Search4Code

Search4Code数据集由真实世界的用户查询和相应的最常点击的url组成。每个查询还具有一个标签，该标签表示查询是否具有代码搜索意图，如本文所述的弱监督模型所预测的那样。 目前，Search4Code数据集包含6596个Java查询和4974个C # 查询。 以下是Search4Code数据集的一些优势: 真实查询: 查询是从匿名的Bing搜索日志中采样的。我们相信这提供了个人如何搜索代码的现实表示。 单击url: 每个查询都有一个列表，列出了搜索结果中最常单击的三个url。对于给定的搜索查询，最常单击的url代表提供令人满意的解决方案的搜索结果。 人气评分: 每个查询都会根据出现的频率分配一个人气等级。 大规模: 数据集包含数千个查询。因此，可以使用需要大量训练数据的更复杂的模型。The Search4Code dataset comprises real-world user queries paired with their corresponding most frequently clicked URLs. Each query is additionally annotated with a label indicating whether the query carries a code search intent, as predicted by the weakly-supervised model introduced in this paper. Currently, the Search4Code dataset includes 6,596 Java queries and 4,974 C# queries. The following are several notable advantages of the Search4Code dataset: 1. Real-world queries: All queries are sampled from anonymous Bing search logs, which we believe offers a realistic portrayal of how developers search for code. 2. Clicked URLs: Each query is accompanied by a list of the top three most frequently clicked URLs within its associated search results. For any given search query, the most frequently clicked URLs represent the search results that deliver satisfactory solutions. 3. Popularity rating: Each query is assigned a popularity rating based on its occurrence frequency. 4. Large-scale nature: The dataset contains thousands of queries, thus supporting the training of more complex models that require substantial amounts of training data.

搜集汇总

背景与挑战

背景概述

Search4Code是一个包含真实世界用户查询和对应最常点击URL的数据集，附带代码搜索意图标签，涵盖Java和C#的数千条查询。数据源自Bing搜索日志，由Microsoft Research于2021年发布。