Output apparatus and non-transitory computer readable medium
An output apparatus includes a processor configured to receive an input word expressing a feature of a matter; and, by inputting the input word to a generation model trained on relation between a feature term extracted based on a descriptive text describing the matter and an associative text associated with the matter, the associative text being generated from the descriptive text describing the matter, output an associative text corresponding to the input word.
1. An output apparatus comprising:
a processor configured to
receive an input word expressing a feature of a matter;
by inputting the input word to a generation model trained on relation between a feature term extracted based on a descriptive text describing the matter and an associative text associated with the matter, the associative text being generated from the descriptive text describing the matter, output an associative text corresponding to the input word;
outputting impact of each word or phrase included in a descriptive text on an associative text through a different generation model including an attention mechanism; and
extract one or more words or phrases included in the descriptive text as one or more feature terms, the one or more words or phrases being selected in descending order of impact on an associative text.
2. The output apparatus according to claim 1 , wherein:
in a case where there is feature term information that is generated in a course of training the different generation model for generating the associative text from the descriptive text and that defines a feature term in the descriptive text, the processor is configured to refer to the feature term information and extract the feature term from the descriptive text, and
in a case where there is no feature term information, the processor is configured to calculate importance of each word or phrase included in the descriptive text from details of the descriptive text, and extract the feature term in accordance with the importance of each word or phrase.
3. The output apparatus according to claim 2 , wherein the processor is configured to determine input order of inputting one or more feature terms to the generation model in accordance with importance of the one or more feature terms or impact of the one or more feature terms on an associative text.
4. The output apparatus according to claim 2 , wherein, using an estimation model for outputting an estimated sentence that contains estimated details of a descriptive text from a feature term, the processor is configured to output an estimated sentence estimated from the input word.
5. The output apparatus according to claim 2 , wherein the processor is configured to refer to total information that totals a number of appearances of every word or phrase used in a descriptive text including the input word; calculate, for each of a plurality of associative texts obtained in a case where a combination of the input word and a word or a phrase other than the input word is input to the generation model, similarity with a reference associative text generated in a case where a combination of the input word and a word or a phrase with a greatest number of appearances is input to the generation model; and, using the similarity with the reference associative text, select a related word related to the input word received from a user.
6. The output apparatus according to claim 1 , wherein the processor is configured to correct the impact on association in accordance with a user's command.
7. The output apparatus according to claim 6 , wherein the processor is configured to determine input order of inputting one or more feature terms to the generation model in accordance with importance of the one or more feature terms or impact of the one or more feature terms on an associative text.
8. The output apparatus according to claim 6 , wherein, using an estimation model for outputting an estimated sentence that contains estimated details of a descriptive text from a feature term, the processor is configured to output an estimated sentence estimated from the input word.
9. The output apparatus according to claim 1 , wherein the processor is configured to extract, as a feature term, a combination of words or phrases causing a smaller loss indicating an error between an associative text generated by the generation model and an associative text related in advance to the descriptive text, the loss being obtained in a course of training the generation model trained to generate, for an input feature term, an associative text related in advance to the descriptive text.
10. The output apparatus according to claim 9 , wherein:
the generation model includes a decoder, and
the processor is configured to extract, as a feature term, a combination of a word or a phrase included in the descriptive text, and, among words or phrases included in the decoder, a word or a phrase that is not included in the descriptive text.
11. The output apparatus according to claim 1 , wherein the processor is configured to determine input order of inputting one or more feature terms to the generation model in accordance with importance of the one or more feature terms or impact of the one or more feature terms on an associative text.
12. The output apparatus according to claim 11 , wherein the processor is configured to input one or more feature terms to the generation model in descending order of importance of the one or more feature terms or impact of the one or more feature terms on an associative text.
13. The output apparatus according to claim 1 , wherein, using an estimation model for outputting an estimated sentence that contains estimated details of a descriptive text from a feature term, the processor is configured to output an estimated sentence estimated from the input word.
14. The output apparatus according to claim 13 , wherein:
the processor is configured to output an estimated sentence for every input word received from a user, and
by inputting an input word corresponding to an estimated sentence selected by the user from among a plurality of the estimated sentences to the generation model, the processor is configured to output an associative text generated for the input word.
15. The output apparatus according to claim 1 , wherein the processor is configured to refer to total information that totals a number of appearances of every word or phrase used in a descriptive text including the input word; calculate, for each of a plurality of associative texts obtained in a case where a combination of the input word and a word or a phrase other than the input word is input to the generation model, similarity with a reference associative text generated in a case where a combination of the input word and a word or a phrase with a greatest number of appearances is input to the generation model; and, using the similarity with the reference associative text, select a related word related to the input word received from a user.
16. The output apparatus according to claim 15 , wherein the processor is configured to select a predetennined number of associative texts in ascending order of similarity with the reference associative text, and output, as one or more related words, one or more words or phrases other than the input word that correspond to the selected associative text or texts.
17. A non-transitory computer readable medium storing an output program causing a computer to execute a process, the process comprising:
receiving an input word expressing a feature of a matter;
by inputting the input word to a generation model trained on relation between a feature term extracted based on a descriptive text describing the matter and an associative text associated with the matter, the associative text being generated from the descriptive text describing the matter, outputting an associative text corresponding to the input word;
outputting impact of each word or phrase included in a descriptive text on an associative text through a different generation model including an attention mechanism; and
extracting one or more words or phrases included in the descriptive text as one or more feature terms, the one or more words or phrases being selected in descending order of impact on an associative text.
18. An output apparatus comprising:
a processor configured to
receive an input word expressing a feature of a matter;
by inputting the input word to a generation model trained on relation between a feature term extracted based on a descriptive text describing the matter and an associative text associated with the matter, the associative text being generated from the descriptive text describing the matter, output an associative text corresponding to the input word; and
outputting a warning in a case where a feature of the input word satisfies a predetermined condition,
wherein the predetermined condition comprises a similarity between the input word with an already-input word being greater than or equal to a reference similarity, a similarity between catch-phrases generated by the generation model being greater than or equal to a reference similarity, or an unextracted word being contained.