Self-adaptive theme story ending generation method and storage medium

Abstract

The invention provides a self-adaptive theme story ending generation method and a storage medium. The method comprises the following steps: initializing a story ending generation model; randomly sampling data of a specific theme of a theme, and dividing the data into a training set and a test set; generating theme characteristic parameters corresponding to the specific theme of the theme through a parameter generator; combining the theme feature parameters with theme sharing parameters extracted from a story ending generation model through a theme adapter, and generating story ending generation model parameters corresponding to the specific theme of the theme; and generating a story end generation model parameter for each topic-specific topic. The unified model offset parameter related to the theme is generated and acts on the theme sharing parameter part in the unified model, so that the unified model can integrate theme specific guidance generated by the meta-learning framework while reserving theme sharing knowledge, the adaptation of the unified model to the target theme is accelerated, and finally, the self-adaptive story ending generation of the low-resource theme is realized.

Description

technical field [0001] The invention relates to the field of task-oriented story ending generation, in particular to an adaptive theme story ending generation method and a storage medium. Background technique [0002] Among task-oriented story generation tasks, story ending generation is a popular task. Its purpose is to give a story text and complete an ending for it. How to produce a coherent and logical ending to the preamble of the story is the key to research. For this task, the existing well-labeled data sets are very limited and expensive. Therefore, how to use the rich data of high-resource topic stories to enhance the performance of the system on low-resource topics and link stories of different topics is an important issue in academia and industry. research questions of great concern. [0003] In related art, previous research attempts to introduce external knowledge to guide the system to encode text and discover hidden information beneath the surface of the te...

AI Technical Summary

Problems solved by technology

However, when the meta-learning framework is combined with the story ending generation system, the quality of the topic-sharing parameters learned in the training phase is closely related to, or even limited by, the distribution of source topic data used for training.

At the same time, in the test phase, the meta-learning framework can only be effectively adapted to similar topics, and it is still difficult to improve the performance of dissimilar target topic models.

Images