Title generation model training method and device

• Main Authors: LUO PENGSHUAI, ZHOU JINGUO, DONG DEJUN, DI PENG
• Format: Patent
• Language: Chinese; English
• Published: 05.04.2024
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The embodiment of the invention provides a title generation model training method and device.The title generation model training method comprises the steps that code merging data of a code merging request are obtained, and when the data length of the code merging data is larger than the preset data length, the title generation model is trained; compressing at least one of a code description field, a code branch field and a code difference field contained in the code merging data according to the compression priority to obtain compressed code data, constructing a training sample based on the compressed code data, and constructing a sample label according to title information of the code merging request, and training the pre-training model according to the training sample and the sample label to obtain a title generation model. 本说明书实施例提供了标题生成模型训练方法及装置，其中，一种标题生成模型训练方法包括：获取代码合并请求的代码合并数据，在代码合并数据的数据长度大于预设数据长度的情况下，按照压缩优先级对代码合并数据包含的代码描述字段、代码分支字段和代码差异字段至少一者进行压缩处理，获得压缩代码数据，基于压缩代码数据构建训练样本，并根据代码合并请求的标题信息构建样本标签，根据训练样本和样本标