Knowledge Graph Construction and Applications for Web Search and Beyond

One table and seven figures of this paper. Table 1 presents Wang Zai’s personal information for more consistent question answering. Figure 1 is an overview of Sogou knowledge graph construction framework. The framework could be divided into three parts: Data Preparation contains operations including collecting data from various sources, extracting data from both structured source and free text and normalizing data; Knowledge graph construction contains all models to build a knowledge graph based on the extracted and normalized data; Application is composed of applications or services of a knowledge graph. A box with solid line represents an operation or model to process data while a box with dashed line represents the intermediate data. Figure 2 shows overview of Sogou knowledge graph storage architecture. Figure 3 shows after query entity linking, entity-related information and pictures are shown in search results. Figure 4 shows that search engine also gives related entity recommendations, due to entity relations in our KG. Figure 5 shows long text entity service used for Anchor Text generation. Figure 6 presents KBQA in Sogou Search Engine, which answers the query directly in the first search result. In Figure 7, a user is having a dialogue with our conversational assistant, Wang Zai.

本文包含一张表格和七幅图表。表1展示了王在的个人详细信息，以实现更为一致的问答。图1概述了搜狗知识图谱构建框架，该框架可划分为三个部分：数据准备，包括从多个来源收集数据、从结构化和非结构化文本中提取数据以及数据规范化；知识图谱构建，包含所有基于提取和规范化的数据构建知识图谱的模型；应用则由知识图谱的应用或服务组成。实线框代表处理数据的操作或模型，虚线框则代表中间数据。图2展示了搜狗知识图谱存储架构的概览。图3展示了查询实体链接后，搜索结果中显示的实体相关信息和图片。图4展示了由于知识图谱中的实体关系，搜索引擎还提供了相关实体推荐。图5展示了用于锚文本生成的长文本实体服务。图6展示了搜狗搜索引擎中的知识库问答（KBQA），该系统直接在第一条搜索结果中回答查询。图7中，一位用户正在与我们的对话助手王在进行对话。