Generation device, generation method, and non-transitory computer readable storage medium

A generation method which a computer executes, the method comprising:

• an acquisition step of acquiring content targeted for disclosure and policy information which indicates a disclosure policy of the content;
• a generation step of generating modified content in which a predetermined part of the content is modified based on the disclosure policy indicated by the policy information;
• an evaluation step of evaluating the modified content based on the disclosure policy to generate evaluation result;
• and
• a generation model learning step of performing reinforcement learning of a generation model based on the evaluation result, wherein
• the generation model is used by the generation step to generate the modified content;
• the evaluation step is performed using an evaluation model, wherein the evaluation model determines whether the modified content matches a context corresponding to the content, and the reinforcement learning of the generation model is performed when the modified content matches the context corresponding to the content;
• if a moving image is acquired as the content, the content is modified based on a sound associated with the moving image to generate the modified content;
• and
• if the sound is determined to be in a first language, the sound is modified to dub in a second language and the moving image is modified to correspond to a speech speed of the modified sound.

A non-transitory computer readable storage medium having stored therein a generation program which causes a computer to execute:

• an acquisition procedure of acquiring content targeted for disclosure and policy information which indicates a disclosure policy of the content;
• a generation procedure of generating modified content in which a predetermined part of the content is modified based on the disclosure policy indicated by the policy information;
• an evaluation procedure of evaluating the modified content based on the disclosure policy to generate evaluation result;
• and
• a generation model learning procedure of performing reinforcement learning of a generation model based on the evaluation result, wherein:
• the generation model is used by the generation procedure to generate the modified content;
• the evaluation procedure is performed using an evaluation model, wherein the evaluation model determines whether the modified content matches a context corresponding to the content, and the reinforcement learning of the generation model is performed when the modified content matches the context corresponding to the content;
• if a moving image is acquired as the content, the content is modified based on a sound associated with the moving image to generate the modified content;
• if the sound is determined to be in a first language, the sound is modified to dub in a second language and the moving image is modified to correspond to a speech speed of the modified sound;
• and
• if a part of the sound is determined to be inappropriate based on the disclosure policy indicated by the policy information, the content is modified to modify the part of the sound and a part of the moving image associated with the part of the sound.