representations encoded by the input text encoder and the reference text encoder are cross-referenced in a multi-layered network.